How do you beat Michael Jordan?

Joel Greenblatt used to ask this question to open his Special Situations class he taught at Columbia Business School between 2002 and 2006.

The expected answer was simple—you play something other than basketball.

You have to play something where you have an edge. In basketball, you aren’t winning.

The same applies to investing.

If you want to win in the market, you have to fish in a pond where you have an edge.

Think of the giant investment firms, hedge funds, equity research institutions, and investment banks as Michael Jordans of the industry. To make money, you have to win against them. That’s only possible by fishing in a different pond than them.

These businesses usually fish in the world of large caps, as small companies don’t move the needle for them.

This doesn’t mean that you can make money in large caps. You can, but it’s easiest in times of a market crash, where exceptional businesses are sold off due to panic, despite no problems in their fundamentals. This is when you can buy them cheap enough to make substantial money.

Conversely, in good times, it’s extremely hard to make money on large caps. As all these bigger institutions split the hair to see if there is any upside left in large caps, which often leads to them being fully priced.

However, these big players often totally ignore small-caps as they don’t move the needle for them. This is the pond where smaller investors can have a real edge.

This is what every smart investor does when they start.

Look below at Warren Buffett’s portfolio in 1951. His biggest position, GEICO, was a $7 million company at the time, which would be $91 million in today’s dollars. He didn’t own any large-cap equities.

If you want to have a definitive edge, small caps are the way.

We too have increasingly turned to small-caps since the beginning of this year, as the broader market has remained largely overvalued and large-cap opportunities are rare.

We have done very well in small caps. We have bought 3 small-caps this year, and we literally have a 100% hit rate. Our average return from these positions is 45%.

There are a few reasons pushing me to fish more in global small-caps:

• Concentration in the US market has peaked; unraveling of these positions could have a catastrophic impact on the market.

• Macro risks are accumulating; 82% of the surveyed banks expect an interest rate hike in September.

• Margin debt is at all-time highs, which may lead to more than a justified correction if a bearish trend takes over.

In this environment, global small and micro caps are perfect for investment:

• They provide diversification away from US market risk.

• They are less liquid, limiting the portfolio variance.

• They are cheap as capital is focused on the US.

Thus, I have long been looking for another hit in the small-cap realm. I have finally found one.

Take a look at this:

• Growth has taken off since a successful business model pivot in 2023.

• Earnings are expected to inflect and turn positive this year.

• It has just started its international expansion.

What’s even better? It’s trading at just 7x NTM earnings.

Yes, 7x, despite an excellent management track record, 40% annual expected growth through 2028, and a recurring revenue business.

Market cap below $180 million. Management sees it over $500 million in the next three years. As it achieves inflection this year, re-rating to 13-14x alone will provide a 100% return in just a few quarters.

One of the best setups I have seen in a long time. I am a buyer.

So, let’s cut the introduction and dive deep into this amazing gem!

Subscribe now

🏭 Understanding The Business

If I were to describe a perfect business, there would be two ingredients.

First would be recurring revenue.

Something like a subscription model. If you have recurring revenue, you don’t need to find new customers to sell your product every time. That makes growth easier.

Second would be a defensive product/service.

If the offering is defensive in nature, churn would be lower; thus, the number of new customers you need to find to grow the business would be lower as well.

This business has both.

Read more

Everybody remembers the atomic bomb, but there were actually many technological leaps that happened during the war that were no less marvelous than the atomic bomb.

One of them was the German V-2 Rockets.

It was the world’s first long-range guided ballistic missile and the first human-made object to reach space. It paved the way for modern space exploration.

It wasn’t developed suddenly in the course of the war. The research trajectory was extending back approximately 20 years, including systematic military rocket development from the early 1930s. It required breakthroughs in liquid-fuel engines, turbopumps, guidance, aerodynamics, materials, and high-altitude flight.

Even after all the breakthroughs were achieved and a physical rocket program started in 1932, it took 10 years to make the first successful test in 1942. It’s estimated that Nazi Germany spent around 2 billion marks, $40 billion in contemporary dollars, to develop V-2.

10 years and 40 billion dollars.

When the Soviets occupied Germany, they found a few V-2s, though not a complete blueprint. They had to open it up, understand, reverse-engineer, and replicate.

Soviets launched the first V-2 replica in 1947, just two years after they found the original ones. Their estimated spending is around 1 to 2 billion rubles, or $3.5 to $7 billion in today’s dollars.

10 years and $40 billion on the one side, 2 years and $7 billion on the other.

This is a recurring theme throughout the history of technological advancement. Original innovators generally end up assuming excess risk and cost, while the copiers, followers, stealers- say what you want- end up with faster development and lower cost.

This is not necessarily bad. This is how the world develops collectively. It happens again and again over time.

From computers to chips, from smartphones to autonomous vehicles, we have seen this many times, and we’ll keep seeing it as long as innovation exists.

Yet, people keep getting baffled by it every time it happens. This time, it’s happening with the AI models.

Chinese model Kimi 3 was launched last week and delivered SOTA-level performance. Many people shared charts showing how little China’s hyperscalers were spending on AI compared to the US, and yet they managed to make competitive models.

This created a panic in the market, prompting investors to question the capex rationale. If the Chinese can replicate top models and even make them cheaper by spending less, how can we get a return on investment when spending shifts to these cheaper models?

There is a deep divide on this issue, and I recently observed that divide in our community as well.

Is the AI trade done? Should we sell, or is it a temporary dip we should buy?

To be honest, I was going to publish a deep dive on an attractive opportunity today, but I see that we more urgently need a direction on this AI topic before anybody could think about buying something. After all, if this is a real issue, the whole AI trade could collapse, and the market would come down with it.

I spent my early career as a competition lawyer specializing in high-technology sectors; then I made my investment career by focusing on competitive advantages in technology sectors and currently doing a PHD on the side in Competition Law & Economics focusing on technology markets. Most of my research draws on historical innovation cycles to understand how innovation shapes future competition.

Thus, I think I have a few words to say on this one.

I’ll try to explain what’s actually happening, what it means for the AI sector, and how we should think and position ourselves as investors. I think everybody will have cool heads at the end of the write-up and a grounded perspective as to what’s happening.

So, let’s start from the basics—what the actual hell is going on with the Chinese models?

Subscribe now

How come Chinese models are so efficient?

There is no free lunch, but it’s a recurring theme in human history that we are often dragged to believe and act like there is a free lunch.

You can see this everywhere. In investing, for instance, people often think trend following would work well for them once they see a momentum in place. It doesn’t take long time to understand it’s not the way to riches. You can’t replace knowledge, research, and patience with herd following.

Most of the time, when we think there is a free lunch, it’s because the price is paid by somebody else, and this is why we don’t see it.

It’s the same in technological innovation.

Whenever someone achieves building and selling an existing product way cheaper, it’s often because the capex was frontloaded by the original innovators.

IBM spent roughly $5 billion building the System/360 mainframe. Then, Gene Amdahl, the chief architect of System/360, left IBM and founded Amdahl Corporation in 1970. His company spent just $40 million to build the Amdahl 470V/6, an alternative mainframe. Capex to discover the original know-how was assumed by IBM.

Same goes for branded drugs/generic drugs, first iPhone/copycats, Tesla/Chinese EV companies, etc. Original innovators assume the capex and go through development periods spanning years; followers/adoptors get there faster and cheaper.

Thus, whenever a company comes up with cheaper versions of the existing products and systems, we have to ask whether it’s really a free lunch, or just a lunch paid by others.

This is the question we should ask about the current Chinese AI models.

The widespread perception is that China builds cost-optimized models that are comparable to the state-of-the-art American models in capabilities:

As we said above, the perception is that they are achieving this by spending way less on R&D. For reference, the total capex of the top 20 American AI companies was way above $500 billion in 2025, while this was only around $65 billion for Chinese peers.

Yet, it looks like China has managed to catch up with the American frontier.

Kimi K3, an LLM released by the Chinese lab Moonshot.AI, is placed in the frontier level while reportedly costing way less per token compared to the American SOTA models like Claude Opus 4.8, ChatGPT 5.5, etc.

Claude Opus 4.7 costs $5.00/M input and $25.00/M output; Fable 5 costs $10/M input and $50/M per output token. ChatGPT 5.6-Sol costs $5/M input and $30/M per output token. Kimi K3 costs $3.00/M input and $15.00/M output.

When you look at the surface level, it seems like China made the best of both worlds. It looks like a free lunch.

Remember, whenever something looks like a free lunch, there is always somebody paying the price.

You have two options:

• You can believe that China achieved a breakthrough unmatched by the US labs.

• You can believe Chinese labs are somehow building on the paid-up innovation.

The first option is nonsensical. It’s way too audacious to think that the best of the experts in the world working in Silicon Valley with literally infinite resources are actually stupid enough to not discover doing it all at a fraction of the current costs.

That leaves the second option as the more plausible one since it’s also consistent with the history of technological development. This is exactly what these Chinese labs are doing.

They are basically distilling American models to create similarly performing models without assuming all the upfront costs related to research & development.

What the hell is distilling, you may ask.

Well, you basically take a big teacher model and use its outputs at scale to train your own model to behave similarly, thus perform similarly.

It’s basically training a smaller model on the outputs of a bigger model as the developer queries the teacher model for answers, reasoning traces, code, tool-use examples, or preference data.

Distillation can happen during post-training and in pre-training by mixing synthetic teacher-generated data into the main training corpus. The result is that we end up with a model with similar capabilities to the teacher but with substantial training efficiencies.

So, how do we know those Chinese models are distilling American models?

First, American labs are aware of the usage patterns that signal distillation. Over the past few months, Anthropic detected that several operators linked to Chinese firms carried out almost 29 million exchanges with Claude using thousands of fraudulent accounts. What are they doing? Coding Alibaba’s new front-end? I don’t think so.

Second, model specs and actual performance validate this.

Kimi K3 is a 2.8 trillion parameter model. For comparison, American SOTA’s like Claude Fable 5 and ChatGPT 5.6 are estimated to have more than 5 trillion parameters. This makes the student model perception stronger for Kimi 3.

The most important indication, however, is Kimi’s failure to pass its training efficiencies to inference.

If there was a real breakthrough, you would expect to see it in inference as well. However, despite seemingly lower cost per token, Kimi’s cost per task is at the same level as GPT 5.6:

Of course, this is not to say that Chinese didn’t do any innovation; they did many. It takes a whole different set of breakthroughs to achieve this level of performance even with distillation. They basically made the best out of their limited resources.

This way or another, these models are reality. They are here, and they look cheaper to use, though not by much. So, there are two relevant questions here:

• How will this change the inner dynamics of the AI race/buildout?

• What does it mean for investors in terms of positioning?

Let’s turn to these now.

Share

Will frontier open-source kill the AI buildout?

This is the most relevant question now. After all, if you are creating a product spending billions and your competitor can somehow distill it into a comparable product at a fraction of your cost, how can you stay competitive and thrive?

First, we have to acknowledge that there wouldn’t be an open-source frontier without American capex. It’s not a free lunch. American investors are paying for it.

Up until now, they have seen positive signals related to their investments primarily through the insane ramp of frontier lab revenues. Anthropic and OpenAI’s combined ARR reportedly exceeded $70 billion as of April 2026:

This revenue ramp is seen as what justifies the whole buildout.

Hyperscalers are getting into agreements to provide capacity to those frontier labs, and they are using free cash flows to buy hardware to build that capacity for AI labs. And those AI labs are financing these capacity commitments also by investments they raise from VCs, hedge funds, wealthy individuals, and other tech firms.

So, if Chinese models can make cheaper models with similar performance, then usage can shift to Chinese models, and revenues of American labs could erode, which would send the whole AI capex down with it because of the circular nature of the deals, and AI investments will freeze. This is the catastrophe scenario the market fears.

I don’t think it’ll happen. Or more accurately, I don’t think it can happen.

First, as we saw above, the cost per intelligence of the frontier Chinese models is very close to that of American SOTAs. As long as this is the case, not too many companies in the West will prefer Chinese models due to strategic and regulatory risks. Thus, American AI lab revenues won’t stall anytime soon.

Second, there is no other option but to keep spending.

If we look at the circumstances and expect decisions to flow from the consideration of the circumstances, it misleads us to think that there are options. If we start from the actual possible realm of decisions, we get a clearer picture.

There are two possible decisions:

• America cuts spending.

• America doesn’t cut spending.

Let’s create a basic decision matrix considering probable outcomes of AI spending decisions of the US and China.

Now, the status quo is useless for the US. If we were talking about 12-24 months lead of American models over the Chinese ones, the status quo could be acceptable. However, in the current situation, we are looking at models basically on par, and China holds the cost advantage, however little it is.

So, obviously, from the game theory perspective, the only reasonable decision for the US is to keep spending, and even increase its expenditures, however painful it could be in the short-term.

This is not just about international dominance of the country; it’s also about the dominance of private companies. If China wins AI, Chinese companies will also win semiconductors, biotech, health-tech, cloud, etc. The only plausible decision for American companies is to keep spending as well.

If you see AI as a system and not just an LLM race, you see more clearly that it makes sense.

If you think narrowly about AI and see it just as a race to dominate the LLM market, you would have a hard time justifying all the spending. After all, you would assume constant capability improvement to justify spending, but if the capability will constantly improve, what stops people from using cheaper models that are just good enough?

Suddenly, you would find yourself in an environment where 90% of the tasks are done by “good enough” models. There’ll be many of them, and it’ll be a fragmented market. Due to this fragmentation, not enough value will accrue to the top firms to keep them making new and better models. It’ll be a dead end.

But if you see LLMs as the tip of the iceberg, that paves the way toward more general AI systems that will pervade our lives and change the world, then you can justify it.

You can assume that the system development will hit a level where it’ll be impossible to dilute all the aspects, capabilities, components, etc.

ASML’s lithography machines are an example here.

They spent €22 billion and 20 years to develop the whole system. Now, despite the blueprint being largely known for every component, no company is anywhere near replicating the whole system. ASML reportedly holds a 10-year lead over the competitors.

It makes sense to spend on system-level development to hit a milestone like this in AI.

What kind of an AI system are we talking about here?

It’s impossible to know even with our most visionary state. Think about it. Who could expect Tesla Roadster when Ford Model T was first launched? Who could expect iPhone when the first telephone was made? Who could imagine Netflix when we first came up with digital video? It’s impossible.

And honestly, it’s even more impossible for AI given the endless opportunities it opens up. We just don’t know what will be made possible by AI and how its end-state will look. It’s just a big unknown.

All this being said, it’s almost certain that the introduction of these models and the tension between the proprietary American models and the Chinese open-source models will create fluctuations in the market. I think it’s just impossible to be totally insulated from these fluctuations while still being exposed to AI to ride the upside.

If you don’t want the volatility, the option you have is cutting AI exposure altogether.

I don’t think it makes sense to completely ignore a new technology that will almost certainly have a revolutionary effect on the world. Fortunately, I think it’s possible to position yourself in a way you can hardly be a loser in the long-term, which allows you to comfortably ride short-term volatility.

I think this is the practical dimension that investors should put more thought.

So, what’s the best positioning given the trajectory of AI and LLMs? Let’s discuss.

How to optimally position in the AI trade?

We have to acknowledge two things to reason our way to the optimal positioning.

First is that demand for intelligence will always increase.

When you read this superficially, you can object and say that demand for anything never always goes up. But it does for intelligence. You think about the times AI didn’t exist and you’ll see this. Demand for intelligence has always gone up in human history. We have become more and more educated, specialized, and knowledgeable.

Because intelligence is the ultimate Jevons Paradox.

We may not know now what we can do with additional intelligence, but when we have extra intelligence, we have always found a way to make productive use of it. In old times, we were intelligent enough to create basic tools; then we were more intelligent, and we created the iPhone.

Thus, if intelligence is now something that can be represented by tokens, we can say token demand will ever increase. And given that ChatGPT was launched just in 2022, we are still in the very early innings of this ever-increasing token demand.

Second, we have to acknowledge that not all work requires the same level of intelligence.

The basic tool and iPhone example applies. Turning raw steel into a fork doesn’t take the same level of intelligence as designing the iPhone. The natural consequence of this is that we hire people from different intelligence levels to do different work.

A barber doesn’t need to be as intelligent as a chip designer at Nvidia.

Generally, people who do higher-intelligence work get paid more than those who do work that requires lower intelligence. This is how we maximize the economic benefit from a given activity. If a hairdresser got paid $1 million a year, haircuts would be too expensive for most people; the benefit of having a haircut wouldn’t be enough to justify the price.

If intelligence is now shifting to models, the same divide applies to models as well.

We will inevitably employ cheaper models for lower intelligence work and more expensive models for higher intelligence work.

The problem is that AI models develop in generations. For human beings, development starts from 0 for everybody; thus, variance is higher. Given generational leaps, variance is way lower for AI models.

Thus, at some point, even the cheapest AI models will be smart enough to do most of the work that is currently in the realm of only the so-called flagship models.

Think about Opus 4.8 today. It does almost all of the coding work that human developers could do. 12-18 months from now, non-flagship, cheaper models in the market will be performing at the same level. So, we’ll likely use them for most of the work and switch to expensive frontier models only for the top 1% of the work.

Take a look at this, and it’s obvious:

As you can see, the smaller, next-gen price/performance model Gemini Flash 3.5 performs better than the flagship models released just a year ago.

Next year, smaller and cheaper models will perform comparably or even better than today’s flagship models for most tasks. As task complexity doesn’t magically go up, spending will likely shift to these models.

We are already seeing this happen.

Chinese models have now overtaken American models in token share on OpenRouter:

If you read this as a shift from Chinese to American, you’ll be misled. All things being equal, no Western enterprise has any incentive to pick Chinese models over Americans.

If you like to see it, look at the cloud market shares. How many Western enterprises use Chinese cloud providers and how many use Amazon, Microsoft, Google, Oracle?

This is not a shift from American to Chinese; it’s a shift from expensive to cheap, and it’ll accelerate as cheaper models achieve better capabilities.

If you see it this way, you clearly see the options:

First, American labs ignore this and keep pushing the frontier harder, hoping that they’ll push so hard that the systems will reach a level not achievable by distillation.

This could work, but I don’t think it’ll stop token share gains of Chinese labs. Even if they improve only incrementally from here, they can already do most work, especially coding, in a satisfactory way. Most tokens will be Chinese, and people will turn to the frontier only when it’s needed.

In this case, what we should be aware of is that the token consumption will likely go through American cloud providers anyway. They’ll be hosting those Chinese models, and Western enterprises and individuals will be more comfortable using models through Western cloud platforms because of privacy and regulatory concerns.

The second option is that American firms take notice of this and also make cheaper models. It’ll be like distilling their own frontier by themselves and releasing cheaper-to-use models alongside the frontier.

This is easily doable for them as they are already making the frontiers from the ground up. As price-performance will be equal to Chinese models, there won’t be any incentive for anybody in the West to pick a Chinese model.

In this case, models will be distributed through Western cloud platforms as well. The only difference is that the inference margins of the cloud platforms will likely be lower, as the American labs will maximize their gross margins thanks to their long-term contracts with the cloud platforms.

So, in both cases, we’ll see token demand surge. This is not surprising, as this is what we have seen so far, and we are only in the early innings:

In the first case, however, hyperscalers will benefit both from increasing volume and higher margins. In the second case, they’ll likely have lower margins, but volume will surge anyway.

Thus, it’s actually a no-brainer to hold hyperscalers right now.

This reasoning could be passed onto the neo-clouds as well, with a reservation on their direct dealings with the American AI labs.

If American labs pick the wrong strategy and lose the token volume and revenues, their direct contracts with neo-clouds will be at risk. Hyperscalers, on the other hand, will have Chinese token demand automatically replace American AI lab commitments, so they are not exposed to the same risk.

However, a similar bullish argument can be made for neo-clouds as well. If inference demand skyrockets, even from Chinese models distributed over the cloud, we can think hyperscalers can end up absorbing excess-neocloud capacity.

However, this doesn’t change the logical order of exposure:

• Buy hyperscalers; prefer higher customer diversification.

• Buy neo-clouds; prefer higher exposure to hyperscaler contracts.

So, let’s look at where cloud backlogs come from, to the extent that we can know:

When it comes to neo-clouds, we know that virtually all of Nebius’ backlog is from Microsoft and Meta. Coreweave, on the other hand, has around $22 billion in commitments from OpenAI and an undisclosed amount from Anthropic in its $100 billion backlog.

The problem is that the market already reflects Coreweave’s higher risk due to both its financing structure and direct exposure to AI labs. Thus, Nebius remains overvalued, while Coreweave is discounted, and perhaps more than it deserves.

Among the hyperscalers, Microsoft is obviously the top pick at the time given lower concentration on Anthropic+OpenAI and lower valuation. I wouldn’t own Google and exclude it because of the inevitable demise of the search business. Oracle is obviously the riskiest one due to its insane OpenAI exposure.

So, assuming valuations were all fairly attractive, hyperscaler picks would be Microsoft in the first place and Amazon in the second; while Nebius would be first neo-cloud and Coreweave would be second.

Counting in the valuations changes the order. Currently, I would favor Coreweave over Nebius as the valuation discount reflects more than the incremental risk that comes from its ~30% exposure to OpenAI+Anthropic.

This is what I am already doing. I entered Microsoft two weeks ago and increased my Coreweave position. I put my money where my mouth is.

Overall, for the given reasons above, I think it’s very hard to lose money on cloud providers over the long term. Still, risk/reward could be improved by considering customers and valuations and determining priorities based on that.

Conclusion

Every major technology goes through a period where capabilities rapidly improve and then flatline. At this curve, there is a point where capabilities are good enough for most of the use cases.

Up until that point, the most important thing is the capability improvement. A substantial premium could be charged for incremental improvement in capabilities. After that point, the importance shifts to price. Willingness to pay a premium for incremental capability declines rapidly.

I think current AI models have reached that point for many tasks.

Up until this year, every new generation of models exploded the demand despite a rapid rise in the cost of intelligence. This year, however, we have seen demand shift from frontier to cheaper models, as they are often good enough for tasks at hand.

Cheaper models with frontier performance will accelerate this trend.

This is not something created by Kimi 3. It’s a frontier-level model with cost of intelligence on par with GPT-5.6. But we know for sure that there’ll be cheaper models next year and they’ll perform better than today’s frontier models.

Frontier labs should decide the strategy they’ll implement to adapt to these new token economics. This is a strategic necessity.

When it comes to AI spending, higher performance of Chinese models won’t likely lead to a collapse in AI capex. After all, without American capex, there wouldn’t be the current Chinese models, as they wouldn’t be able to find a frontier model to distill.

Cutting spending means the status quo, and China would like the current status quo better than America and American firms. So, American companies will likely see AI as a system-level development, rather than an LLM race, and increase their R&D efforts even further to reach a state that others can’t reach by just copying.

The only way out is through.

Luckily, we don’t need to speculate on this, as we know for certain that the demand for intelligence will go up, one way or another. For most of the West, this demand will flow through Western cloud providers, regardless of what model wins the token share. These cloud providers get a part of their compute capacity from neo-clouds.

So, it doesn’t take a genius to see that hyperscalers and neo-clouds with established customers offer the optimal risk/reward position within the AI trade.

If you bought speculative names that depend on the overall market vibe about AI, you may have every reason to panic. However, if you are positioned in cloud providers at attractive prices, there is no reason to panic at all.

On the opposite, I am highly inclined to see this as a buying opportunity and grow my Microsoft and Coreweave positions.

I hope this helps. Let me know your position and thoughts in the comments.

Share

Leave a comment

That’s all friends!

Thanks for reading Capitalist-Letters!

Please share your thoughts in the comments below.

👋🏽👋🏽See you in the next issue!

At least, this is the type of investing we practice here, and it’s worked exceptionally for us so far.

The definition clearly states the two critical elements—an attractive price and an exceptional business.

The importance of the “exceptional” part can’t be overstated. For reference, 3% of the firms generated all the wealth in the US stock market since 1926. Buying and holding exceptional companies is the key.

The problem is that it’s hard to practice this simple strategy as “undervalued” and “exceptional” are two properties that rarely come together. This is by its nature.

Exceptional companies are rare; thus, they tend to be very well known. Especially those that are above a certain size. Since they are well known, their stocks usually trade at a significant premiums to the market.

This changes rarely.

Looking at the market history, there are three cases where exceptional businesses can be categorically undervalued:

• General crash in the market.

• Industry-specific recession.

• Emergence of a new revolutionary technology.

The first two are self-explanatory, but the third is more nuanced.

Since the Industrial Revolution, firms at the frontiers of technology tended to be bigger than others, as higher technology often means higher value-add. Thus, at times of technological changes, the market questions how incumbents will fare.

Will they be the winners, or losers?

This often leads to fluctuations in their valuations, and more importantly, makes their valuations increasingly driven by sentiment. The more the market believes they’ll be winners, the more optimistically they are valued; if the market believes the opposite, their valuations get depressed.

This is exactly what we are going through right now.

The so-called “Magnificent 7”, as a group, is now trading at its lowest relative valuation premium to the rest of the index:

When you see a situation like this, by default, it should be where you look to see if there is an opportunity here before embarking on a treasure hunt.

The reason is simple—exceptional companies tend to stay exceptional.

Thus, when you find those stocks at a discount, by default, they are more likely to be buying opportunities than not. Naturally, you are closer to making money on them than you would be buying lesser-known stocks, as they expose you to obscure risks.

This doesn’t mean you shouldn’t look elsewhere. It just means that when something obvious comes up, you should first determine whether it’s really an opportunity. If yes, you have to prioritize them rather than setting sail for lesser-known waters.

This is what we’ll discuss in this write-up—does the declining mega-cap premium really create a buying opportunity in mega-caps?

To understand whether there is an opportunity or not, we have to:

• Establish whether some of them are really discounted, and not just cheap compared to their own historical metrics.

• If yes, we have to understand the reasons and see if they are justified.

This is what we’ll do.

Subscribe now

Do any of the mega-caps trade at a discount?

To understand whether the declining mega-cap premium has actually led to a real discount in some of those names, we need to understand the pricing mechanism of the market.

At any given time, any stock’s valuation is made up of two parts:

• Value of the current earnings.

• Present value of the future growth expectations.

Michael Mauboussin and Dan Callahan recently published a paper titled “Opportunities and Expectations” explaining how current earnings and growth expectations shape the current value of stocks. If you haven’t read it, I highly recommend you read it as soon as possible. We’ll draw on this paper here.

The core idea is simple.

We can estimate the steady-state value of a company’s current operating earnings by assuming that those earnings continue indefinitely with no growth. Under that assumption, the earnings stream can be valued as a perpetuity by capitalizing them using their cost of capital.

Why?

A business is an economic asset that generates cash flows for its capital providers, including both lenders and shareholders. Those cash flows should therefore be discounted at the minimum return required by all providers of capital, the company’s weighted average cost of capital, or WACC.

Thus, we can estimate the economic value of $1 of sustainable net operating profit after tax, or NOPAT, received every year indefinitely by dividing $1 by the WACC.

If we assume that the WACC is 10%, the present value of $1 of annual NOPAT received in perpetuity is $10, as (1)/(10%)=10.

This represents the steady-state enterprise value generated by each $1 of sustainable NOPAT. To determine the portion of that value attributable to shareholders, we then subtract net debt.

The result is the estimated equity value of the company’s current earnings power. If the equity is priced at a premium to this value, it means the market is assigning some additional value to future growth.

Important conclusion that follows from this is that a higher share of future growth in the current market implies higher expectations. Naturally, when expectations are high, it’s harder to deliver for companies. Mauboussin and Callahan clearly establish this in their paper.

Their research found that while the connection between the share of present value of growth opportunities (PVGO) in stock price and future performance is largely sporadic, the returns in the highest quartile are substantially lower. This means the inverse correlation strongly manifests itself at the extremes.

This is consistent with the general market behavior.

At 19x P/E, you can’t say much about the future direction of the market. It can go to 22x or 16x. Both are equally likely. However, at 25x, the mid-term direction of the market has always been downward.

According to JPM Research, every time the market traded around 23x earnings, annualized returns for the next 10 years have always been between -2 and 2%. No exceptions.

So, it’s extremely helpful to see how much of the current market cap is from future growth opportunities to understand where the expectations are for the mega-caps we are looking at today.

I have taken their TTM NOPATS, capitalized by cost of capital, subtracted debt, and calculated the ratio of value from current earnings and forward growth to market caps. I excluded Tesla from this practice as it’s obviously overvalued.

As you see, nearly half of the current market values of Meta, Google, and Microsoft come from future growth opportunities, while this is way higher for the others.

Now, these figures don’t mean much by themselves. Even if we look at historical figures for these businesses and compare them to where they stand today, it’s hardly indicative, as they are at much larger sizes right now. So, we have to compare it to some benchmark.

That benchmark is, of course, the S&P 500.

However, as the S&P 500 is an index directly based on market caps of the components, we don’t need to use NOPAT and subtract debt. We can just use TTM earnings and the current cost of equity, which is estimated as the equity risk premium + risk-free rate.

TTM earnings for the S&P 500 are around $293. Equity risk premium is around 4.23% as per Aswath Damodaran, and the 10-year Treasury yield is 4.55%, giving us an average index cost of equity of 8.78%.

So, capitalizing earnings with cost of equity gives us 3,337. Given that S&P is currently at 7,575 points, this means only 44% of the current market cap of the S&P 500 is from current earnings power of the index components. According to Mauboussin, the historical average has been 65%.

This means the market is extremely optimistic about the future. Though remember that the future returns of the index when PVGO on the 4th quartile have been substantially lower.

However, what’s even more interesting than this extreme confidence in the future doesn’t reflect on three companies—Google, Meta, and Microsoft, as they are the only ones for which a lesser portion of current value is PVGO compared to the S&P 500.

At first look, it’s easy to make sense for a few names.

Apple is not a capex spender, and it’s now valued like a consumer staple, so it’s natural that the market thinks most value lies in the future. Nvidia rides the capex wave, so it’s understandable as well. Amazon is a big capex spender, but its core e-commerce business is valued like a durable consumer staple, like Walmart, so it offsets the capex-heavy tech business.

Among the names whose future potential is discounted relative to the market, Google is easy to make sense of, as its search business is seriously threatened. Personally, I don’t remember the last time I did a deep Google search on a topic as I would back in the 2010s.

Some people could make the same argument for Microsoft software as well, but it’s obviously not correct. We are ditching search gradually, but I haven’t met any single person or a business entity that has ditched Microsoft software.

So, we are clearly looking at a special situation with Microsoft and Meta.

These are known as way above average quality assets with above average growth expectations. For reference, Meta is expected to grow its top line by 20% annually over the next three years while Microsoft is expected to average 17% annual growth in this period. In the meantime, the S&P 500 is expected to deliver 8.5% annual sales growth.

Despite their superior prospects, Meta and Microsoft are actually and substantially discounted relative to the market.

Thus, the question is whether this discount is justified or not.

Let’s dig.

Share

Why are Meta and Microsoft discounted?

There are three common reasons we can observe for the discount they get. We have to look at each one of them and explore whether they are justified.

1️⃣ Concerns About ROI on AI Capex

The common problem for Meta and Microsoft is that they are both big capex spenders. These companies’ combined capex is projected to reach $325 billion this year, while total capex for hyperscalers will surpass $800 billion:

Now, if this is the case, why are Google and Amazon more optimistically valued than Meta and Microsoft? What makes them different?

The difference is having a cutting-edge AI lab.

Google is competitive in the AI model race, and the market thinks it’ll be one of the winners. This translates to higher confidence to generate positive ROI on AI capex.

Microsoft has a strong AI unit, but they aren’t very competitive in the model race right now, while Meta’s Superintelligence Labs is just getting traction.

But Amazon also isn’t competitive in the race, right?

Yes, and it’s punished for that. When you look at the headline numbers, you may think Amazon is optimistically priced, but most of the premium is due to the durability attributed to its retail business. The market sees it as a Walmart-like consumer staples business.

Michael Mauboussin’s historical PVGO numbers for Amazon validate this. Despite seemingly high PVGO relative to Meta and Microsoft, it’s at its lowest level since it became public. This illustrates that the market doesn’t attach much premium to Amazon’s AI infrastructure business.

So, the main problem is ROI concerns rather than purely high capex. If you have a competitive AI lab, the market thinks you have a better shot at generating ROI; if not the market doesn’t think path to ROI is easy by just selling compute.

Is this really the case? I don’t think so.

Economics of the recent deals are illustrative. SpaceX reportedly generates ROI between 70%-120% from its deals with Anthropic and Google.

Yes, SpaceX compute business is designed to address acute demand. It’s called the short-term & large-scale market for compute where prices are 3x-4x the long-term contract prices of neo-clouds.

However, if this is the case, hyperscalers should also be generating above 20% ROI on the compute they sell. This has been hard to track due to the lag between capex outflows and revenues, but we can see we are in the right direction.

According to the latest Bloomberg figures, global AI revenues excluding China have reached 1.2x of D&A.

Assuming 8% cost of capital, 30% operating margins, and 5 years of depreciation, hyperscaler capex turns positive at roughly 1.7x revenue/D&A. Note that this is a very conservative assumption and the actual value is likely lower as D&A is longer than 5 years in practice. Nvidia A100 chips released in 2020 are still being used.

This means that hyperscalers can generate significant ROI in their AI infrastructure businesses even without making cutting-edge AI models.

Microsoft is currently not pushing hard on the model race; however, Meta has been pursuing this market relentlessly, and they have released pretty satisfactory image and language models lately. The market reacted positively.

If this strategy of Meta holds, its compute business will largely operate in short-term/large-scale market since bulk of the compute will be used internally, and it’ll only monetize what it doesn’t “currently” need. In this market, prices and ROI are way higher.

However, even if its attempts to be a competitive player in the models market fail and it has to monetize bulk of its compute through a cloud business more like that of Microsoft’s and Amazon’s, there is a clear path to ROI as illustrated by the figures above. Bloomberg itself also saw the numbers as a clear path to ROI:

In short, both Microsoft and Meta have a path to ROI on AI capex even without competitive model businesses. This is validated by both ROI numbers on SpaceX deals and the general ramp of the global AI revenue ahead of D&A.

2️⃣ Possible Capacity Overbuild

This is the second reason the market is skeptical about huge AI capex.

In a stabilized market, ROI is a function of capacity, so they are closely connected. But the main worry here is not an imbalancing oversupply in an otherwise stabilized market, but a massive capacity overbuild where most of the capacity won’t be utilized.

Those who are afraid of this frequently share the chart below:

As you see, we are just a couple of years into this investment supercycle, but AI capex is already around 1.5% of the US GDP. Given that annual AI capex is projected to at least double by 2030, we’ll likely see this number peak around 2.5% of GDP, which is largely in line with the previous investment booms.

When people see this, the logical conclusion that follows is that there will be a big bust, as all the major investment cycles were followed by a bust.

This is correct, but I think there is also a different perspective here.

Look at all these booms, and you’ll see that, despite the busts, we always ended up with less than we needed over the long-term.

Think about railroads. Despite the railroad bust of the 1870s, we currently need much more railroads than we have. And it’s harder to build railroads now, so the industry will likely stay supply constrained for a very long time.

Look at the telecom boom. Despite the bust, we are still expanding the infrastructure today as the internet bandwidth we need keeps growing.

This is basic Amara’s Law, which suggests that we overestimate the potential of a new technology in the short-term and underestimate it in the long-term.

As a result, we tend to overestimate the investment needed in the short-term, but underestimate it in the long-term. Continuing investments in telecoms, railroads, and hardware illustrate this.

So, why does this happen?

The answer is simple—diffusion.

Up until now, diffusion of new technologies has been slower than we can mobilize the investment.

Again, think about computers and phones. It took a long time to do the required investments, but also there was an adoption curve involved as these were the technologies that people didn’t have a clue about before their productization. This extended the learning curve, and diffusion has been slower.

As there were no global networks, news and know-how about new technologies took time to spread in the US, let alone the globe. And when they did, it again took time for other countries to take steps and adopt the tech.

Public data networks emerged in the late 1970s, commercial internet service providers emerged in the US in 1989, yet most of the world didn’t connect to the internet before 1993.

This lag between investments and diffusion led to busts as the owners of the investments couldn’t extract the benefit they expected from their spending.

This changed with the internet. As it’s a global technology now, every new technology provided on the web diffuses at an unprecedented pace.

This is exactly what we have seen with AI.

ChatGPT has a 3x to 4x faster diffusion rate compared to the internet itself:

The result of this faster diffusion is that the lag between capacity investment and utilization shortens. This increases the foreseeability for the capex spenders.

In the early 1800s, a railroad investor was acting with a lot of assumptions about 5-10 years ahead. Today, an AI capex spender sees how fast the incremental capacity it brings online is consumed in days, not months. Thus, they have a far better sense of short- and mid-term demand.

This is why we may not get the short/mid-term overcapacity that we saw with the previous cycles. The fact that we have a market for short-term & large-scale demand proves this.

Therefore, I don’t expect a post-investment bust at a scale because of this speed of diffusion.

What’s more interesting, the argument could be this: As the capacity ramps and technology diffuses, AI infrastructure will be like utility and can thus have utility margins even without a major capacity oversupply.

Let’s discuss this.

3️⃣ Margin Collapse in AI Infrastructure Business

The most relevant comparison here is probably the change in wholesale internet prices. As infrastructure expanded, the prices collapsed:

It’s even cheaper now, with some sources reporting as low as $0.08 per Mbps in key American and European cities with fully developed infrastructure.

This is basically utility-level pricing.

As a result of this, major telecom companies haven’t become mega-caps like Microsoft and Meta. They are basically like utilities.

The argument here is as follows, and I used to think this way as well: If AI will change the world, it has to diffuse like the internet and electricity first, and to reach that level, prices should collapse so margins of the companies in this business will be utility-like.

However, I changed my mind on this as I thought further on the subject.

Two important points led to this change.

First, look at the margins in the traditional cloud business.

Utility margins argument could have been made for the traditional cloud business, but Amazon unveiled +20% operating margins in 2015, and that climbed to over 35% since then. Instead of shrinking margins, we got expanding margins.

Why? Because of the difference in asset lives.

Telecom infrastructure, for instance, is composed of long-lived assets. Useful lives of the telecom assets are above 30 years. However, they are amortized in shorter periods. This leads to collapsing prices as price competition intensifies post-D&A period.

Server chips, on the other hand, have a depreciation period of 5 years on average, and can produce returns above cost for around 7 years. So, when you are operating with assets that you need to replace in 5-7 years, and every new generation of chips costs more than the older ones, you can’t expect utility-level pricing.

Second reason why margins may not collapse is that we are looking at an oligopolistic market where cost of ownership declines for incumbents with every new generation of hardware.

Due to the insane capital expenditures needed, the entry barriers are already high. Potential entrants can’t cover the cost of new equipment at scale. Incumbents, however, can do this. As a result, their cost of ownership declines over time; however, they don’t feel that much pricing pressure due to the oligopolistic market structure.

There are only a handful of companies that can provide compute at large scale and the highest level of reliability. Plus, most customers don’t switch for a few pennies of discount as it exposes them to a business risk. Customers also seldom cut prices on their existing models, so they are fine with paying the ask anyway.

This is a bit like retail.

Everybody can open a mini-market, but it’s almost impossible to reach the scale of Walmart and Costco.

The difference is that a bottle of water bought from a mini-market and Costco has the same value to the customer. In the AI compute business, 1 MW has 0 value for the customers that matter. But if you have +10 GW, they are literally dependent on you, allowing you to have satisfactory margins.

Because of shorter replacement cycles and an oligopolistic market with dropping total cost of ownership, I think the margins will stay at satisfactory levels and possibly even at higher levels than the traditional cloud.

🏁 Conclusion

Whenever you see exceptional companies at a discount, buy them. Don’t be distracted by thinking about what else you can add to your portfolio. It exposes you to higher risks and dilutes returns.

Currently, we have so much discussion about the mega-cap discount, as their huge premiums over the index have eroded lately. However, when we analyze it in more detail, we see that only three companies are really trading at a relative discount to the market—Google, Meta, and Microsoft.

Among them, I eliminate Google because of the search business. That leaves us with Meta and Microsoft.

Both of these businesses have a lower share of their current market cap come from future growth than the broader index. Yet, both have higher mid- and long-term growth expectations than the market average.

So, here the discount stems from the lower likelihood the market attributes to that growth, and possibly lower value of growth due to lower ROIC.

I don’t agree with this for the reasons I explained above.

If you buy the arguments, this means the market attaches lower future potential to two companies of which there are no questions about their superior quality.

This is why I see a buying opportunity in Microsoft and Meta, especially Microsoft, as Meta was bumped up by ~20% last week.

I have owned Meta for a long time but entered Microsoft last week. I’ll be looking to grow my Microsoft position going ahead and consider opportunistic purchases in Meta if it drops again due to sentiment change.

Subscribe now

Share

That’s all friends!

Thanks for reading Capitalist-Letters!

Please share your thoughts in the comments below.

👋🏽👋🏽See you in the next issue!

This is exactly what we aimed for at the beginning of the year.

If you remember, I said we had an extremely imbalanced market in late December and early January:

• AI names made up 40% of the S&P 500 market cap.

• The market overall was trading near 22x forward earnings.

• The equal-weighted index was trading at 17x forward earnings.

Thus, we expected a broadening and avoided buying names that already had outsized weight in the index. Instead, we hunted for opportunities in forgotten sectors and foreign markets.

This strategy played out perfectly.

We have witnessed a smooth broadening as the market-cap weighted index has come down to 20x forward earnings and the equal-weighted index has gone up to 19x:

This led to underperformance of the portfolios focused on the well-known AI names like Nvidia, as it’s up just 4% YTD vs 9% of the index. Meanwhile, we generated substantial alpha from focusing on ignored corners of the markets. Most of our gains came from three buckets:

• Foreign markets.

• Ignored S&P sectors, like healthcare.

• AI plays that were ignored up until this year, like AMD.

In the first 5 months, we observed a portfolio that was still correlated with the broader index moves, though it was weakening thanks to this strategy. Over the last few weeks, we saw a portfolio that positively diverged from the index as broadening was underway.

That broadening is largely complete in my view, as the valuation gap between the equal-weighted and the market-cap-weighted indexes has narrowed down to 0.7x.

The fact is that Mag7’s net contribution to index performance this year is just 28 points, while 609 points came from the rest. This used to be the exact opposite. For reference, Mag7 made up 68% of all gains in the S&P 500 from 2023-2025.

Mag7’s drawdown has also stressed some other top-of-the-funnel AI names, as they are seen as leveraged bets on Mag7.

This created another lopsidedness in the market.

I think the following clearly explains what I mean:

Meta, Microsoft, and Nvidia all have way higher growth expectations than the broader market. However, they are now trading at 18x, 17x, and 15x 2027 earnings, respectively, while the S&P 500 is at 19x 2027 earnings. The market is assigning no premium to the highest-quality businesses.

You may say it’s because of the capex worries, but that doesn’t explain Nvidia.

Indeed, the Mag7 premium to the S&P 500 has collapsed to the lowest level since 2025, and all that premium is basically made up by Apple and Google, if you exclude the inexorbitant valuation of Tesla:

Let me be very straight about it: If this rally continues, money should flow back into these names to some extent and attach them some substantial valuation premium over the index. These are the highest-quality growth in the index, and the earnings growth of the rest will fall short of keeping valuations at these levels.

If we think that money will flow back into these names, we should also see top-of-the-funnel AI names to get some love. These two spaces are where I think opportunities exist now.

Plus, we should also see that the recent rotation was still not broad enough to bring sectoral balance to valuations. Money has largely flowed into healthcare and equipment/industrials. Some other sectors are still ignored, and there are opportunities to capitalize on.

This is the general direction I’ll take today:

• Trim one position and totally exit another position.

• Buy one new foundational position.

• Top up two existing positions.

This way, we’ll be positioned well for the next wave of rotations while locking in gains from our positive divergence so far.

I’ll explain the trades below and add the link to the portfolio spreadsheet at the end of the write-up. I’ll be updating it after I execute the trades.

So, let’s get started.

Subscribe now

Here are the exact trades I am making:

The first trade is obvious for us.

Read more

Look at other successful investors, and you’ll see they have said similar things.

• Terry Smith: “Good businesses tend to stay good, bad businesses tend to stay bad.”

• Warren Buffett: “Time is a friend to a good business and an enemy to a bad one.”

• Pulak Purasad: “Great businesses remain great. Bad businesses remain bad.”

It’s a recurring theme because it’s true. Look at all the legends, and you can probably name one defining investment that’s done amazingly well over decades, not years.

Buffett bought American Express for Berkshire back in the 1990s, and it’s still holding it; Chuck Akre held American Tower for over two decades; Philip Fisher bought Motorola in 1955 and held it until his death nearly 50 years later, etc.

All these businesses were subject to a lot of talk about disruption and about why they wouldn’t perform well over the next decade while these investors held them.

Look at American Express financials in 1964 when Buffett bought it, and you’ll see:

Its revenues grew by 17% and net income by 11%.

Over the next 6 decades, many times people thought American Express would be disrupted. When credit cards started killing travel checks, when Visa pioneered open-loop card networks, when PayPal spearheaded mobile payments, when Apple launched Apple Card, etc.

Yet, fast-forward 6 decades, American Express grew revenue by 10% last year and is projected to grow by 15% this year.

Exceptional companies tend to stay exceptional.

You may object and say, “But many companies thought to be exceptional have been disrupted like Kodak, Blockbuster, Blackberry, etc.”

What distinguishes exceptional companies from simply the strong ones is their adaptability. Exceptional companies are often defined by a timeless function rather than a particular product. This gives them adaptability and durability.

Again, take American Express. It fills a timeless function—facilitating commerce.

It used to do it with travel checks; now it does it with credit cards and mobile payments. Times have changed, and it adapted because what defines it is a function.

Of course, everything can eventually be disrupted. But thanks to being function-driven and the adaptability that comes with it, 9 out of 10 cases where the market suspects disruption prove to be a great buying opportunity for long-term compounding.

If you can analyze the circumstances objectively and assess the company’s position, you can tell an opportunity from a threat and capitalize on it.

Google is a recent example.

The market thought it could be a potential AI loser and hammered the stock in early 2025. Yet, if you scratched the surface a bit, it was obvious that Google was perhaps the best-positioned company to win AI. The stock is up 2.5x since then.

Now the same thing is happening with Microsoft:

The stock recorded its steepest quarterly decline since the 2008 Financial Crisis as the market is worried about AI’s impact on software and its returns on AI capex.

The question is simple—is Microsoft an exceptional company that occupies a critical function with adaptability just like American Express, or is it simply a strong company that’s still receptive to disruption like Kodak, Nokia, Blackberry, etc?

If it’s the former, we are looking at one of the best buying opportunities in the market; if it’s the latter, there is no low enough price to justify taking the risk.

So, let’s cut the BS and dive deeper into Microsoft to answer this question.

What are you going to read:

1. 🏭 Understanding The Business

2. 🏰 Competitive Analysis

3. 📝 Investment Thesis

4. 📊 Fundamental Analysis

5. 📈 Valuation

6. 🏁 Conclusion

Subscribe now

🔑 Key Takeaways

🎯 Microsoft is one of the strongest businesses with the ability to spawn new businesses over different cycles of technological change.

🎯 The market currently sees the software business as the weak link and is questioning ROI on capex. This has sent the stock spiraling down.

🎯 Its software business isn’t threatened by AI, as it’s the default medium of exchange for knowledge work and will remain so. AI capex is also closer to generating positive ROI than the market thinks.

🎯 There are ample growth opportunities ahead thanks to the AI boom and changing pricing model of software.

🎯 The current price implies around a 13% discount to fair value even based on conservative assumptions.

🏭Understanding the Business

“Spawners.”

What the hell is a spawner, you may be asking. I heard this word for the first time from Mohnish Pabrai. Up until I heard it, I was using “anomalies” to describe the same phenomenon, but I think spawner is a much better fit.

It simply refers to a group of companies that manage to successfully expand out from their core markets and dominate several other markets as well, related or unrelated.

What makes them different from conglomerates is that spawners create or buy assets that are substantial growth drivers in their own capacity, while conglomerates usually deal with more mature assets. Thanks to this relentless focus on growth and domination, successful spawners also tend to be the best-performing assets.

Look at Amazon, for instance. It currently has 4 businesses each making more than $50 billion a year—online stores, third-party seller services, cloud, and advertising.

Spawners existed for a long time, but at no point in history have they become as big as the spawners we see in the digital markets. This is for two reasons:

• Digital markets removed physical boundaries on growth.

• No physical boundaries among product markets made it easier to leverage dominant position across different products.

Microsoft is arguably the first company that discovered the potential of spawning in digital markets, and thus can be considered the original spawner of tech.

Its 50-year history is defined by expanding into markets adjacent to its core products, dominating those markets as well, and then using them as a stepping stone for expanding into even more markets.

When it first started in 1975, it had nothing.

Bill Gates and Paul Allen were friends from Lakeside High School. Both were obsessed with coding, which was the basis of their friendship. They were a part of the small group that colonized the Teletype room in Lakeside. It was called the Lakeside Programmers Group, and other members included Ric Weiland and Kent Evans.

They had one common belief that glued them to the Teletype room—computers would change the world.

They parted ways as Allen headed to Washington State University and Gates to Harvard. In 1974, Allen dropped out of Washington State University and moved to Boston, largely to be closer to Gates and pursue entrepreneurial projects together. They had nothing in mind.

One day, Allen saw that January 1975 issue of Popular Electronics magazine was featuring a device called Altair 8800, a $397 hobbyist computer with no keyboard, no screen, and no software:

He bought the issue and rushed to Gates’s dorm room at Currier House and said they were about to miss the revolution they had been waiting for.

They called the manufacturer, MITS, and claimed they had a BASIC interpreter ready for the machine. They didn’t. They hadn’t even seen a real Altair. They wrote the code in eight weeks on a Harvard mainframe simulating a chip they had never touched, and when Allen flew to Albuquerque in early 1975, the code ran on the first try.

Suddenly, they had a company.

They named it micro-soft because this was what they were doing—software for small computers.

It looks coincidental now, but they actually had the spawning thesis early on, though in different words. Their original mission statement was “a computer on every desk and in every home.” This meant a massive TAM expansion which would unlock larger opportunities, and Microsoft would capture these opportunities as they emerge.

This is exactly what happened since then.

They spent the early years selling their catalogue of programming languages to microcomputer makers like Commodore, Tandy, etc. This elevated them in the market as the one platform-agnostic provider that everybody already knew. So, when IBM rushed into the personal computer market in the early 1980s, Microsoft was already a perfect fit to supply an operating system.

At the time, Microsoft didn’t have any operating system, and they had neither time nor R&D to make one. Gates made one of the most famous opportunistic moves in the history of technology entrepreneurship. He bought an operating system called 86-DOS from a small outfit named Seattle Computer Products for roughly $75,000, renamed it MS-DOS, and licensed it to IBM. So, the first IBM PC shipped with MS-DOS in 1981.

The crucial thing for Microsoft was that IBM didn’t ask for exclusivity.

So, when other box makers reverse-engineered the IBM PC, they flooded the market with IBM clones that also had Microsoft as the operating system, thereby making it the dominant company in operating systems.

This is not the only spawning Microsoft did in the 1980s. As they were making inroads in the operating system market, they were also building another business with giant potential—productivity software for computers with graphical interfaces.

As mentioned above, by the 1980s, it was already well-known among the box makers as a platform-agnostic provider of software. This led to Steve Jobs hiring Microsoft to create software for its upcoming computer with a graphical interface, the Macintosh.

Most of Microsoft’s productivity software was developed for Macintosh. They became very useful when IBM clone box makers that were shipping with Microsoft's operating system also upgraded to graphical interfaces, and Apple refused to license macOS to them. Microsoft developed the OS and bundled it with its productivity software.

Suddenly, both its operating system and productivity software businesses were booming as box makers flooded the market and Apple declined following Steve Jobs’ departure. Thus, just within two decades of its founding, Microsoft had spawned two giant businesses—operating systems and productivity software.

Spawning is self-perpetuating in the sense that the more businesses you spawn, the more adjacent and complementary markets you have, so the more opportunities you see.

By the early 2000s, Microsoft was the dominant company in productivity software and operating systems. Thus, when the internet got popular in the late 1990s, hundreds of millions of people were relying on Windows machines as their gateway to the internet.

Microsoft thought it could leverage its dominant position in operating systems to spawn another giant business—consumer services hosted online, like MSN, Hotmail, Bing, etc. Yet, this was a very different business from packaged software because it required running hyperscale data centers to host millions of global users.

They knew that it would take a lot of capex and would be loss-making, as they would have to offer them mostly for free. It indeed played out exactly as expected.

Microsoft spent a lot of money in creating its data-center infrastructure, but ended up losing these markets anyway. However, thanks to these efforts, it was one of the few companies on earth that knew how to build and operate a hyperscale infrastructure.

When cloud computing started to take off in 2007, Microsoft was already ready to capitalize on it, so they entered cloud in 2008.

The first years of it were a disaster.

Its first commercial cloud was a platform-as-a-service. You would write an application in a specific format, and they would run it for you.

It was architecturally elegant but commercially wrong because IT departments were used to running applications on physical servers in the company basements. Microsoft was asking them to rewrite applications in a specific format. What customers wanted was Amazon’s cruder offer—give me the virtual machine and leave me alone.

Naturally, it failed to take meaningful market share from AWS initially.

They renamed the platform to Azure in 2010 and appointed Satya Nadella as its CEO in 20211.

Nadella rebuilt Azure as an infrastructure-as-a-service, a direct AWS competitor. That was the right architecture and the right offer for the demand. So Azure took off and started to take market share from Amazon. This turnaround won him the CEO role post-Ballmer.

Now Azure is at $100 billion annual run-rate with 21% market share in enterprise cloud infrastructure. It’s another giant business Microsoft has spawned over 5 decades.

And this is exactly how you should see Microsoft.

If you see it just as a big technology company with several businesses, you’ll miss the essence. Microsoft is a spawner.

• It generates insane amounts of cash flows thanks to its dominant position in existing markets.

• Then, when it sees an opportunity in an adjacent or a complementary market, it leverages both cash flows and customer base to expand into that market.

• It turns the new business into its new growth engines and cash cows for the future and repeats the whole process. Self-perpetuating cycle.

At this stage, Microsoft is one of the most successful spawners in history that’s spawned dozens of businesses, and generates over $318 billion in revenue, $125 billion in net income, and $72 billion in cash flows over the last twelve months.

As you could guess from the process above, two things are crucial for Microsoft’s spawning:

• Viable adjacent/complementary market for expansion.

• Cash flows that enable it to capitalize on the opportunities.

It has both now.

AI cloud is an exciting new market adjacent to the traditional compute cloud, and it has all the cash flows to capitalize on it thanks to two businesses it previously spawned—productivity software and OS.

Looking at its spawning track record, the market should have been extremely excited, and the stock should have been trading at a substantial premium.

Yet, it’s posted its worst ytd performance since 2022.

• Why is it?

• Is it justified?

These are the two questions all the alpha will be derived from when it comes to Microsoft.

Let’s dig.

🏰 Competitive Analysis

I received the single best investment advice when I was in law school. It belongs to one of the top Business & Tax Law Professors in the US, Steven Bank.

It reads exactly as follows— “if you can find a company that’ll be around 20 years from now, you’ll likely make a lot of money.”

That stuck with me.

It feels mind-numbingly simple, but if you think about it, you see that it captures all the essentials of value creation.

• If it’ll be around 20 years from now, it’ll capture natural growth from inflation.

• It’ll also benefit from increasing population, as it’ll drive volume growth.

• Even if you overpay, 20 years is long enough to offset most overvaluation.

All that suggests the single most important thing you should decide is the durability of the business, and everything else is secondary.

What provides durability is the competitive strength of the businesses, or the so-called concept of “moat.” If you read competitiveness as durability and not as the narrow fight for market share in the Porterian sense, you easily understand it has two aspects:

• Durability against alternatives that emerge over time thanks to technology.

• Durability against the existing alternatives.

So, if you decide that durability exists, the only thing left to be decided is valuation.

You can assume base growth will be given due to inflation and population growth, but you can’t be so sure about potential growth from innovation. So, what you should do is make conservative assumptions based on inflation+volume growth and buy when valuation looks low enough.

You basically have two core decisions to make. That’s it.

For decades, the durability part has been taken for granted for Microsoft due to its dominance in operating systems and productivity software. Most computers shipped were Windows, and they carried Microsoft Office as their default productivity software, which made Office the standard among individuals and enterprises.

This gave Microsoft immense pricing power, which led to an explosion of cash flows as its marginal cost per additional product was nearly 0:

Now this is changing as the market thinks AI is a software killer and Microsoft is software. So, suddenly its cash flows from software don’t command the same premium they used to. It further worries the market that Microsoft is spending its software cash flows on AI compute, which is yet to generate a solid return on investment.

So, suddenly, in the market’s eyes, Microsoft is using less durable cash flows to fund low-quality investments. And the stock naturally got hammered.

The key thing to focus here and decide is the former part—whether software cash flows are really less durable.

Because if cash flows are durable, the worst-case scenario is Microsoft wasted a few years of cash flows, which doesn’t detract much from terminal value, and the stock would recover when Microsoft stops wasting cash flows.

This is exactly what happened with Meta in 2022-2023. The stock dipped because it burned billions of dollars in the Metaverse, which didn’t generate any ROI, but the stock recovered when Meta committed to efficiency.

Given the current reaction, the market is afraid that they are not as durable as they used to be. I don’t agree.

What the market is doing now is confusing low replacement cost with the ease of replacement.

Before AI, producing software was costly, especially new software. Developer/hour was already high, and the products would generally require a few rounds of iteration before they achieve product-market-fit. For this reason, software used to come with pretty high sunk costs.

AI basically collapsed the cost of development.

The market thinks that now the cost of development is converging to zero; the cost of replacement will also drop drastically. As a result, the overall software industry has been going down, dragging Microsoft alongside it.

It’s wrong to see Microsoft software through this lens.

Most of Microsoft's software was developed in the 1980s and 1990s.

Microsoft Office bundle was launched in 1988 and came to Windows in 1990. So, it’s actually pretty primitive software that we know how to replicate for decades. Development cost or hardship was never the reason Microsoft has proved durable all these years. Thus, AI can’t remove a barrier that didn’t exist in the first place.

You should think of Microsoft software in the same way as WhatsApp.

Development hardship or costs were never the reason WhatsApp hasn’t been disrupted. There have been thousands of messaging apps launched after WhatsApp, proving that replicability was always there.

And objectively, there have been better productivity software than Microsoft Office and better messaging apps than WhatsApp, but both Microsoft Office and WhatsApp still dominate; why?

Two reasons; one is obvious, the other one is less obvious.

The obvious one is that both WhatsApp and Microsoft Office software are networks. This is less obvious for Microsoft software because it creates externalized network effects, unlike WhatsApp, which creates internalized network effects.

The difference is simple. In internalized network effects, value is captured by the network operator; this is why they are generally in the form of platforms where it’s easy to capture value for the operator. In externalized network effects, value flows outward to users by empowering them.

Microsoft software creates substantial externalized network effects as it’s the default medium for the exchange of knowledge work. The value added increases with every exchange, every revision, and this is captured by the transacting parties themselves, not by Microsoft.

The less obvious one is that this external value capture by the users of the network leads to what economists Joseph Farrell and Garth Saloner describe as “excess inertia” in their influential paper titled “Standardization, Compatibility, and Innovation”.

The setup is simple: Two firms sit on an old technology standard, and a better one appears. Each firm prefers the new standard if the other also moves, as compatibility is more important than the upgrade itself. As no firms want to take the risk of incompatibility, a comprehensive switch happens only rarely.

This effect is elevated when the capability ceiling of the standard is not that high, meaning switching to a new standard doesn’t offer much capability expansion. When this is the case, excess inertia is stronger.

This mechanism almost always leads to continuity of the incumbent dominance, as delay gives opportunity to the incumbent standard to bridge the capability gap. Once the capability gap is closed, there is no incentive to switch at all.

This means a displacement is most likely when there is:

• High capability ceiling

• Wide capability gap between incumbents and the newcomers.

In the case of Microsoft software, the capability ceiling is pretty low. According to surveys, 90% of customers use less than 10% of Microsoft Word functions. For Excel, it’s less than 5%. As a result, excess inertia is very strong. This will likely give Microsoft the opportunity to bridge any gap that could emerge with competing products or AI models directly.

The fact that people still transfer their AI-generated work to Microsoft software before sharing it validates this excess inertia stemming from compatibility. They don’t want to leave Microsoft software; they want AI to use Microsoft software for them. Leaving Microsoft software doesn’t obviate the need for a common standard, and why leave an existing standard if novelties don’t promise much excess value?

This is why WhatsApp has remained the standard, not because of lack of competition or better alternatives. This is exactly why Microsoft software will remain the standard medium of exchange for knowledge work as long as computers remain the main device where knowledge work is generated. This is the critical function of Microsoft software, just like facilitating commerce is for American Express, ensuring adaptability which leads to durability.

This reverses the script that software cash flow is no longer S-Tier.

They are, and they’ll remain S-Tier. So, durability is still there. Change of perspective from “no longer durable” to “still durable” changes how AI capex should be seen as well. In the worst-case scenario, Microsoft is spending money on a low-ROI investment, which doesn’t detract much from the terminal value, provided the durability.

We saw this with Meta’s investment in the metaverse. That proved to be a literal negative-ROI spending, yet once Meta was done with it, the stock recovered in no time.

I believe this makes the case that Microsoft is as durable as ever.

We should emphasize here that this durability of cash flows also reinforces its position in the cloud computing market and ensures durability there.

It’s because cloud computing is basically a scale business, as you are making margins on compute. Larger scale means more bargaining power and priority with suppliers, which means faster build times and more attractive offers to customers, which brings even more demand and leads to an even larger scale.

As a result, the market is dominated by companies that could scale faster and further by reinvesting gargantuan cash flows generated by their other businesses. Look at all the cloud giants- Amazon, Microsoft, Google, Alibaba, Oracle- and you see that they are the companies that generate enormous cash flows in their other core businesses.

This creates a vicious cycle where those that already have the scale end up with even larger cash flows they should allocate, which leads to even larger scale, keeping the market as an oligopoly.

So, in short, Microsoft has no problem of durability. Software cash flows are as durable as ever, and it also contributes to the durability of the cloud business, which also self-perpetuates itself thanks to the oligopolistic market structure.

As the durability is established, the investment decision turns to valuation, which is closely related to growth prospects.

So, to determine whether the valuation is right, we have to understand what optionality we have on the table so we can price it accordingly.

Share

🏰 Investment Thesis

My Microsoft investment thesis relies on three pillars. These are the key optionalities I see for Microsoft going forward.

1️⃣ Cloud computing will grow secularly.

To understand the growth potential of cloud, you have to see that AI will be a General Purpose Technology (GPT).

What the hell is a GPT?

GPTs are the technologies that are critical inputs to other innovations. This is why demand for GPTs grows secularly for long periods of time.

Take electricity as an example.

It’s a GPT, as it enabled the industrial revolution and computer revolution. Today we are still making new gadgets and developing technologies that consume electricity. So, electricity demand keeps growing secularly even though it’s been two centuries since it was discovered.

The internet is another GPT, and it follows the same trend. It’s enabled the invention of many other technologies like social media, streaming, e-commerce, etc. So, demand for internet bandwidth has grown secularly as illustrated by the internet traffic over time.

AI is also a GPT as it’s an input to many other technologies that we currently use and the ones to be invented. Take search. It’s now AI-powered. AI does the web search instead of humans and fetches the information. It can use the computer for you to do everything you do on a computer and more.

Many gadgets and devices we are currently using will be AI-pilled and made smarter this way. Cars are the classic example. Then of course we’ll produce new and more capable types of robots running on AI. Even home appliances will likely be made smarter by AI integration.

All this AI demand will flow through data centers, so the cloud computing market will grow secularly, just like demand for electricity and the internet have grown secularly since the day we first used them.

Gartner’s revising their 2029 market size estimates by 48% in 18 months proves that demand growth outpaces even the most optimistic assumptions. This is indeed what we see in GPTs.

Microsoft reportedly holds 21% market share in this market currently, according to Statista. So, based on Gartner’s model and assuming no increase in market share, Microsoft can generate around $310 billion cloud revenue in 2029. This is more than 2x its total cloud revenue over the last twelve months.

And this is the base case. As explained, GPTs drive secular demand. So, these figures will likely be revised upward, as they were twice in the last 18 months.

2️⃣ AI Capex Will Turn ROI Positive Faster Than Expected

One of the main concerns surrounding AI capex is the worries about positive return on investment. Initially, the market didn’t think much about it and assumed all compute capex would turn out to be wildly profitable as traditional cloud has been.

Then, as capex exploded and reports came out to suggest that enterprises were struggling to generate positive ROI on spending, the market became more skeptical of hyperscaler ROI.

However, the latest revenue ramp of the top AI labs led to a revision in models. Anthropic and OpenAI together reportedly reached a $100 billion ARR run rate. This is way ahead of their internal projections published later last year. Total AI revenue excluding China has surpassed capex depreciation decisively:

Let’s do some rough math to understand where AI capex turns ROI positive, shall we?

Assuming 8% cost of capital, 30% operating margin and 5 years of depreciation, hyperscaler capex turns positive roughly at 1.7-1.8x revenue/D&A.

According to the data above, we are currently around 1.2x. This is pretty promising given that only around 3-4% of the world population are currently paying for AI. If this number reaches just 7-8% over the next few years, it’ll be more than revenues to justify capex.

Latest Bloomberg reporting also corroborates this:

Consensus analyst estimates also imply that AI revenues will start generating net ROI from 2028 on, as hyperscaler FCF is estimated to explode post-2027 even though capex will also keep climbing:

In short, one of the primary reasons that depressed the stock was the uncertainty over the ROI on AI capex. We now see a clear path to positive ROI as AI revenue growth has decisively outpaced the D&A growth for the first time since the beginning of this investment cycle.

Once AI investments start being accretive to earnings, we’ll likely see a substantial multiple expansion for all hyperscalers. When will the market do this pricing? Nobody can know. Could be three months from now or could be next year. But given the growth rate of AI revenues, it looks inevitable.

3️⃣ Software Pricing Model Will Change

The software pricing model has changed a few times since the invention of computers. Earlier, in the 1950s and 1960s, software came free with a mainframe, which is why a “software company” was an oxymoron.

In the 1970s and 1980s, enterprise software was priced per computer as computers were wildly different in power. Mainframe software came to be priced in MIPS, literally millions of instructions per second.

From the 1980s, pricing has shifted to per seat. And per-seat pricing shifted from perpetual licenses that led to bumpy sales to a subscription model in the cloud era, which made the model like annuities and sent the operating margins above 40%.

In the AI era, software pricing will change once again as AI agents will become the main forces of knowledge work and humans will act like orchestrators.

This enables essential software operators to create three-layered pricing:

• Base price per seat for the software itself

• Admission price per seat to use AI in the software

• Consumption-based usage of AI agents after admission

We are already seeing this as Microsoft has started to push through this model already. They are currently charging extra to unlock Copilot in Microsoft Software. This is required to use Copilot Cowork, which is agentic AI that can work across Microsoft apps and deliver the end product. Cowork pricing, however, is consumption-based.

This will be a significant driver of both software and AI revenues. Microsoft currently has over 450 million commercial Office/M365 customers, and only around 20 million of them are paying for Copilot, which is around 3-4%.

There could be two counterarguments here.

First, I see some people think integration of third-party LLMs like Claude would work better than Microsoft’s own implementation so people won’t buy Copilot. Those who argue this simply don’t know the history of platform leveraging in tech. Microsoft Teams is a recent example.

It was launched after Slack, and the user experience was way inferior. But it has overtaken Slack in just three years, as distribution is always the king:

Microsoft will surely leverage this and can make the user experience better than all other competitors, as we are talking about an application that sits on its own platform.

Second, some people argue Copilot AI is inferior to others. It was, but Microsoft solved this by making Copilot in the form of a model aggregator. You can pick from Claude and ChatGPT models depending on the task. Microsoft will still likely have positive margins here as consumption will flow through Azure.

As a result, Microsoft software and licenses revenue can even accelerate while contributing to AI revenues at the same time.

So, in short, Microsoft has ample growth opportunities going forward. It’s very well positioned to be one of the biggest winners of the AI boom, while its software business can still keep growing thanks to its position as the default medium of knowledge and changes in pricing structure forced by AI.

As we established durability and optionalities that should be reflected in valuation, we should see whether there are any red flags in fundamentals before creating the valuation model.

📊 Fundamental Analysis

➡️ Business Performance

Well, there is nothing much to say for Microsoft’s past performance. It’s a rare event, and one of the most powerful forces in the history of business.

We are looking at a $3 trillion company that is 50 years old and still growing like a young-growth company:

What’s even better than this is the underlying transformation of Microsoft’s business, which illustrates its spawning ability.

In 2023, it generated a total of $212 billion in revenues. This has grown to $318 billion over the last twelve months, meaning it added $106 billion in revenues since then. The cloud division was responsible for 51% of this growth, as it grew from $73 billion in revenues in 2023 to $128 billion over the last twelve months.

For Google and Amazon, only 27% of their total growth came from cloud in the same period, meaning Microsoft is ahead of others in shifting its core business to AI.

We are looking at a 50-year-old company that still managed to position itself as one of the pioneers of an emerging revolutionary technology and derives substantial growth from it. I can’t imagine a better scenario for a 50-year-old, $3 trillion company.

➡️ Financial Position

If there was something like a gold standard of financial health, it would be a perfect declining line through total equity, EBITDA, and total debt.

Microsoft has it:

Thanks to its gigantic cash flows, it doesn’t need to utilize much debt in its operations. This is the most important thing for public shareholders, as it eliminates the possibility of permanent loss of capital.

This means, in the worst-case scenario where AI capex goes completely trash, all Microsoft is losing is a few years of cash flows, which doesn’t detract much from terminal value, as we already established the durability of the business.

This part doesn’t require much further deliberation. It’s one of the strongest companies in the world and will remain so.

➡️ Capital Allocation & Profitability

⏺️ Margin Profile

Throw gross, net, and free cash flow margins of Microsoft onto the same chart, and you’ll clearly see why it’s an amazing investment.

Look at the gross margins first.

Microsoft’s gross margins have been exceptionally stable since 2018. More importantly, gross margin is now exactly where it was in 2022 before the launch of ChatGPT, which means AI has not affected its software business at all.

Net margins, on the other hand, substantially expanded in the same period, from 31% in 2019 to 39% over the last twelve months. Considered in relation to stable gross margins, this shows its incredible operating leverage.

Plus, as you see, despite heavy AI spending, Microsoft's free cash flow margin is still above 22%, making the case that Microsoft is not drowning in sunk AI costs.

Finally, looking at gross and net margins together also refutes a bearish argument that’s taken off lately. AI bears make the case that cloud infrastructure economics get mega-cap tech companies closer to capex-heavy industrials than capex-light tech businesses. Microsoft’s gross margin has remained stable since 2022, and net income margin expanded, which clearly refutes this argument.

⏺️ Return on Invested Capital (ROIC)

Microsoft’s ROIC remains at an elite level despite heavy AI infrastructure capex, which comes with ~18 months of lag between cash out and first revenue.

We clearly see this as Microsoft’s ROIC peaked in 2022 before the ChatGPT release and has steadily declined since then. If you know what happened with AWS, this doesn’t stress you.

Amazon got heavy criticism for absorbing huge AWS costs between 2011-2014, where stock appreciated only by 50%. Then it unveiled +25% operating margins in the cloud business in 2015, and the stock made 3x by 2017.

We’ll likely follow the same path with AI infrastructure capex as well. Microsoft’s ROIC will likely dip in 2027-2028 as the 2024-2026 capex ramp will hit, but then recover as we experienced with AWS.

In the meantime, Microsoft will get heavy criticism for investing large sums in a business that compresses ROIC; however, you should note that there is not much of an alternative. These are giant businesses and opportunities to deploy hundreds of billions in capex at a higher ROIC than WACC don’t come every day.

Alternatives for these businesses have been buybacks, which are less accretive for shareholders than even below-par ROIC, as stocks of these companies generally trade at a premium and buybacks themselves also tend to increase that premium. So, in most cases, they end up overpaying for own shares, which isn’t accretive to shareholder value over the long term.

So, Microsoft didn’t have any other opportunity but to go all in on AI infrastructure, and it’s going well so far.

Summing up—Microsoft’s financial position is rock solid.

The business has been performing at an elite level, the balance sheet is one of the strongest on earth, if not the strongest, margins clearly illustrate it’s not getting inferior economics from AI infrastructure, and ROIC is still well above satisfactory.

It’s as good as it gets. So, only one question left—is the valuation right?

📈 Valuation

As I have said above, the most important thing in investing is minimizing the permanent loss of capital. This is why margin of safety is a central concept.

It’s primarily applied in three layers:

• First by picking a business with high durability.

• The second layer is keeping future assumptions conservative.

• The third layer is demanding a price below what even conservative assumptions suggest.

With Microsoft, we established the durability part and future optionality. The mistake many investors make is incorporating the optionality as accelerated growth beyond the baselines, which rarely works well.

In Microsoft’s situation, it could be justified to assume acceleration in growth given the potential of AI. Analysts project its cloud division can generate +$800 billion in revenue in 2030.

However, it’s already a $3 trillion company, so I wouldn’t project much acceleration from here to be conservative. Instead, I think optionality will reflect itself by extending the current growth period.

Here is what I would conservatively assume:

• Its TTM growth rate is 18%. I assume 18% annual growth for the next 5 years, which then gradually converges to 3% in the 5 years after that.

• The effective tax rate is currently around 18%; I assume the marginal tax rate will be 25%.

• Operating margin will decline from 40% to 45% as cloud makes up more of the company.

• Cost of capital will decline from 8.3% now to 7.5% in maturity, which is the average of mature companies in the US.

• Terminal ROIC will be 15%.

Here is what this gives us:

As you see, even based on very conservative assumptions that price no acceleration in growth despite opportunities ahead, we end up with $445 per share fair value for Microsoft. The stock is currently trading at $390 per share, implying a 13% discount to the share value.

I would easily buy Microsoft even if it was only fairly valued, which makes the current price a no-brainer for me.

🏁 Conclusion

Exceptional companies tend to stay exceptional, and bad companies tend to stay bad.

This is why when there are discussions about potential deterioration of an exceptional business, you can find the best long-term opportunities.

Microsoft has been depressed by such discussions for some time now. However, if you really understand the model and go deeper to understand what actually makes Microsoft’s moat, you’ll see that it’s as strong as ever and possibly getting stronger, as it’s well positioned as one of the top beneficiaries of the AI revolution.

Sooner or later, the market will see this as well and re-rate the stock.

Meanwhile, we are getting an opportunity to buy one of the best companies in the world at a nice discount to fair value.

As I explained above, its software business is as strong as ever and will likely make even another jump in growth thanks to the changing pricing model of software. AI infrastructure business is already generating the bulk of the growth and will likely keep accelerating here over the next few quarters.

I am a buyer here.

Will make it one of my foundational positions next week if the price doesn’t suddenly jump before I can buy at the current levels.

Remember, investing is as simple as buying exceptional businesses at fair value. If you can catch them at a discount, even better. This is the case for Microsoft now.

Subscribe now

Share

That’s all friends!

Thanks for reading Capitalist-Letters!

Please share your thoughts in the comments below.

👋🏽👋🏽See you in the next issue!

Portfolio is now up 49% over the last twelve months against 18% for the S&P 500!

There is something I need to do before I start talking about the portfolio.

I have received so many nice messages lately, as this month was nothing but pure alpha creation. It’s my duty to thank our members. They are not just readers; most of them are active contributors, and many of them are friends I can call by name.

The community has always been an integral part of our success, as it’s a medium where decisions and thoughts about the prospects and current positions mature. So, it’s my duty to give our great people credit here. They are not passive beneficiaries but active contributors to our performance. Once again, thank you all.

This has been a hard but amazing month for our portfolio.

Value reached its high-water mark while ytd performance remained stable around 21%, and TTM performance came at 49%. If our portfolio were an official fund, it would have outperformed all but one of the major Wall Street funds YTD:

Most people focus on the headline performance, and indeed our headline performance is great. However, what’s better for me is that the portfolio has behaved exactly as it was designed to. This is the most important thing for me as it shows our knack in controlling portfolio posture and positioning.

If you have been following this publication for a while, you know that one of the themes I talk most about is “base effects.” This is the biggest problem of high-performing portfolios like ours.

It’s very simple in principle.

When you buy an undervalued stock, and it reaches full value, it sets a high base. From there on, it’ll be harder for the stock to go up as mean reversion turns to downward pressure from upward driver. If you remain inert, you are more likely to see the price decline than rise. This happens as long as fair value, since it tends to be the mean in the long term, doesn’t rise faster than price in all periods:

The way to overcome this is portfolio turnover, i.e you dump/reduce fully valued stocks and buy new prospects instead that have room to go higher until they reach fair value.

Portfolio turnover comes as a natural urge to most investors. The problem is that most investors don’t want to tolerate intermittency in their performance. When they see their portfolio up 50%, they want to buy things that will go up fast and make that 50% to 60%. This is often what kills long-term returns.

The problem is that when the momentum becomes undeniable and easily observable, stock is often either about to reach full value or already beyond it. Thus, mean reversion has either already turned to downward pressure or is about to turn.

The correct move here is the exact opposite, and it’s actually what prevents performance intermittency at the portfolio level.

When we exit or trim a position, we always reallocate capital to what’s undervalued at that stage. The natural consequence of this strategy is that new positions tend to take longer to perform, as we can’t know where they are on the downward-sloping curve. They are rarely at the bottom tip.

This creates an intermittency at the individual holding level. But, if we are right on the thesis, it eventually climbs the upward sloping side of the curve, driving portfolio performance.

When this is practiced consistently with discipline, portfolio-level performance intermittency also declines substantially as positions perform at different times. Thus, as we wait for new additions to perform, older positions that already started their move on the upward sloping side of the curve carry the performance.

For instance, in 2025, one of our most substantial performance drivers was SoFi, which we aggressively bought in late 2023 and early 2024. Meanwhile, our Oscar position that we bought in late 2024 didn’t do much throughout 2025. This year, Oscar has become a multi-bagger for us, driving substantial alpha.

This way, portfolio performance kept improving without any major intermittencies.

At the beginning of this year, I made three important claims according to this strategy:

• One of our big positions that lagged in 2025 would drive alpha in 2026.

• We would look at international markets and derive alpha from there, as the US was already expensive.

• We would keep making opportunistic purchases in the AI space and the US generally.

I am very happy that all these have materialized.

• The position that lagged in 2025 is up nearly 30% this year, driving alpha.

• The non-US position that we took in January has become a multi-bagger.

• We made an opportunistic purchase in the AI space and exited after a 30% run.

We are not deriving alpha from randomness; the strategy we set out earlier drives alpha. This is what makes me happier than the headline performance because it’s what counts. It’s what shows that themes can come and go, sectors may explode or implode, some countries can stall, and some countries may grow fast; regardless, our strategy can drive returns.

We have derived performance from AI, from semis, from healthcare, from serial acquirers, from education, from financials, all in 12 months thanks to our consistent application of this strategy. It could be AI today and something else tomorrow. We don’t care; we have a strategy well suited to drive returns wherever the alpha is.

We’ll keep implementing the strategy with discipline, and I am confident that we’ll derive alpha regardless of changing market conditions, as we have done so far:

Our portfolio is now outperforming the market by 31%!

The following transactions took place in our portfolio in June:

• Exited 2 positions.

• Trimmed 1 position.

• Opened 4 new positions.

In the previous updates, I had provided my outlook for each company in the portfolio. Below, I’ll provide an outlook for our new positions as well and share overall portfolio commentary/strategy going forward.

So, let’s dive in.

🚨Our portfolio is proprietary to the members supporting the publication🚨

Members get portfolio updates every month!

Here is a 25% discount to celebrate new readers!

Valid only until Monday!

Get 25% off for 1 year

📊Here Is Our Full Portfolio!

As of today, we have 22 holdings in our portfolio.

Read more

The reality is different.

Warning signs emerge before all the great market collapses, but people just choose to ignore them. We just choose to believe it’s not yet time for the music to stop.

Paul Cohn is a partner in Agility Equity Partners, and he was a VC in Silicon Valley during the dotcom boom. This is what he wrote about the environment in the Valley at the time:

As you see, unlike the popular narrative, signs were there, and nobody was blind. Everybody understood what was going on, but just chose to ignore the signs to pursue quick profits.

Similar warning signs are flashing today.

• It’s forgotten that what matters is growth + profitability, not just growth.

• Valuations have already hit unsustainable levels.

• People ignore valuations to chase quick profits.

It’s arguably even worse today as the market has never been more expensive, not in the dotcom bubble, not in the 1929 bubble, never.

Yet, one thing history has taught us is that high valuations themselves are not a sufficient condition for a correction. Something needs to stop the money flow. As Ray Dalio says, “something needs to prick the bubble.”

Back in 2000, these valuations were sustained by VC money flowing to dotcom startups, which then went public, so the public market capital flowed back into VCs, allowing them to repeat this until public market capital stopped flowing as those IPOs started to go bankrupt and the Fed raised interest rates.

Today, high valuations are sustained by hyperscaler cash flows.

They are the ones funding the AI buildout. When the earnings go up across the AI value chain, it’s because hyperscalers are buying. The problem is, we are nearing the limits of their funding ability.

Capex guidance for this year is already well above 2025 operating cash flows for all of them. And cash flows are not catching up anytime soon, as projections by Epoch AI show:

This doesn’t mean capex numbers will suddenly collapse, but we likely won’t see aggressive upward capex revisions from now on, unlike what we have seen over the last three years.

What does this mean? Should we cut all the AI/semiconductor/hardware exposure?

Of course, no. Let me articulate once again where I stand:

I am a huge AI bull. As I explained in one of my previous articles, I think AI is a general-purpose technology that will pervade almost every electronics and digital processes we are using today. Over time, that will lead to a productivity boom, which will further increase demand for AI. Thus, I think computing demand will secularly grow to infinity, just like we saw with electricity demand and the internet bandwidth.

But this doesn’t mean all this buildout should happen fast over the next 3-4 years. We’ll likely see some stalls and reacceleration once the ROI catches up. This is why I believe we should remain bullish on AI while not ignoring valuations.

The straightforward implication of the potential slowdown in AI buildout due to draining hyperscaler capex is this: If you are buying stocks valued at 25-30x sales based on the assumption of upward capex revisions, you may get disappointed. Capex numbers may not grow fast enough to create the assumed sales boom for those stocks.

In this environment, what makes more sense is prioritizing the companies with contracted revenue, as these are already included in the current capex projections.

This is what underlies one of the transactions I’ll be making today. I’ll be opening a position in an AI enabler trading at a substantial discount to its contracted revenue.

Second, we should also be aware of the systematic risk we are facing today.

As I said above, the market is currently the most overvalued as it’s ever been. What makes this worse is that this overvaluation has mostly been led by one sector—semiconductors.

We also have a lot of money in this sector as we bought AMD in early 2025 and exited with a 5x return just two weeks ago. However, we shouldn’t be blinded by the amazing performance of the sector. It’s not normal. In the history of the market, the semiconductor index delivered over 230% in the rolling 14 months only twice:

This is alarming because capex growth hitting a wall may create a domino effect. It could lead to a correction in valuations across semis and hardware, which will trigger a market-wide correction as 49% of the S&P 500 market cap is made up by AI-enablers.

Given this systemic risk, it’s prudent to diversify away from the US whenever an attractive opportunity presents itself. Indeed, one of our best performers this year is Tasmea, which is an Australian consolidator of the skilled service businesses:

I have recently found a similar opportunity in the European market, and I’ll be buying my first shares today, as it’s already attractive in and of itself and allows us to diversify away from the increasing systemic risk of the US market.

To summarize, I’ll be making three transactions today:

• I’ll trim a position in an overvalued AI enabler and shift it to an undervalued AI enabler with contracted revenue.

• Open a position in a fast-growing European small-cap.

This way, we’ll keep our exposure to the AI trade without exposing ourselves to the risks related to a potential slowdown in capex growth, and also diversify away from the systemic risk of the US market while not sacrificing performance.

I’ll explain the trades below and add the link to the portfolio spreadsheet at the end of the write-up. I’ll be updating it after I execute the trades.

So, let’s get started.

Subscribe now

Here are the exact trades I am making:

This is one of the most obvious valuation disconnects I currently see in the market.

Read more

We are going through such times.

Since the beginning of the year, everything that can be somehow connected to the AI trade has skyrocketed, regardless of whether the connection promises real growth or not.

This isn’t something specific to now; it’s a characteristic…

Read more

This is also what steered our investment decisions earlier this year, when the market was at its peak before the Iran War. AI euphoria was pervasive, the market was at all-time highs, and investors were expecting rate cuts. Naturally, sexier, high-growth tech businesses were restrictively expensive across the board.

At that point, instead of forcing those expensive stocks, we shifted our focus to simple, boring, and consistent businesses that were ignored due to AI euphoria.

We picked Tasmea with that mindset, and it’s been one of our best-performing positions this year, alongside Nebius and AMD, which we bought in early 2025.

In my Tasmea write-up, I said it could double within the next 12 months. We are almost there, and it’s been only 6 months since my write-up:

It performed incredibly despite not being a tech stock in an environment where investors only wanted tech companies with AI exposure because its fundamentals and prospects were undeniably strong. The execution we saw so far validated this.

This is a clear example of the market creating opportunities somewhere else when everybody focuses on specific sectors and names.

I think we are going through a similar period again.

It doesn’t take a genius to see this:

• The market is at a near all-time high despite war and a jump in inflation.

• Valuations are high as the S&P 500 forward P/E is at 21 and shiller P/E is at 35.

• Stocks that can be even remotely connected to the AI boom go up 30% overnight.

It’s again time to shift our focus to simple, boring, and consistent, as the concentration of the current rally on the AI theme has created opportunities in other, less sexy corners of the market, opportunities like Tasmea.

Today, I am introducing you to one such opportunity:

• 40% annual revenue growth since 2021.

• 90% annual income growth in the same period.

• Insiders still own around 15% of the company

What’s even better is that it’s trading at just 13x this year's EBITA while guiding for +15% annual growth in the long-term.

Stock can easily double over the next 12-18 months if the management delivers on medium-term targets, and it could be even faster, like Tasmea, if they keep up exceptional execution.

It’s a true gem in an environment where almost everything is priced for perfection.

So, let’s cut the introduction here and dive deep into this hidden gem.

Subscribe now

🏭 Understanding The Business

One type of business, when done right, has consistently outperformed the market.

Read more

Our portfolio reached all-time highs, one position became a 10x-bagger, while another became a 5x-bagger, and we had opened these positions just a year ago in 2025.

Result? Our margin of outperformance reached 30%, and TTM performance reached 58%.

Just as I was thinking that it couldn’t get any better, it got better.

A high conviction position we opened earlier this year has officially doubled after going up ~20% on Monday. I got many kind messages for this from our readers, and I want to take this as an opportunity to thank everybody who sent these messages.

This is, of course, great, but it’s not without challenges.

As a result of their stellar performances, many positions in our portfolio have now gone well beyond their fair values, leaving us with limited & speculative upsides and pretty stretched downsides.

And this comes at a time when stock valuations are at all-time highs:

So, it’s challenging to find new, undervalued positions in which we can put proceeds from our overvalued positions. Yet, they exist.

Interestingly, I see many more opportunities than I would normally expect to see when the market is at this valuation.

This is because the incredible market performance since 2022 has created an insane amount of FOMO. Yet, people are also aware that valuations across the board are high, and we may be nearing the end of this cycle. Thus, everybody is trying to find stocks that can go up +100% in a short period of time.

That fundamentally changes the game from one of valuations to one of predicting what people will concur to buy next, and narratives are the vehicles creating these concurrences.

As a result, names that have positive narratives behind them go up no matter how high their valuations already are, and names that don’t have these narratives or have weaker narratives lag in price regardless of their quality and low valuations.

This is illustrated by the fact that the S&P 500 without AI is actually lower since the start of the Iran War:

This means the breadth is actually very low, and higher valuations are a making of a handful of high-flying stocks, not a market-wide phenomenon.

This is why there are still very high-quality businesses trading at attractive prices.

Today, we’ll reposition the portfolio to capitalize on these opportunities by:

• Exiting two well-performing positions that have become pretty overvalued.

• Buying two high-quality undervalued positions instead.

This way, a substantial part of the portfolio will have its profile change from “limited upside-stretched downside” to “limited downside-high upside.” So, we’ll have a better shot at beating the high bases set by those well-performing positions and keep advancing the portfolio performance in the remainder of the year.

I’ll be explaining the trades below, and you’ll also find the link to the portfolio spreadsheet at the end of the write-up. I’ll be updating it after I execute the trades.

So, let’s get started.

Subscribe now

Here are the exact trades I am making:

Let’s start with one of the most obvious trades I currently see in the market.

Read more

Portfolio is now up 58% over the last twelve months against 28% of the S&P 500!

I’ll be straightforward—it’s been an amazing month for our portfolio.

One position we entered last year has officially become a 10-bagger, while another position we took around the same time has returned 5x so far, and all this happened in just 14 months.

One position we took earlier this year is already up by 75%, and another one is up 30%. A position I took just last month is also up by 25%. This is supported by strong performance from our existing positions. One position in particular, which had been down for a long time, has recovered substantially, removing a long-standing drag.

Result? YTD performance has reached 21% and annual performance 58%.

No leverage, no chasing so-called bottleneck themes, no position bigger than 15%.

Pure stock picking alpha.

I am very proud of and happy with this. Obviously, this is an incredible performance as we are on the course to outperform the market by wide margins for the 5th consecutive year.

However, what’s more important than the returns themselves is that our strategy continues to work as intended. This is our main long-term asset.

If you have been following this publication for a while, you know that I frequently touch on the concept of base effects. It’s because I see it as the main challenge to consistent outperformance.

The mechanic is simple.

When positions find or exceed their fair value, two things happen:

• Portfolio goes higher, obviously.

• Downside risk surpasses upside potential.

This is because fair value exerts a gravitational pull on stocks. The more overvalued they become, the stronger this pull is.

As a result, when positions in the portfolio hit or exceed fair value, they create a very high base of performance, which is hard to exceed with existing positions due to that gravitational pull.

At that point, you have two options—sell or stay.

Most investors decide to sell. For us, selling has a high threshold because of two reasons:

• Great businesses surprise on the upside (and we only own great businesses).

• Proceeds should be put in a new position, which exposes us to a new risk.

Thus, we don’t easily sell due to overvaluation. We make valuation-based exit decisions only if a position becomes egregiously overvalued. At fair value, we always keep, and we may trim between overvalued and egregiously overvalued.

Once we exit or trim a position due to overvaluation, we need to redeploy that capital into something with more upside potential than downside risk, so that capital can exceed its prior high watermark—and, over time, so can the whole portfolio.

That’s obvious, and the urge to beat the base effects is natural to investors. What’s not obvious is that they want to keep performance in a straight line, which leads them to reallocate capital to positive momentum stocks. Yet, due to that positive momentum, upside is often already shrunk for those stocks, and downside is already stretched.

We don’t do that. We make our reallocation decisions purely based on the raw upside potential and downside risk, not momentum.

We seek and find high-quality & high-potential businesses ignored by the market for this reason or that. Investors generally pass on these businesses as they don’t believe they can generate quick returns within 1-2 months. This allows us to create meaningful alpha by taking a bit longer view.

Among those prospects, we are willing to invest in only those with long-term durability, as it’s what limits the downside. Then, we pick the ones with the highest upside potential so that we can consistently get above-market returns from the positions even after short-term opportunity costs.

I call this period when their performance is below the opportunity cost “incubation period.” When they are in this period, they tend to lag the rest of the portfolio, or even create some degree of drag, while existing positions carry the performance.

Over time, existing positions flatline as they reach their full valuation, the market recognizes the true value of new positions, bumps them up, and they come out of incubation, driving the performance upward again. We repeat this process by exiting/trimming positions that reached full valuation and incubating new positions in their stead.

Thus, when zoomed out, our portfolio cycle resembles the following (excluding systemic drawdowns):

AMD is a recent good example of this.

We bought it earlier last year for around $100 per share. At the time, its data center segment had missed expectations, and the market dumped it. We thought it would be temporary, as customers would inevitably diversify away from Nvidia and AMD, which were priced as they would never get another percentage of market share in data center GPUs.

It first declined to $80 levels before turning up. This was the incubation period. Once it delivered on its targets and announced GPU deals with major AI labs and hyperscalers, the market recognized the potential. It is now a 5x position for us.

This is the strategy we deploy with discipline to beat the base effects and generate consistent outperformance, and it has worked incredibly well for us so far:

💥Our portfolio is outperforming the market by 30%!

The following transactions took place in our portfolio in May:

• Trimmed 2 positions.

• Opened 2 new positions.

As our portfolio has widely outperformed the market, many of our positions have either reached or are about to reach their full value. Thus, we’ll have to exit/trim these positions and put the proceeds in new opportunities.

In the previous updates, I had provided my outlook for each company in the portfolio. Below, I’ll provide an outlook for our new positions, share overall portfolio commentary/strategy going forward, and also talk about the particular trades I am planning, i.e., stocks I am thinking to exit/trim, and those I want to buy.

Let’s dive in!

🚨Our portfolio is proprietary to the members supporting the publication🚨

Members get portfolio updates every month!

Here is a 25% discount to celebrate new readers!

Valid only until Monday!

Get 25% off for 1 year

📊Here Is Our Full Portfolio!

As of today, we have 20 holdings in our portfolio.

Read more

We finished last year up 44% while S&P 500 delivered 17.9%. We are currently up by 57% over the last twelve months against just 26% of S&P 500.

I am not telling this to brag; I am telling this because I will explain how we have achieved this so far, as this is o…

Read more

Twitter and Substack are inundated with people posting about semiconductor stocks or AI-cloud stocks that can run further or about the next so-called bottlenecks..

We have also made a lot of money on these themes.

I was one of the first people who wrote about Nebius on Twitter and this platform, and my early shares have now made…

Read more

• Our AMD position that we entered in early 2025 has become a four-bagger.

• We turned out to be right on Oscar as it delivered a monster beat. Stock hit $20.

• Fluence Energy, which we picked last October, jumped by 100% over the last week.

Thus, I have received many kind DMs over the past week. I tried to respond to everybody, and I want to take this as an opportunity to once again thank everybody for their kind words. Thank you.

Among all the wins above, the Fluence thesis was possibly the most divergent alpha, as nobody was watching it when I published the thesis in October.

Most people think the stock moved up violently because the hyperscaler deals it announced were a surprise. It wasn’t. It’s always been the thesis.

When I looked at it in October, the case was obvious:

• Almost all AI data centers need batteries as the cost of intermittency is high.

• Fluence was the best-positioned one as it was a JV between AES and Siemens.

• Fluence was the first mover in reshoring, with 100% American manufacturing supply chain.

As a result of being a JV by AES and Siemens, Fluence had already inherited most, if not all, hyperscalers in its book of business.

In the thesis I published in October, I wrote this:

So, it already had relations with hyperscalers, and its 100% American manufacturing supply chain made it even more attractive, thanks to tariffs on Chinese products. Thus, it was only a matter of time before it announced a new hyperscaler deal.

This was the core thesis, not a surprise, unlike what many people take from the aggressive re-rating.

After the stock moved, many people tried to chase it. You don’t need to.

Power is currently the biggest bottleneck for building data centers and scaling AI. There are many opportunities across power generation and distribution supply chains. Demand for power distribution equipment is especially tight as data center demand is growing way faster than the manufacturing capacity for these products.

Today, we’ll dive into one of these companies that is riding the exploding demand for power distribution equipment, and it’s still under the radar. Literally nobody talks about it.

Tailwinds are already hitting the wheel. Revenue jumped by 48% over the last twelve months, thanks to five data center contracts it received in that period:

In January, it received a data center contract half as big as its total revenues in 2025. Yet, this is just the beginning as the data center buildout in its market is still in very early innings, so it’ll likely receive dozens of these contracts over the next decade.

What’s even better is that this optionality hasn’t been priced into the stock as it’s still trading near 17x earnings against the industry average of 30x.

This is one of the best risk/reward opportunities I have seen in the power equipment industry for a long time, as the whole industry has been aggressively repriced due to the data center buildout.

So, let’s cut the introduction and dive deep into this hidden gem.

Subscribe now

🏭 Understanding the Business

Power demand from data centers will triple over the next decade, increasing from 500 TWh in 2025 to over 1,500 TWh in 2034.

Read more

Almost all our positions are appreciated, some have gone vertical, and our TTM performance climbed above 65%.

If you go back to the environment in early April, this would look unlikely. All-time high markets, conflict in the Middle-East, and increasing energy prices, declining hopes for rate cuts, etc.

It would be total insanity for any sane person to expect strong performance in the market under these conditions. I remember, at the heights of the Iran War, just as when valuations in the market started becoming attractive, many people exited the market, thinking that the risk was elevated.

Ironically, that was the time the market was most de-risked since April 2025. Many high-quality businesses were trading at attractive prices, and decisions were driven by extreme fear, exactly the conditions you would love to see to succeed in the market.

I said keep holding the quality businesses and even buy if you already have a big enough cash cushion. After all, newspaper headlines eventually wane in importance against the fundamentals.

The market once again humbled all those people who think they can better position themselves by looking at the headlines and forecasting the futures than by looking at fundamentals.

The market is up by almost 6% since the war started:

As the market was dipping in this period, we insisted on keeping our positions. We sold nothing, as we believe all our positions were very strong fundamentally. We even bought a new position.

Result? The past two weeks couldn’t have been better for us.

Our portfolio was up by 61% TTM in late April; now it has climbed over 65%.

Almost all of our positions posted positive returns in this two-week period, and a few of them literally went vertical.

What’s even better for me is that we haven’t derived this performance through a heavy exposure to one single theme like AI.

For instance, we own AMD, and we bought it in early 2025 when it was hated by everybody. That position is now almost 4x for us.

But that’s a 6% position for us. We also generated heavy alpha from financials last year and healthcare this year, alongside AI-driven sectors like semiconductors, cloud, etc.

That sectoral diversification is one of the core strengths of our portfolio.

However, it’s less of a deliberate pursuit and more of a by-product of our investing philosophy. We believe inefficiencies occur when people get concentrated too much on a sector, theme, or a specific name, completely ignoring others.

We are willing to look at those ignored places and buy inefficiencies there.

This was exactly the case when we bought AMD. Everybody was focused on Nvidia, and they thought AMD was trash. Same for Nebius. When we bought it, everybody was focused on Coreweave, and Nebius was there trading at $20 with a $5 billion market cap and almost $1 billion ARR guidance.

We now see this case again, and we see it more emphatically on a sectoral basis:

I recommend you to completely ignore the benevolent picture provided by P/Es relative to S&P 500, as S&P 500 itself as the current market P/E is between the 87th and 95th percentile. When you look at the absolute valuations, you see where you may find opportunities going forward. Even then, it’ll be really hard because no sector is undervalued relative to their 10-year and 30-year historical valuations.

Under these conditions, it’s imperative to look beyond what’s obvious and seek opportunities in overlooked sectors and industries to limit the downside risk while still retaining optionality.

This is what made us succeed so far, and we’ll keep adhering to it.

Today’s transactions will reflect this strategy once again. I’ll make four transactions:

• Trim two positions

• Open two new positions

Together, these transactions will broaden our optionality, reduce the concentration risk, and position the portfolio for further alpha generation going forward.

Let’s get started.

Subscribe now

Let’s start with the trims as always.

Read more

What followed this was perhaps even more striking:

He ended up warning investors against succumbing to these behaviors, as it always, at some point, ends up in disappointment and loss of their hard-earned money.

What he warned against is exactly what’s happening in the markets today.

People are chasing small-cap stocks trading at 200-300x forward earnings under the name of AI bottlenecks, with the assumption that a future evolution of the technology will create demand dynamics that will justify those prices.

Sounds familiar? If not, I recommend you read some of the investment theses about the new internet companies of the dot-com era.

I am not saying bottlenecks don’t exist; they do. But this doesn’t mean they’ll evolve exactly as we see today, and it definitely doesn’t mean paying 100x sales (yes, sales) for a stock is sensible. It’s not, and I guarantee anybody that they’ll end up losing money if they consistently pursue this.

Only group that makes money out of this feverish behavior, as Warren Buffett says, is the promoters of this behavior, i.e., the promoters of these stocks.

They take positions in small-caps before, then post about it in coordination, creating the hype. Some people naively follow them, believing in the future potential of the stocks, while others follow the greater-fool theory, thinking that they’ll be able to ride the hype and sell way higher if they are early enough.

And nobody is immune to this kind of behavior, even the best of the game.

Druckenmiller shorted $200 million in internet stocks in 1999, expecting the bubble to burst. The market kept rising, he got FOMO, and covered his shorts with a $600 million loss. He then went long, buying $6 billion of tech stocks. The bubble burst, and he lost $3 billion on his overall tech exposure.

When asked what he learnt, he said—“I didn’t learn anything. I already knew that I wasn’t supposed to do that. I was just an emotional basket case and couldn’t help myself.”

Nobody is immune.

In an environment where getting carried away is so easy, we have to consistently remind ourselves that almost nobody made a fortune trading dotcom stocks in the 2000s, only the shameless pumpers.

Look at all the great investors, Buffett, Munger, Lynch, etc. You can’t find a single one who made money playing the speculation game we are seeing in the markets today.

For long-term wealth creation, the recipe is simple—buy quality and never overpay.

But.. Are there still high-quality businesses with solid growth prospects and attractive valuations left in the current market environment?

The answer is an emphatic “yes.”

Fortunately, people’s clustering in the sexy industries and hype zones leaves the boring corners of the market ignored, creating opportunities for those willing to look.

I have just found such an opportunity:

• Revenues grew by 32% annually since 2023.

• Net income has almost doubled since 2024.

• Return on equity expanded from 14% in 2024 to 18% now.

What’s even better is that the stock is trading at just 10x this year’s earnings, and the management is aggressively buying back shares.

It’s a rare, mispriced opportunity in this market. I have been following the company for a while and recently decided to pull the trigger after their latest earnings print came even better than I expected.

It will be a multi-bagger over the next few years, even if it can hit conservative growth targets.

So, let’s cut the introduction here and dive deeper into this gem.

Subscribe now

🏭 Understanding The Business

The more niche you get in the market, the more inefficiency you’ll find. This has always been a well-known fact among investors.

If you are managing a small sum, go niche, overlooked corners of the market. You’ll see more inefficiency because the companies in these spaces generally lack broad analyst coverage and are too small for institutions to be interested.

The same principle applies in business as well.

The more niche you go, the more supply gaps you’ll find. Given that these gaps will be relatively small, you can create a defendable business if the product is sticky, as it’ll keep the contestable portion of the market small enough to make the entry unfeasible for potential competitors.

Read more

Portfolio is now up 62% over the last twelve months against 30% of the S&P 500!

It’s been a wild ride! — This is probably what best explains how I feel about the last few months.

It’s truly been a wild ride, and I am extremely happy and proud of where our portfolio has reached today after this wild ride. I’ll give you the update and talk about the strategies and investment philosophy, but first, let me start by extending my thanks to the great community we have here.

If you are a human with a conscience, writing about finance and stocks is very hard because you know that people make financial decisions based on what you write. The only way to have a clear conscience is to fully align yourself with what you write.

When something you write wins, you win; when it loses, you lose as well.

This is what I have always practiced here. If you have been in the community for a while, you have probably heard me say, “I can’t say buy because I am not holding it.” And not getting a concrete answer might have disappointed you at that point, but that’s the only way to have a clear conscience on my end.

As a result, sincerity has emerged as the default mode of operation in this community, and we have developed a strong sense of friendship. We have exchanged ideas, poked holes in our theses, and restored sanity through discussions when things got volatile.

So, thank you.

When it comes to our performance, I’ll say again what I have said a few times in the past—forget the performance, what’s important is how well your positioning played out.

Indeed, this performance could be 45% today instead of 62%, but this wouldn’t change anything in its impressiveness because what’s important was being able to position to generate this level of return.

If you remember, at the end of last year, I said two things:

• The market is overvalued.

• 2025 was hard, and 2026 will be harder because there is no single asset class that is clearly positioned to go higher.

Yet, despite all this, I also said “stay in the market.” I recommended exiting fundamentally weak or overextended positions, building a nice cash cushion, and leaving the rest in the market and allocating to opportunities as they come.

Then the war happened, and we still didn’t sell a single share.

Because our position was this:

• All our holdings were very strong fundamentally.

• In the worst-case scenario, they would go undervalued, which is never a reason to sell.

In the end, optimism won once again, and parties to the war entered a de-escalation path, so the market has come back. Then, AI optimism peaked as Anthropic revenues tripled in Q1 and compute demand skyrocketed again, driving further optimism.

Result? Most of our positions are now at or near all-time highs.

What’s even more valuable is that we have achieved this without any leverage or mass buying speculative stocks.

Fundamentals of all our positions today justify either the current valuations or way more than the current valuations, but not less. Having said this, I have to talk a bit about a dangerous trend I see on Substack and X (Twitter), under the name of “bottleneck investing.”

This is pure speculation, given a fancy name to make it look more reasonable. Bottlenecks, of course, exist, and it’s often productive to see the bottlenecks ahead and invest in them. If the bottleneck proves legit, companies operating there will likely have strong pricing power, and this will be enough to boost earnings by a lot.

Many positions in our portfolio are there because we believe they address certain bottlenecks as well. There is nothing wrong with this.

But… This doesn’t mean every price is justified or that the stock should be priced as if the bottleneck will persist forever. When this happens, the result is almost always a disaster.

Here is an example:

When you look at Corning stock, you see the last time a price movement like this happened was back in 2000. We know how it ended. This won’t be any different.

In some small-cap stocks, the situation is even more absurd.

You should be careful. If the fundamentals of those stocks were able to take them to highs, those who own them wouldn’t need to shill them every day. Those positions will be dumped on some people at some point.

We have never engaged in such speculative behaviors, and will never do.

Our philosophy is simple. We run a barbell portfolio where:

• We buy companies with durable competitive advantages, i.e., foundational stocks.

• We combine foundationals with growth positions that have strong balance sheets.

• We don’t sell unless fundamentals deteriorate or the price goes off the roof.

This has worked incredibly well for us so far:

Our portfolio is now up by 62% against 30% of S&P 500!

The following transactions took place in our portfolio in November:

• Exited 1 position.

• Opened 1 new position.

In the previous updates, I had provided my outlook for each company in the portfolio.

In this update, I’ll also provide an outlook for our new position and share my overall portfolio commentary and strategy going forward.

Let’s dive in!

🚨Our portfolio is proprietary to the members supporting the publication🚨

Members get portfolio updates every month!

Here is a 25% discount to celebrate new readers!

Valid only until Monday!

Get 25% off for 1 year

📊Here Is Our Full Portfolio!

As of today, we have 18 holdings in our portfolio.

8 of these positions can be considered foundational.

Read more

We are now up 60% in the last twelve months, against just 33% of the S&P 500.

How have we achieved this? Well, looking back, I think there are three factors:

• We let the winners run.

• We timely opened new positions trading at deeply undervalued prices.

• We waited patiently for our existing undervalued positions to play out.

Before the year started, we had closed three groups of positions:

• Those that went too high above their intrinsic value.

• Those whose fundamentals were deteriorating.

• Those that were too small to make a difference.

We rolled a portion of the proceeds from these exits over to new undervalued positions that could create alpha as they were starting from a lower base, and kept the rest to create a bigger cash pile as the market was obviously overvalued.

But we kept the exceptional companies we owned, even if they appreciated as their fundamentals were improving fast, and we didn’t see a safe selling price. In these positions, we would rather ride the downside volatility than give away the optionality.

Then, within this year, we have only opened three new positions. One of them is up by 30%, and the other two are largely flat. The good thing with other positions is that they tested their lows a few times since we bought them, and bounced back. They have now broken the downtrend and are ready to go up if favorable market conditions continue.

Another driver of alpha was the recovery in one of our most undervalued and biggest positions. We heavily increased our allocation in this name last year after it got heavily undervalued and it finally started to play out, creating alpha while the market has been largely flat YTD.

We had finished all these positional arrangements before the Iran War started. So, at the outset of the war, we didn’t have any “egregiously overvalued” position that could stress us during the downturn. Thus, we managed to do exactly “nothing” while everybody was panicking.

For me, this is the gold standard of positioning.

If you can afford doing nothing during downturns and it looks like the only logical thing is buying more if you are to do something, then the positioning was right.

That’s exactly how it was during the downturn. We didn’t have any position to stress us, and we could only consider buying more if the prices came down to our “no-brainer” zones. In the end, most positions didn’t come down enough for us to buy, given our ~20% cash position, and we ended up doing nothing.

Now, the market has recovered and recorded a new all-time high, and valuations have swung from “somewhat attractive” to “unattractive” again.

Indeed, the market is now trading at around 21 times earnings, only 200bps below where it was in late 2025:

And the credit spread is still below 3%, meaning the money is still abundant, and optimism dominates the capital allocation decisions.

This is not an environment where you find one fat pitch after another.

In this kind of environment, the best opportunities come when the qualitative conditions that led to the discounting of particular stocks start changing, and the discount looks increasingly unjustified. The market is sometimes slow to recognize these changes and to bump up those stocks to their full expected value.

I think we have one such opportunity, and it’ll be an obvious one in hindsight.

Yet I still want to keep my guard up, as the market's valuation makes me nervous, and I don’t want to reduce my cash position. This is why I will fund that position by closing a position that is probably the lowest conviction for me in the current portfolio set-up. It’s in the money, but I think the upside scenario is getting weaker.

So today, I’ll be making two transactions:

• Close one position.

• Open one new position.

This way, we’ll be able to get exposure to an additional optionality while eliminating a potential weak link in the portfolio in terms of market power.

You’ll also find the link to the portfolio spreadsheet at the end of the write-up. I’ll be updating it after I execute the trades.

Let’s get started.

Subscribe now

Here are the exact trades I am making:

Let’s start with the exit, as always.

Read more

Founder mode happens when an entrepreneur acts like he is still the business, because this is how it is when companies are young. There is nothing as “business” on the first day; it’s a founder and an idea that would …

Read more

“As fast as we put the capacity in, it's being consumed.” — Andy Jassy, Amazon CEO, said this in Amazon’s Q1 earnings call last year.

Almost a year later, after he said the words above, we read in his annual letter earlier this week that:

• Tarinium 2 capacity was sold out.

• Trainium 3 is almost fully subscribed.

• A part of Trainium 4 is already reserved.

Not ju…

Read more