The AI Trade Deconstructed: AI Builders, AI Suppliers and AI Product firms!

    When I posted my valuation of Anthropic a few days ago, I noted that turbulence lay ahead, both for the company and the market. I did not expect that turbulence to be this forceful and arrive this soon, but markets don't have an obligation to wait until it is convenient for investors, to adjust to new realities. With the election results on November 3 acting as a catalyst, investors initially responded positively with stocks up on Wednesday, but reversed course sharply on Thursday, with the NASDAQ down close to 6% and the S&P 500 declining 4%. This drop has again brought to the forefront questions about the AI trade and whether it has run its course. 

    In this post, I will start by referring to my earlier note on AI as a business, and use the ideas in it to deconstruct the many players that are lumped together as AI companies, from those spending the money building the data centers (hyperscalers) to the firms benefiting from selling the builders the ingredients for these data centers (chip makers, electrical equipment, power and even real estate) to the companies that are spending considerable amounts , leasing these data centers for long periods  (with the LLMs in the lead). I will argue that while all of these groups benefit, at least in general, from AI's success at creating large and profitable markets, we are at a point in the AI disruption, where their interests and payoffs will diverge, with implications for both the businesses and investors. I use last week's market carnage in technology and examine how the market treatment of these different groups may be a reflection of a more discerning market, less willing to mark up and mark down all AI companies, and more likely to treat them differently, as their business interests diverge.

The AI Transition 

    My journey on AI started with a selfish and narrow intent, which was to evaluate an investment I had made in Nvidia in 2018, for reasons entirely unrelated to AI. I thought that Nvidia, at the time, was a innovative chip company with more pricing power than many in its peer group, and felt that it was undervalued, given its prospects. While the stock outperformed my portfolio early on, it was the ChatGPT debut on November 30, 2022, and the subsequent AI buzz that put its stock price on steroids. That left me with a challenge, albeit a welcome one, of deciding whether I should continue to hold on to a stock that had already become a twenty-bagger or sell some or all of it. Since much of Nvidia's supersized success was being propelled by AI demand, that required me to take my first look at how big the AI market would be, and how well Nvidia was positioned to take advantage of that market. If you have read my multiple posts on Nvidia over the last three years, you may be aware that I sold all my shares in the company in four installments, with the last sale happening earlier this year. As many of you have been quick to point out, my decision to gradually decrease my Nvidia position has led me to leave money on the table, but I have no regrets. 

The Players

    In the process of trying to understand AI and disentangle the many players in the AI space. I break them down broadly into four groups:

1. AI Suppliers: While AI is viewed as a technology business, we have learned over the last few years that it is much more capital intensive than any other tech endeavor in history. The data centers and infrastructure that are needed to power AI agents and models has created a surge in demand for ingredients. The biggest portion of the cost, at least so far, has been on high-powered  chips, with Nvidia claiming a lion's share, and enough left over demand for other chip companies to benefit. This infrastructure build out has increased demand for electrical equipment, with power and real estate companies also gaining as data centers get built.
2. AI Builders: Alphabet, Amazon, Meta and Microsoft, four companies that were part of FANGAM, the acronym that predates the Mag Seven, bought into the AI hope, and joined by a handful of other players (Oracle, Coreweave), they have invested hundreds of billions of dollars into AI cap ex (data centers and infrastructure). The market cut these hyperscalers slack early on, rewarding them for their capital spending, even though that spending was denting profitability at these companies and demolishing cash flows, and part of the AI trade question has been whether the market is running out of patience, waiting for payoff.
3. AI Products: In the last two years, companies which are on the frontlines of creating AI products and services, have became the focal point of investors, albeit in the private investing world, with talk of public offerings pricing them in the trillions of dollars. SpaceX became the first of these to list, but Anthropic's IPO is imminent and OpenAI will join, sooner rather than later. There are  other companies (Palantir is the most prominent), albeit only a handful, that are  have made AI a central component of the products and services they sell, but that number will be added to, as private businesses in the space go public.
4. AI Customers: There is a fourth group, perhaps more critical than any of the above, that will determine the value creation potential and long-term standing of this business, and that includes all of the potential customers, individual as well as businesses, for AI products and services. While the premise for AI hype is that its reach (both in terms of numbers and monetization) will be huge, it is undeniable, even if you are an AI diehard, that there is little of substance to show at the moment. Even if you take the ARR numbers that are being leaked by the lead players, the collective revenues from AI products and services right now is about $250 billion, a small number relative to what has been invested in the space. The entire AI ecosystem is built on the premise that this group will become much larger, both in terms of reach across businesses and money spent in the trillions of dollars.

The companies in all of these groups can be classified as being in the AI business, but they generate value in very different ways from that business.

The Linkages

  There are links that hold these groups together, and I have tried to capture those linkages and how they play out as revenues, expenses and earnings in each group:

This simple, and perhaps even over-simplistic, breakdown of AI does, in my view, provide a good starting point for understanding why their interests and market pricing is more likely to diverge in the future, as the business takes form.  The linkages between the AI groups come from three core transactions between these groups that determine not just their current earnings but also their long term value propositions:

1. Building the AI infrastructure: The  AI builders are spending the money building the infrastructure and much of it is being legitimately classified as capital expenditures, though there is a discussion to be had about the life of that infrastructure. The reason that the investment life matters is that it determines the amortization schedule for the AI cap ex, and consequently the negative impact on earnings in the current year; the free cashflows  are lower since they take a direct hit from the cap ex. The suppliers report these same capital expenditures as revenues in the current year, and when netted out against their operating costs, increase earnings at these companies. I wrote about how the net effect of the infrastructure buildout has been an increase in aggregate earnings at US companies in the last year, and has supported stock prices, even as interest rates have soared.
2. Leasing the AI infrastructure: The AI architecture, once built, is leased out by the builders to the businesses that use the data centers to create and support the products and services they offer. The builders report the lease payments as revenues, allowing them to earn a return back on their infrastructure build, but for the AI product and service businesses, these payments are either operating expenses, if annually reset, or financial expenses, if structured as contractual commitments over many years; the latter will create imputed interest expenses and depreciation.
3. Creating and selling AI products: Using the leased AI architecture, AI product and service businesses create products and services that they sell to consumers (individual and enterprise) and record the payment they receive, from either subscription or usage-based models, as revenues. With the LLMs (OpenAI, Anthropic), these products are AI agents and frontier models where users pay either in subscriptions or based upon usage, and net of the costs of creating these products, hopefully into profits. 

The linkages between the companies in the AI ecosystem described in the section above are operating linkages that are part and parcel of doing business, but there are financial linkages, though, the motives for which I will examine later in this post, that are messier. In particular,  AI companies have funded each other, with both equity and debt. Microsoft and Nvidia are shareholders in OpenAI, Alphabet and Amazon in Anthropic and Nvidia has also provided financial backstops for data centers being built out. These financial cross holdings create accounting noise (with mark ups of equity investments reported as earnings) and contribute to opacity.

The Timing of Benefits

    Given that AI infrastructure can take years to come online, there will be timing differences across the groups, in terms of seeing the benefits from their business models, with AI suppliers being the first to see revenues and profits, followed by AI builders, with AI product & service businesses generally being last to see profitability.

1. Hope and Hype        

    In the first phases of a disruption, the visions of a huge market often drive actions on the part of all of the AI groups, but the way they play out and how they show up in results will vary:

• The suppliers of the ingredients that go into AI infrastructure will be the first to see tangible benefits as the spending on AI architecture shows up immediately as revenues in income statement and as incremental profits. 
• For AI builders, no matter how merited the build-up (in terms of delivering value over time), the immediate effect on both earnings and cash flows will be negative. The reason the earnings effects has been muted with AI is because the biggest AI infrastructure investors (Meta, Microsoft, Alphabet and Amazon) have earnings and cashflows from legacy businesses and the cloud businesses (thus playing a role asiAI suppliers) that have provided a buffer, but the free cash flow impact has been real.
• For companies in the AI product group, which are nascent businesses, there will be little to show in terms of revenues, since neither the infrastructure nor the products are ready to go. With R&D and employee compensation expenses, they will be money-losing machines.

For AI, the hope and hype phase was most pronounced in 2023 and 2024, and not surprisingly, the biggest surge in market capitalization during the period was at Nvidia, as its chips became the essential ingredient to get investments going.

2. Investment Buildup

    In this phase, the AI infrastructure building continues at a torrid pace, but portions of it are complete and ready for use, and the consequences again will vary for the groups involved:

• As growth in AI infrastructure continues, AI suppliers will continue to see growth in revenues and profit from the continued building, though growth rates will come down with scale. In addition, new entrants may start to compete on prices, reducing pricing power at least for some of the supplier groups.
• For Ai builders, the portion of the AI that is ready to use can be leased out, allowing for a monetization of at least a portion of the AI investment. Since the investment continues to increase, and much of it is still not operational, on measures like return on capital, the AI buildup investment will not measure up. Markets will often be willing to forgive, at least with managers that they trust, on the assumption that things will get better over time.
• As AI infrastructure comes online, AI product companies can start generating products and services, and to the extent that they are growing from small scale, growth rates can be exponential. Those revenues though will generally be dwarfed by expenses from product development, continued research, and employee compensation, leaving the companies with large losses.
• With AI users,  a mix of early adopters and experimenters, in a subset of industries where AI is viewed as having the most promise, will emerge. Given AI’s strengths, software businesses were among the first to be exposed to AI products, with consequences for competitive position and profitability, but AI agents also have found a foothold in white collar businesses.

Through much of 2025, AI suppliers continued to benefit, AI builders invested hundreds of billions of dollars in AI cap ex, seemingly without much pushback, the product companies drew attention with frontier models that dazzled users with their mix of power, albeit with rough edges, and markets continued to reward them all, based on its perception of market potential.

3. Business Building

    A big end market notwithstanding, business building is hard and in this phase it is inevitable that questions will turn to unit economics, economies of scale and eventual profitability, with each group facing different degrees of questioning.

• By this stage, AI suppliers should have fully formed business models, but revenue growth rates will ease, as new infrastructure spend becomes smaller, at least on relative terms. As builders try to develop cheaper supply sources, either by bringing some of it in-house or by looking at competitors, margins will start to compress, and you will see some suppliers focus on the infrastructure replenishment/replacement market, rather than on new infrastructure. Since markets are forward-looking, the pricing of these companies will reflect expectations of lower growth and the companies will look cheap on a pricing basis (low PE ratios, low EV multiples), relative to their history.
• As more infrastructure becomes operational, AI builders will begin to see more revenues and earnings from existing investments, but the magnitude will depend largely on whether demand from AI product companies is running ahead of capacity (good for revenues and earnings) or behind (bad for revenues and earnings). As returns on invested capital pick up, markets will make their own assessments of whether they will end up adding or destroying value, and reassess the pricing of these companies accordingly.
• AI product companies will jockey for position in the market, looking for ways to differentiate their products and preserve pricing power. With compute power being a key driver of both AI model power and growth in revenues, AI companies will try to not only get more AI infrastructure under control, but, if it is scarce, lock it for the long term. In the aftermath, AI product companies with a duration problem, where AI compute commitments are locked up long term, while their revenue streams are still in their formative phase and subject to short term volatility. Again, the forward-looking nature of markets implies that the pricing of these companies will look high, relative to current operating numbers, but that pricing will be volatile, as markets reassess prospects.
• AI users will find themselves on the receiving end of sales pitches for AI products and services, with the key selling point being improved productivity, with the warning included that not buying these products will put them at a disadvantage, relative to competitors.  For AI to find its business feet, you need the customer base for AI products and services to become both wider (cutting across many different industries) and deeper (where customers spend far more on AI, on a repeat basis). 

There are three broad scenarios that can unfold.  The first is that AI products and services work much better than expected, and that the demand for these products runs ahead of expectations. In this scenario, AI product companies will benefit first, since they will see growing revenues and higher pricing power, which create the unit economics and economies of scale for escalating profits, but the benefits will ripple to the AI builders, allowing them to increase capacity, which in turn will lift AI supplers. The second is that AI products and services work about as well as expected, with demand matching up to expectations. In this case, AI product companies should, at least in the aggregate, hold their own, but as with any competitive business, some will gain at the expense of others. There will be both business consequences, as weaker players drop out, and market implications, as the market adjusts the pricing for the survivors. The third is that AI products and services work less well than expected, with demand and monetization falling well below expectations. This will hurt not just the AI product companies who will struggle to meet revenue targets, but ripple through the other groups, as product companies either break or renegotiate compute commitments, hurting builders, who in turn, will shut down or abandon infrastructure capacity, creating pain for AI suppliers.

Commonalities and Competing Interests

    Even though the businesses in the AI space have very different business models, there is one common element that holds them together; the value of all of the players in the ecosystem ultimately will be determined by the size of the AI market and its profitability. As I noted in my earlier posts on AI, if the end market for AI products and services is small, because of the imperfections as AI tools, regulatory restrictions or limited reach, the entire AI ecosystem (builders, suppliers and product businesses will all feel the pain. Conversely, if AI delivers on its promise as a powerful productivity tool that works across many businesses, all of them benefit, albeit in different forms and different degrees. If history is a guide, it is the product companies that will be the primary benefactors, but the central role of compute power in the AI business may give AI builders and suppliers more of a share. In addition, if one of the groups faces growth and survival challenges, those pains will spread into the other groups as well. That may explain why you see so much circular financing in the AI ecosystem, where financially better- positioned and stronger players like Nvidia invest in or offer financing guarantees to weaker players, either to ensure survival or to provide an impetus for continued investment spending.

    Those common interests notwithstanding, these businesses also have competing agendas, and decisions and actions that affect one group negatively can create benefit for others. Thus, if AI builders choose to build too much infrastructure, too quickly, it will hurt them and their shareholders, but that overbuilding will be a bonanza for the AI suppliers, as chips, equipment, power and real estate players all benefit from the overbuilding. Once built, it is in the AI builders' interests to lease out the infrastructure for longer periods than merited, locking in their revenue streams, but for the AI product companies, those long term commitments can increase default risk. In short, as the AI business matures, I would expect these competing interests will often overwhelm the common interests, creating divergence. 

    For much of AI's early existence, the commonalities have overwhelmed the competing interests, with the hope of a large market for AI products and services pushing up market pricing for all of the groups, with AI chip companies like Nvidia being rewarded for the expected growth in AI architecture, hyperscalers seeing their pricing marked up, when they spend more on architecture, and the compute power commitments atAI product companies being viewed as a competitive plus, rather than minus. That, in a nutshell, has been the AI trade, but as the AI disruption moves through the disruption, it is inevitable that this will change, with the market differentiating across groups, picking winners and losers.

AI Companies - Letting the Data Speak!

    With that long lead-in, I think that we are better positioned to take a look at the AI trade, both in terms of where it is right now and where it may be going next. I do believe that the easy AI trade, where you could buy any company that was part of the AI ecosystem (supplier, buyer or product) and make money from a general optimism about AI’s potential is mostly done, and there are many traders who can point to profits they made from that trade. While last week’s market collapse in AI stocks may have drawn attention, the AI trade has been far shakier in 2026, as strength is some AI groups has masked weakness in others, reflecting a market trying to discern differences across companies in the space. To get an empirical measure of how the different AI groups have performed this year, I create samples of companies in each of the three AI groups (supplier, builder, user) and before you take issue with my choices, I will concede on four points:

• Public vs Private: Data availability and the requirement that there be a market price restricts the sample to publicly traded companies. That not only reduces sample size and creates sampling bias, but it also has disparate effects across the different groups. In particular, two of the biggest players in the AI products grouping, OpenAI and Anthropic, as well as almost all of the smaller players (Mistral, Qwen and Deepseek) do not make it into the sample, because they stayed privately held, and SpaceX enters the public company sample in the middle of 2026.
• Missed companies and changing sample: Since there is no separate accounting categorization for AI-related revenues and expenses, I relied on public descriptions of the companies operating in each of three AI groupings. Those descriptions are likely to be incomplete, but the companies that are missed are smaller and lower-profile companies, in terms of both operating metrics and market capitalization. In addition,  as companies go public or exit the market, there will be changes in the sample.
• Not pure plays: Most of the companies in this sample operate in other businesses, many of which pre-date their entry into AI. This is especially true of the hyperscalers, where Amazon, Alphabet, Meta and Microsoft have legacy businesses that generate large earnings and market value. That will make it difficult to isolate the AI effect on market capitalization, because changes in market cap at these companies will be driven not just by their AI actions, but by legacy business performance.
• In multiple groupings: Some of the companies in this sample operate in more than one grouping. Meta, Alphabet and Microsoft, in addition to be major investors in AI cap ex, are working on their own versions of LLM (Muse for Meta, Gemini for Alphabet and Co-Pilot for Microsoft) to enter the product market. I have included SpaceX as an AI product company, but at the moment, almost all of its revenues on its AI arm come from leasing out its compute power to Anthropic and Alphabet, making it more of an AI builder than AI product company.

With those caveats in place, here are the public companies that I have in each grouping:

Not surprisingly, given where AI is in the disruption cycle, with AI architecture is in its build phase, and products still nascent,  we have far more publicly listed companies that are involved in supplying the materials needed for the architecture - chips, electrical, power and real estate/construction, than in the other group.  To get a measure of how each grouping has performed as the AI boom has unfolded, I looked at the aggregated market capitalization of companies in each group from 2022 to 2025:

As you can see, the market capitalization of each of the groupings surged over the three year period, with AI chips companies having the biggest increase, in both dollar and percentage terms, over the period; Nvidia was the biggest contributor. The AI product segment also surged, but the value of the publicly traded companies in this space is small, as the biggest companies (OpenAI, Anthropic and xAI) were still privately owned during this period. The stock returns of AI builders lagged other AI groups but, but  their market cap increased 180% between 2022 and 2025, even as they were spending hundreds of billions in AI cap ex, with little to show as returns on that investment. Cumulatively, AI companies outpaced the rest of the market substantially over this period, rising as a percent of the market.

    Have things changed in 2026? To answer that question, I looked at how these AI groups have performed during the year:

AI products group percentage change does not include SPCX

Looking at the year-to-date numbers for 2026, AI chips and AI real estate have run ahead of the market, and the rest of the AI groups lag the market; again, you should read the AI products and services grouping with caution, because it has only three companies in it, and SpaceX made its entry into the group mid-year. In fact, AI companies, in the aggregate, increased their share of overall US equity market capitalization for much of the year, peaking at about 37% on September 30, 2026. Since the end of the third quarter of 2026, though, every AI group has struggled, with the exception of AI real estate & construction. Last week, in particular, was a testing one for AI companies, with Thursday (November 5) standing out and in the figure below, I report the dollar value changes of each group by day:

On November 5 alone, AI stocks lost almost $1.8 trillion in market cap, with chip stocks taking the brunt of the hit. During the course of the the last week, the AI chip and AI electrical groupings took a beating, but most of the other groups held their own. In fact, the hyperscalers, with the four Mag Seven companies in their midst, actually saw an aggregate increase in market cap during the week. It is a small sample, and there is a danger that I am overreaching, but if the immediate aftermath of last Tuesday's elections is a slowing down of data center construction, it stands to reason that the first companies to be affected will be the AI suppliers, who are dependent on growing construction for their growth. In fact, given that shareholders in the hyperscalers have become increasingly concerned about the magnitude of AI cap ex spending, they may very well welcome a slowdown.

An Assessment of AI as investment: The Big Market Delusion!

    I would like to think that I have kept an open mind about AI's business potential, in my writings, and I have been taken to task by AI advocates, for not being optimistic enough about the size of the AI market, and by AI skeptics, who believe that I am inflating its size. In almost all of my writing about AI, though, I have argued that it is exposed to what I call the big market delusion, where the potential of a big markets in conjunction with overconfidence on the part of founders, VCs and CEOs, leads to over investment and a correction. The picture below is the generic version:

I am seldom certain about anything in markets, but I believe that AI, because of its reach and potential, almost certainly has led to over investment, by AI builders, over commitments to compute power by LLM companies and over pricing, by the market, at least in the aggregate. The correction, when it comes, will cause pain, but the pain will be distributed unevenly, and for those who believe that AI has potential but have found the companies in the space over priced, it will provide opportunities. Before you decide that this (the over investment, over pricing and subsequent correction) is a problem, it is worth noting that this phenomenon is a feature, not a bug, that is associated with uncertainty and big change, and that it is, in spite of the costs of the cleanup, is a pre-requisite for change.

    This big market delusion suggests that you should either not buy any of the companies in the AI space or even sell short on these companies, but here is why you may not arrive at that conclusion:

1. Uncertainty about correction (when and what form): Even if you are right that investor and business overconfidence has lead to overreach, you need the market to come to agree with you, and it is difficult, if not impossible, to try to predict when in will happen, and what form it will take. The obvious response that many investors have is to just avoid AI companies for the moment, but since it is just a large portion of the market, this becomes a variant on market timing, with questionable payoffs.
2. Catalysts for correction: The catalysts that trigger the correction can  be bad news about a leading Ai company, a macro or political event that makes investors reassess the landscape, and history offers little useful guidance. The big market delusion is ultimately built on a behavioral characteristic (over confidence) and the correction has as much to do with investor psychology, as with financial fundamentals. It is one reason why I am not quick to dismiss charts and technical indicators (price and volume) as investing voodoo, since these tools may be better attuned to detecting these mood and momentum shifts. 
3. Acting in the fact of uncertainty: Since the correction process is difficult to time, if you are an investor of long standing in one of AI’s many winners, you should use each correction phase to sell a portion of your shares. You will give up upside, by doing so, but you will not be completely exposed, when the correction happens. If you decide to buy AI stocks after the correction, don’t wait for the bottom, since you may never hit it. Find the companies in the AI marketplace that you would like to own, set a trigger pricing or a valuation at which you would buy, and automate the process. Thus, if you like SpaceX, but only if its market cap is $1 trillion, put in a limit buy at that value, and let it get filled automatically. 
4. Go where it is darkest: One of the themes of my valuation and investment philosophy classes is that the payoff to assessing fundamentals and valuing companies is greatest in sectors and at times of maximum uncertainty. If you are an AI advocate, and you have generated supersized returns over the last four or five years from just buying companies in the AI space,  you are entering a phase, where  even if your AI optimism holds, you should prepare yourself for more periods like last Thursday, where the market mood on AI darkens, and use those periods to be more discriminating in your AI investment choices.

For those of you have read this far expecting a clear signal of whether now is the time to sell or buy, I know that you may find my conclusions to diffuse to be actionable. 

Implications for Anthropic IPO

    Since I have spent much of the last few weeks valuing Anthropic, ahead of its IPO, and following up on that valuation, there is no avoiding the question of what the market and political turmoil in the two weeks will do that IPO. I have seen rumors floating around of a cut in the IPO pricing (from the initial estimate of $1.5 to $2 trillion to a lower number and some that the IPO may be delayed into next year. 

• As I noted in an earlier post, IPOs are priced, and in young businesses, with uncommon metrics. After not hearing talk of annualized revenue run rates for a few weeks, there was a leak that Anthropic's ARR for October hit $80 billion, which was touted as both good news (by some) and bad news by others. The truth is that pricing is a mood and momentum game, and it is undeniable that the mood has darkened not just for AI, but also for the market in the last two weeks. If you consider bankers who are terrified of over pricing a trillion-dollar plus offering and picking up the pieces, the pressures to lower the offer price only get stronger.
• On the valuation front, it is not market mood that matters but events that change the fundamentals. The most critical of these were the election results, where voters sent a clear message about their aversion to data centers. In Texas, which is home to about a third of the new data centers coming online in the near term, it may not stop existing data center construction, but it will slow the growth of data centers, at least in the United States. It is also possible that the newly elected congress will write in tighter restrictions on AI, creating a bigger drag on margins.  For those who are more focused on seeing the glass as half full, this will also vindicate Anthropic's rush to lock in compute power commitments, before the rules of the game changed, and the limits on data centers may give Anthropic a moat against competitors. Just to give you a measure of how sensitive the Anthropic valuation is to these changes, increasing how much Anthropic will have to invest to add compute power in future years (by lowering the sales to cap ratio from 3 to 2.5) and increasing the drag on margins from regulation (from 5% to 7.5%) reduces the estimated equity value to $1.25 trillion, even if you leave the target revenues at a trillion dollars.

I have no sources in investment banks, but I think that this IPO is too far along for bankers and the company to withdraw or delay, without markets reading worst case scenarios into their actions. That said, I would not be surprised to see Antrhopic priced closer than $1.25 trillion, with perhaps a smaller cash raise (with fewer shares being offered), allowing for that "price jump" on the trading day and a buzz that carries over.

Conclusion

    As AI moves from the hype phase to business building, I hope that we stop talking about  "the AI trade",  as a catch all for AI companies, and start differentiating between the different groups for two reasons. The first is that as the AI business starts to take form, there will be developments that are good for one group and bad for others; the election results clearly affected AI suppliers more negatively than it did AI builders. The second is that when you have crises, the market will lose its head and mark down all AI companies indiscriminately, offering you the chance to buy those that you believe should not have been marked down.   I also think that it is time that we stop making overly broad assessments of the AI business that lock us into our priors. Thus, if you start with the presumption that the entire AI phenomenon is a "bubble" (notice that this is the first time in the post that I have used that word), your mind is already made up and no AI company, no matter what group it belongs to, will be worth its price. That view, no matter how strongly held, would have kept you out of a segment that has added almost $18 trillion in market capitalization between 2023 and 2026, and even with a strong market correction, it is likely that staying out of the AI bubble would outweigh what you would have earned by investing in some of these companies. At the other extreme, if you are so dazzled by the technology that you see this as a sure win, you are setting yourself for disappointment, because not only are there no sure wins in equity markets, but the pathway to AI nirvana will be come with challenges that test your faith. In short, stay with your beliefs about AI but leave the door open for different opinions.

YouTube

Spreadsheets

1. Anthropic IPO valuation: A post-election update

Blog Posts on AI in 2026

1. To a Trillion Dollars and Beyond: A SpaceX Odyssey!
2. Revisiting SpaceX: A post-Prospectus valuation
3. OpenAI, Anthropic and SpaceX: The S&P 500 Inclusion Question and Investment Consequences
4. AI Scenarios: From Doomsday to Do-nothing!
5. AI's Bar Mitzvah: From hope & hype to business questions!
6. Stock Prices, Earnings and Cashflows: The AI Effect!
7. From Eyeballs to ARR: Buzzy Pricing Metrics and the Anthropic Pricing
8. The Anthropic Prospectus Arrives: My (but not the) valuation of Anthropic
9. Heeding the Anthropic Feedback Loop: Cost of capital delusions and distractions!

The Anthropic Feedback Loop: The Cost of Capital Delusion and Diversion!

I knew that my post on Anthropic would draw strong reactions, when I made it public last week. I did mention in the course of the post that I have no interest in selling my estimate of value to the rest of the world, both because I don't own the high ground here and because it is meant to guide my decisions, not yours. That is why I made my model open to changes that you can make to my story and inputs, and while you have to trust my mechanics, I tried to carry your inputs into value. Much of the pushback I have received has been thoughtful and productive, some of it has been emotional (either from people who love Anthropic or from those who would not touch it with a ten-foot pole) and some of it, especially on the cost of capital, reflects a fundamental disconnect with market history, current market pricing and just basic business sense. I plan to play devil's advocate with my own valuation, looking at my story inputs, and ranking them in order of fragility, partly to stress-test my value estimate and partly to reinforce some corporate finance first principles, and then move on to your feedback and what I take away from it.

Playing Devil's Advocate: A self-critique of my Anthropic valuation

    Since I posted my valuation only a week ago, rather than rehash how I approached the valuation and my end results, you can read the gory details in the post that accompanied it. I will start with a summary of the valuation, though, and then list  the parts of my story that I believe are most directly driving my value estimate on both magnitude (how much it is affecting value) and fragility (how uncertain I feel about the part). 

The IPO Valuation

    I did summarize my valuation of Anthropic in a figure in my original post, but since there was that much to compress into that picture, I will try a different format for the same here:

Download updated spreadsheet

If you check against my estimates from my earlier post, you will notice that the equity value at $1.45 trillion is slightly lower than my prior estimate of $1.475 trillion, as I adjusted the trailing 12-month revenue and operating income numbers to include the just-leaked third quarter numbers. Note that replacing the third quarter revenues from 2025 with the third quarter revenues of 2026 raises the base revenues, and since I still believe that the revenues for the coming twelve months will stay at my original estimate of about $100 billion, the growth rate in revenues in year 1 has been adjusted down to deliver that result. This reframing allows me to add a few details that I might have glossed over in my last post, as well as correct errors (none of which thankfully changed by value estimate: 

1. Value per share: Though I have a share count, created by adding the shares reported as outstanding in the balance sheet to the restricted shares granted to employees, that share count will change before the public offering, as bankers adjust it to deliver on a per share value that they believe will make the company more "sellable". I will revisit this number, and update it when the company does (and it has to, before setting an offering price), but it should not change the aggregate value of equity.
2. Cost of capital choices: I used a 9.74% cost of capital which incorporates risky technology cost of equity and a cost of debt based on a BB+ rating for the company, putting the company at close to the 90th percentile of US companies in 2026. I know that some of you felt this was too low, and I will come back to address what a cost of capital can and cannot do for you in valuation, later in this post, and why I don't think that this is the input that most moves the valuation.
3. Compute commitments: I wish that Anthropic had created a table in their prospectus, where they specifically listed total commitments, both contractual and total, by year, similar to the table that they have for purchase commitments. Instead, they have sprayed content commitment numbers in the text, with little detail on timing and contractual standing, which results in differences of opinion in what those commitments are., with estimates ranging from $400 billion to $800 billion, with no clarity on timing or contractual terms. In my valuation, my estimate of contractual commitments results in an aggregated value of just over $450 billion and a present value of about $373 billion. It is possible that these numbers will need revisiting as details emerge of these commitments, but this will remain a long term issue for anyone valuing or pricing the company. To offer an imperfect analogy, I struggle with dealing with content commitments at Netflix, whenever I value the company, and especially so in the early years, when the entire Netflix library was leased or rented from content providers.
4. Dilution Risk: Anthropic, like many young companies, has used its equity as currency, paying employees with restricted stock and options, and used it as an add-on to borrow money in the form of convertible bonds. In addition, if the company needs to raise capital in future years, it will have to issue additional shares for that purpose, and the larger its ambitions, the greater these issuances will be. When valuing Anthropic, there is no escaping this dilution overhang, and here is how it is incorporated into the value per share. With stock-based compensation, I valued the equity options (at their option value, rather than exercise value) and netted that number ($472 million) from my estimated equity value to get to value of equity in common stock; I also included the restricted shares as part of my share count to compute value per share. With the convertible debt, I have assumed that the marked-to-market number (it was the marking to market that created the $72 billion hit to earnings in 2026) is close to fair value, and subtracting it out will remove the conversion option from the equity. The future share issuances are to fund negative free cashflows, and since those negative cashflows are already weighing down my present value, there is no need to add the shares issued to the share count; adding it will be double counting.

If you are surprised that adding the third quarter numbers did not move my value by much, it is because it was an open secret that Anthropic's third quarter revenue growth had slowed and I had already incorporated that into my valuation story.

Value Drivers

    The equity value that I obtain for Anthropic is close to the lower end of the offering price estimates, but as I noted at the time that I posted it, and will reinforce again now, it is a function of not just my upbeat story about the company, but also one about how the AI products and services market will evolve over time. To the question that some posed about whether I was concerned that I could be wrong, of course I am, but as I will argue in the next section, I cannot let that concern paralyze me. I would, of course, rather be right than wrong, but I am trying to forecast the future, and it is a guarantee that I will be wrong. That said, I am more uncertain and exposed on some parts of my story line than others, and here is how I would rank them for Anthropic, from most critical to less so:

1. Target Revenues: If I were forced to pick one variable around which my valuation revolves the most, it is my estimate of the target revenues for Anthropic, which I estimate to reach a trillion dollars in ten years. Not only would this put them at or close to the top of all companies, in terms of total annual revenues, but they will also be sharing that grouping with companies like Amazon and Walmart that deliver on scale, but at the expense of margins. The revenue assumption is also inextricably linked to not only what you think AI agents and products can do, but also how demand for these will be monetized. In addition, there is a company-specific component to this story, where Anthropic emerges as a winner in the premium portion of the AI market. capturing a dominant market share, with pricing power.
   
   
   
   The fragility of the target revenue assumption comes from it being susceptible to any of the parts failing, and if a part fails, revenues will end up below a trillion, and perhaps significantly so. That, in turn, will have ripple effects on the other parts of your story and inputs, with lower revenue growth pushing down margins, creating idle compute power and increasing the risk of failure. 
2. Governance: I did use harsh words to describe the governance structure at Anthropic in my last post, describing the combination of its structure as a public benefit corporation and a voting share concoction that leaves founders in control, no matter how much capital the company raises in equity markets, as a red flag. I did get some pushback from some arguing that this is a good feature, not a bad one, since these are smart people who understand AI better than anyone else in the world. Even if you accept these views  I would still argue, perhaps even more forcefully, that concentrating power in the hands of founders is a bad idea, and here is why. Anthropic has a plausible or even probable pathways to business riches, but those pathways come with setbacks (technological, regulatory, competitive), and you need management that can respond quickly and decisively, sometimes abandoning existing policies. For the latter to happen, you often need outsiders, not as immersed or even as informed on the technology front, but with different  backgrounds and experiences, to call attention to management missteps and push for change. I don't know any of the founders of Anthropic, and this may be unfair, but they seem to share a common worldview and operate in an echo chamber. My concern, as an investor, would be that they will not see changes occurring in business economics, and even if they do, be unable to or unwilling to adapt to a changing environment, and for Anthropic, that may be fatal.
3. Profitability: My story for Anthropic is that it will not only be able to deliver operating profits soon, but do so on a sustained basis, with a target operating margin of 32.5% (before taking the regulatory risk haircut). Implicit in my profit story are assumptions about pricing power, the unit economics of the AI business and economies of scale and each is breakable:
   
   
   
   One of the most frequent critiques I received in responses to my Anthropic valuation is that this is a business without moats, and while I agree that moats matter, I am not sure that I go along with the notion that this business will get commoditized. The truth in technology is that unlike consumer products or old economy companies, where the moats take the form of brand names, patents or cost advantages, the companies that win find ways to make their products stickier to their customers, where stickiness implies that customers will not switch to alternatives, even if these alternatives are technically superior and cheaper. I recognize that Anthropic has work to do in creating stickiness, but I can visualize ways in which Claude agents, twinned with client data on customers and employees, can become sticky, allowing the company to preserve its pricing power and market share.
4. Continuing investment needs: While much of the focus on Anthropic has been on their compute commitments, disclosed (loosely) in their prospectus, my valuation of Anthropic is driven by expectations of what those commitments will look like over time, and especially if my growth story for the company comes into fruition. Delivering a trillion dollars of revenues will require Anthropic to not just maintain and replenish the compute power that is has already locked down, but add to that power over time. Implicit in my story line for the company, where the company is able to add this compute power at lower costs (reflecting declining chip prices) and in time, is the assumption that the compute power story moves along smoothly, working through resource constraints and pushback: 
   
   
   
   If you have been tracking news stories about data centers, it seems clear that building new data centers is getting more difficult, as local communities push back, and politicians realize that opposing them is a net vote winner. 
5. Regulatory risk: You may find this surprising, but risk, in my story, is not as critical as my first four inputs, partly because it is determined by shifts in operating variables. Put simply, Anthropic will become riskier, if revenues do not measure up to expectations and the pathway to profitability becomes rockier, making it more an output than an input variable.
• I know that for some of you, the cost of capital of 9.74% is the main stumbling block, but I will provide a more cohesive (and hopefully) more persuasive argument of why it is not as critical to value as you may believe it is.
• I also know that for others, the notion of a money-losing company borrowing hundreds of billions, either directly as debt or through contractual commitments, creates a failure risk probability that is so high that the rest of the valuation is moot. I agree with you on the general argument that some players in the AI ecosystem, such as Coreweave, have abandoned corporate finance first principles and risk bankruptcy, but Anthropic is no Coreweave, especially if you consider that most of its debt is in contractual commitments spread out over time and that it will have a cash balance large enough to cover its first few years of those commitments.
• It is on the regulatory risk front that I worry the most about being wrong, because regulatory risk is as much driven by politics as it is driven by economics. You could be one major election away from onerous regulations on AI, where unlike my base case, where those regulations cost you only a small slice of your operating profits, they take a major bite out of revenues and profits. 

In sum, not only will I be wrong on any of my key inputs, but being wrong on one (especially target revenues) will cause me to be wrong (in the same direction) on my other inputs, causing value to shift more dramatically in response. 

Investment Followthrough

    At this point, you are probably why, if I am this uncertain about a company's future, I would even consider investing in it and it is a question that I have pondered as long as I have been an investor. Uncertainty is a feature of investing, no matter what you invest in, with the differences across investments lying in how much uncertainty you face, rather than whether you face. There are four responses that an investor can take to uncertainty:

1. Ignore it: There are investors who are capable of ignoring uncertainty and going in fully even on the most uncertain investments. They are able to do so because they are far less risk averse than the rest of us and are willing to live with the consequences of investments (and their portfolios) blowing up. They are more gamblers than investors, but with good instincts and timing, gamblers can get wealthy too.
2. Avoid it: If you completely want to avoid uncertainty, you cannot invest in equities, corporate bonds, which have risk of default, and even treasury bonds, which have price risk from shifting interest rates. Unless you want to hold just hold cash, you have to therefore decide what your threshold for risk is, and stay within that threshold. In conventional value investing, that would lead you to restrict your investments to companies with assets in place (captured in book value) and earning power from those assets (high current earnings and cash flows) and stay away from growth companies, since growth is "speculative" and therefore risky. While I understand the impulse, this choice comes with a cost, which is that your portfolio will be almost entirely invested in mature and declining companies, with no growth companies in emergent technology. 
3. Margin of Safety: The third approach to dealing with uncertainty also borrows from conventional value investing and imposing a margin of safety, where you will buy a stock only if its price is less than your value estimate by a margin (20%, 30%, 40%) that you specify. While your inclination, if your risk averse, is to keep increasing your margin of safety, it is worth remembering that like almost everything else in investing, it creates a trade off between type 1 errors (where having a higher margin of safety protects you from the losses of investing in stocks that you should not be investing in) and type 2 errors (where that higher margin of safety keeps you from investing in good stocks that you should be holding):
   
   
   
   The question of whether you should have a margin of safety, and if so, how large it should be, will be a function not only of how many opportunities you see in the market, relative to your portfolio size, and how much cash you have in your portfolio (with larger cash balances suggesting that you should lower your margin of safety).
4. Law of large numbers: If I had a million dollars in investable funds, would I invest that entire amount in Anthropic, even if it trades at well below my assessed value? The answer would be no, given the uncertainties associated with the future of the company, but that answer would change if my portfolio became less concentrated, since its performance would be less a function of any individual company's performance (good or bad). In effect, I am using the law of large numbers as an ally to not only reduce my risk exposure to individual companies, but also as a basis for making investments in far riskier stocks than I would be otherwise inclined to invest in.

As you assess your story and valuation of Anthropic, it is crucial that you frame the question that you are trying to answer with your analysis. Is it whether you should buy or sell short Anthropic, as a stand-alone investment or are you looking at Anthropic as an add-on to an existing portfolio? If it is the latter, how concentrated (across stocks and sectors) is your existing portfolio? The answers will vary, as a consequence.

Keeping the feedback loop open: Your pushback on my valuation

  In the days since I posted my Anthropic valuation on multiple platforms (X, LinkedIn, Google, Substack and YouTube), I have heard from many of you, and in the interests of keeping the feedback loop alive, I did a breakdown of two statistics - the percent of people who thought that I was over valuing and undervaluing Anthropic, and among the majority who thought I was overvaluing, their reasons for disagreement. I used the Google shared spreadsheet that I created last week, where some of you were kind enough to enter your story/inputs and value for the company, and here is what you found:

Clearly, a lot of people disagree with my assessment, with more feeling that I have over valued Anthropic, rather than under value the company, with the crowd assessment of its value closer to $1 trillion, rather than $1.5 trillion. I also used your inputs from the shared sheet, as well as a reading of the comments on my valuation, across platforms, to see the source of your disagreement, especially if you believed that I have overvalued the company.

Among those who disagreed with me, the reasoning varied, but I was disappointed, albeit not surprised, that the biggest single reason outside of "my gut tells me that Anthropic is worth nothing" was a judgment that my cost of capital estimate for the company (9.74%) was much too low. I was disappointed because I don't think that the cost of capital is at the center of the DCF universe (and I used words to that effect in my original valuation) but the lack of surprise at its prominence comes from observing appraisers and analysts obsess over this number, often without understanding what it is designed to measure and without a frame of reference (as to what a reasonable value for it is), and using it as a receptacle for their hopes, fears and fantasies.

The Cost of Capital - The Swiss Army Knife of Finance

    I teach and practice across three different areas of finance - corporate finance, which looks at the first financial principles that govern how to run businesses, valuation, which looks at those businesses and tries to attach a value to them and investing, which is about finding market mistakes and taking advantage of them. I describe the cost of capital as the Swiss Army Knife of finance, because it is in every aspect of finance, performing different roles for different players (businesses, investors, activists). I won't bore you with an extended discourse of all of these roles, which I have expanded on in a practitioner paper on cost of capital, but you can get a glimpse of its uses in the figure below:

For many analysts and bankers, the cost of capital is the discount rate that they use to value businesses, but its usage in corporate finance is deeper and more impactful. It is the hurdle rate that businesses use to decide whether to invest in a project or make an investment, tailored to the risk in the business, the optimizing tool that businesses can use, if they so choose, to determine the right mix of debt and equity to fund themselves and the divining rod to determine whether businesses should be returning more or less cash to their owners.  It is true you can break the cost of capital into its components, by categorizing the funding into debt and equity and attaching a cost to each:

At the risk of stating the obvious, all debates about cost of capital become discussions of what comprises a reasonable cost of equity, what the cost of debt should measure and how they should be weighted.

Estimating the Cost of Equity - Checks and Constraints

    While there is much to be learned from modern portfolio theory, one of its more deleterious consequences has been a dumbing down of the cost of equity, where intuition and common sense are supplanted by heated arguments about asset pricing models and parameters.  To bring the discussion back to steady ground, start with a recognition that the cost of equity for an investor should reflect what investors can earn elsewhere in the market on an investment of equivalent risk, i.e. it is an opportunity cost,  and understand that is what you are trying to measure, often imperfectly, in every model. That number has subjective elements, and can vary across investors, based on risk aversion, but the requirement that you can earn that number elsewhere creates reality checks. 

a. History

    What do you think the cost of equity for the average US company is today? In an unbounded world, you can give me any answer you want, from 3% to 30% or higher, but one check you can run is against history. On my webpage, I maintain a dataset of historical returns on US stocks from 1928 to 2025, and in the graph below, I compare the average annual returns you would have earned on US equities (all as well as just small cap) over 25 years (2001-2025), 50 years (1976-2025) and close to a hundred years (1928-2025) to annual returns on treasuries (3 month and 10-year):

There are two ways this historical data can be used to frame cost of equity estimation. You can use the average annual return on stocks over the period, which ranges 8.74%, if you look at just this century (2001-2025), 11.92%, if you go back fifty years (197-2025) or 10.13%, if you go back all the way to 1928 (1929 -2025), and use that as your cost of equity. Since investors had the alternative of investing in treasuries, and rates over treasuries have varied over time, a second option is to look at the premiums you would have earned investing in stocks instead of treasuries. That number has a tighter range, between 5.5% and 5.75% over the 10-year bond and 7-7.5% over treasuries; using the 10-year treasury rate of 4.9% on October 26, 2026, that would yield a cost of equity of about 10.4-10.65% for the average risk firm. There is value in knowing history, but it is worth recognizing that these returns capture stock market performance over a century in the United States,  the most successful equity market in the world over that period. There is a selection bias that may be leading to overestimate future returns, if you use historical averages. That would make using a cost of equity higher than what this history delivers even more indefensible.

b. Market-Implied

    If the essence of cost of equity is estimating an opportunity cost, i.e., a return you can make investing elsewhere, I would argue that a better way of estimating the cost of equity for a company is to back it out of market prices. In effect, you can start with what the market is pricing equities in the aggregate, estimate expected cashflows from buying these equities and solving for the expected rate of return. If that sounds familiar, it is the process I use at the start of each month to compute the implied expected return and resulting equity risk premium for the S&P 500:

In a post earlier this month, I reported on what those numbers are at the start of October 2026:

Note that the expected return from investing in US equities, in the aggregate, at the start of October 2026 was 8.99%, and this number is model-agnostic. Thus, you can choose to believe or disbelieve the CAPM and think that betas are a waste of time, but the expected return from investing in US equities will still be 8.99%. I know that there are estimates of growth and cash flows that help determine the number, but almost every pushback I get (the cashflows cannot be sustained, the growth rate is too high etc.) will lead to a lower expected return, not a higher one. 

    So what? If this is the market you are in, contending that the cost of equity for a median US company should be 12% or 15% borders on the delusional, since you cannot make that return on an investment of equivalent risk in this market. To see why costs of equity that are 15% or higher are indefensible, you can check out the graph below where I report the expected return on US equities (using the implied premium approach) going back to 1960:

The expected return on US equities, and by extension, the cost of equity for the average US company has not exceeded 10% since 1994. If you are reaching for 15-20% costs of equity, you have to go back to the early 1980s, when the 10-year treasury rate was 10% or higher. 

c. A Pricing Perspective

    There is a third way to assess the reasonableness of a cost of capital, and this approach is especially effective if you think in terms of pricing multiples (PE and price to book ratios) and have a frame of reference on those (in the sense that you have a measure of what a reasonable multiple of earnings stocks should trade at). There is a simple (and perhaps simplistic) way to think about the cost of equity and capital in terms of pricing multiples. 

• You can convert a cost of equity assessment into a price earnings ratio, with relatively few assumptions. To see how, start with a simplistic valuation model where the value of a stock is the present value of its expected dividends over time, and the company is in stable growth:

If you believe that the cost of equity is 15% for the average US firm, given that nominal growth in the US economy is about 5% and that the average return on equity across US firms is about 20% (and that is a generous estimate), the PE ratio for US firms today should be roughly 7.5. Given that the actual forward PE ratio is closer to 20, you have two choices. One is to sell all your equities, though I am not sure where you would put that money, and the other is to revisit your cost of equity assessment.

• The cost of capital maps on to enterprise value multiples, using a similar approach, with a 12% cost of capital translating into an enterprise value multiples of 9 times operating income and  6-7 times EBITDA for the entire market, numbers that are vastly below today's values.

Put simply, if you content that the cost of equity for US firms is 15% and in the same breath argue that a PE ratio of 15 for the market is reasonable, you are being internally inconsistent, and you have to decide which of these assertions you plan not just stick with, but act upon, as an investor.

The Cost of Debt: Intuition and Estimates

    One reason that the Anthropic cost of capital discussion may have been hijacked is the almost mindless use of debt in funding data centers, a practice that I have attacked as senseless for both the borrowers and for the lenders. With Coreweave as poster child, this has led to an explosion of concerns about default risk and in  market-implied default rates and spreads for some companies in this space. By extension, Anthropic has become part of this discussion because of the compute commitments that it has locked in, for good and bad reasons. In fact, the argument for a higher cost of capital goes as follows. If it costs Anthropic 10% or more to borrow money today, it should have a much higher cost of equity to reflect the residual risk. 

    It is true that the cost of debt will tend to rise as worries about default risk rise, and that ratings operate as a shortcut to measuring that default risk. Using the default spreads that you can observe in markets right now, with a 10-year treasury bond rate of 5% as the base, here are updated default spreads and costs of debt, by bond rating:

As you can see, the default and cost of debt for companies rises as the rating drops and kicks into hyper drive at a B rating. You may have a hard time reconciling the very high costs of debt at the lowest ratings with the earlier table that constrained cost of capital, but that is because at the lowest ratings, the risk of default is imminent and large. It would be inconsistent to incorporate this cost of debt, in its entirety into a cost of capital, a going-concern concept, and that is why I separate out failure risk as a way of capturing that risk. 

    My estimate for the cost of debt for Anthropic reflects my estimate of a bond rating of BB+, and a recent news story in the Financial Times suggests that they will remain below investment grade. Anthropic may owe a lot in commitments, but they are spread out over time, with about $70 billion coming due next year and $76 billion the year after. Those are large commitments, but Anthropic has a cash balance of $114 billion, which will increase to over $200 billion after the IPO. There are lots of reasons to take issue with Anthropic, but worries that it may default in the next year or two or even the next five don't make it to my top ten list. That does not mean that I am ignoring the risk of default, because I am incorporating a failure risk of 10%, where Anthropic does run into headwinds, is unable to meet its contractual obligations and has to have fire sales. If your concerns about default are higher, teh failure risk is the number you should be moving, not the cost of capital. In fact, if you were ignore failure risk, as many analysts do, and try to jam default risk it into cost of capital, you would need a cost of capital of about 11% to arrive at an estimate of value close to mine, in my valuation.

Cost of Capital - The Cross Section

    I know that you may still argue that if you have a very risky company, and you throw Anthropic into that grouping, you could have a much higher cost of equity than a median company. That is true but you may be vastly over estimating the divergence in costs of equity across companies, when you do so. At the start of every year, I estimate the cost of equity for every public company in the world, and graph the distribution to get a sense of what high, low or typical numbers look like, and the start of 2026 numbers are shown below:

In almost every sector, take heed of the tight range between the first and ninth deciles. In technology, a sector where investors perceive wide differences in risk, 80% of all firms have costs of equity between 9.88% and 14.49%, and costs of capital that range from 9.02% to 12.27%.  When you have this little room to diverge, you can see why spending your time finessing the cost of capital is an exercise that will bear little fruit, especially if you also factor in that your starting numbers for these costs in valuation will change over time, as your company becomes larger and more profitable.  This table is in US dollars and for US companies, and if you are working with a different currency and with a company in a different country, you should add the differential inflation and country risk premium to the cost of equity. In the table below, I have the adjustment computation for a subset of currencies/countries:

Thus, if the inflation rate is 1.77% higher in Indian rupees, and the Indian equity risk premium is 2.49% more than the US, you should add 4.26% across the table to get the ranges in rupees for Indian companies. A median Indian company would therefore have a cost of capital of around 12.05%, in Indian rupees, and a median Swiss company would have a cost of capital of about 5.85% (adjusting for an inflation rate about 1.71% lower than the US and an equity risk premium about 0.23% lower).

    It is true, based on this table, that you could push the cost of capital for Anthropic higher, perhaps towards the 90th percentile of technology firms, and give it a cost of capital of 12%, but there are two caveats. The first is that you would then have to set the failure rate to 0%, since the costs of capital in this table do not adjust for failure (and especially so at the higher risk percentiles). The second is that your cost of capital, in your valuation, has to change over time to be consistent with your story line for Anthropic as a company. Thus, if you see Anthropic scaling up revenues over time and turning the corner on profitability, its cost of capital will move towards a more typical cost of capital for tech companies over time. If you don't see that happening, you don't need a high cost of capital to conclude that Anthropic is worth little or even nothing.

    You will notice that nowhere in this discussion have I used a greek alphabet (beta) or a finance acronym (CAPM) because not only do I not need either to make my point about costs of capital but also because they have become easy excuses for laziness on the cost of capital front. I don't estimate costs of capital as an academic exercise, but for purely pragmatic reasons, since I need them in my corporate financial analysis and valuation. I need a relative risk measure for a firm to estimate how much higher or lower than the average company its cost of equity should be, and all that beta provides is an estimate of relative risk. If you like it, use it. If you don't it, come up with your alternative, and if you cannot think of any, try any of the following:

Put simply, if you don't like betas (or the CAPM), you should consider why (and it cannot just be because Buffett or Munger told you so) and follow through to come up with your proxy for risk. I am okay with any of your choices, since I can use them to estimate costs of equity and capital, and to be honest, none of them will make or break my valuation.

Cost of Capital Delusions - Investors and Businesses

    I know that there are analysts and investors out there who have picked a required rate of return as their hurdle rate, sometimes arbitrarily,  for investing and never revisit that number, and almost as many CFOs who do the same for the hurdle rates at companies. That number can be 10%, 12% or 20%, and when you push back, the response is that it their choice to make, and that they are just setting high standards. With individual investors and privately owned businesses, the consequences of setting an artificially high cost of capital are personal, and if, after being made aware of them, they stick with a 25% cost of capital, they are entitled to do so. With institutional investors and publicly traded companies, where you are investing other people's money, you have to consider the costs that they will bear for your overreach.

Professional Investors - Grasp exceeding Reach!

    With subgroups of professional investors, especially venture capitalists and private equity, the numbers that they offer as target or expected returns are so staggeringly out of sync with reality that the cynic in you should wonder whether these numbers are marketing props, rather than  financial metrics. Let's start with venture capital, where VCs routinely set target rates of 20%, 30% or even 50% for their investments, depending on how early in the process they invest, with angel financing commanding the highest rates. Private equity fund sponsors also present potential investors with investments that least on paper deliver internal rates of return of 20% or higher. We know that these target returns are fiction, because the returns that venture capitalists and private equity investors actually deliver are not just much lower but surprisingly close to what investors can earn on public equities, say on index funds. 

Cambridge Associates

Note that not only do VC and PE funds lag public equity indices over long periods, there is not a single number in these tables that comes remotely close to the target returns that PE and VC claim up front. 

    You may be tempted to cut venture capitalists and PE investors slack for using fictional target returns,  viewing them as sales and negotiating tools, but those fictional values feed back into how much founders have to give up in exchange for capital, and may also explain the failure of VCs and PE investors to deliver value for their investors. As a generalization, I believe that active investors who promise their clients too much (as an expected return) will overreach in terms of the risks they take, and if the target returns they promise are set too high (relative to what they should be), they will spend their time looking for ten or hundred baggers, rather than solid long term investments.

Publicly Traded Companies - Inertia and Herding

    While it is true that company hurdle rates are for internal use and not often visible from the outside, surveys of CFOs, asking them about what the costs of capital are for their company, often come back with the conclusion that hurdle are set about 2-3% higher than justified by market numbers. Often, these hurdle rates are either arbitrary or reflect the history of the company, where high returns earned on projects taken in the past are employed as hurdle rates for the future. You may be tempted to let this practice go, as just being conservative, and protecting companies from taking bad investments, but here again, there is a trade off. For companies that estimate returns on projects rigorously, having a higher hurdle rate than justified will mean the good investments will be left untaken. Early in a company's life, when project choices are plentiful, this may not be a deal breaker, but as companies mature, that underinvestment will cost them (and their investors). 

    There is another, and potentially an even more dangerous consequence that undermines the entire process of investment analysis. When hurdle rates are set unrealistically high, relative to what you can earn in the market, managers, recognizing that they will never find projects that earn more than those hurdle rates will start to skew cashflows and growth upwards to compensate for discount rates being too high. Top managers will then push up hurdle rates to compensate for the bias, and when managers respond with even more positive skews in cash flows, you have set in motion a process where both the cash flows and the discount rate become fictional and made-up numbers, rendering the entire process moot.

    Ultimately, though, as with PE and VC, the proof is in the pudding, and with publicly traded companies, the returns on equity and invested capital reflect what they actually earn on projects, rather than what they claimed they would in the original analysis. With my sample of publicly traded companies, in 2025, here is what the numbers look like:

As you can see from the 2025 statistics, and this is with my cost of equity and capital calculations from the earlier table (which you may have found too low), less than 30% of companies earn returns on equity (capital) that exceed their costs of equity (capital) and less than 20% beat their hurdle rates by more than 5%. If you raise the costs of equity and capital to 12% or 15%, that underperformance will only get worse. 

Individual Investors and Privately owned businesses

    As an individual investor, investing your own money, you do have more freedom on setting your required returns, but there will be consequences to setting unrealistically high hurdle rates. 

• If you stick with these "high" hurdle rates. you will find it difficult, if not impossible, to find equities that can deliver those rates. I can assure you that with a 20% cost of equity, for instance, there is not one firm in the S&P 500 that will be undervalued for you to invest in, and if you do find firms that look undervalued in the broader market, it is because they are basket-case firms on the verge of tipping over. Thus, a reasonable and logical follow-up would be to reduce or even eliminate your equity exposure in your portfolio, though if your unreasonableness on hurdle rates carries over into other asset classes, you will find no risky asset class that meets your rate of return criteria.
• More dangerously, it will leave you open to scams, where your search for high return investments will lead you invest with a "friend of a friend" or a "YouTube investment celebrity", claiming to have found a way to deliver more than your hurdle rate with investing alchemy.

I have long heard critics of intrinsic valuation argue that it does not work for them because it seems to find everything that they try to value to be over valued. That is less a problem with the approach and more a reflection of using unrealistically high costs of equity and capital in your valuation.

    For businesses with private owners, setting the hurdle rate too high will lead them to be more conservative in taking investments, opening the door to more aggressive competition, often from publicly traded companies or private business competitors with more reasonable hurdle rates. The bigger danger here again is when you get ready to sell, you will sell your business  for too low a price. You will facilitated in this process by private company appraisers who come up with costs of capital that are outlandishly high, usually getting there with what they call the "build up" approach, a  piling on of premiums that are either thinly sourced (a small cap premium that ceases to exist in 1981), double counted (an illiquidity premium, with an additional illiquidity discount of value) or just made up to (a non a company-specific risk premium,  where a number is added on to get to whatever discount rate you want). 

Conclusion

    This post started with a sell assessment of my Anthropic valuation, and I hope that you can see that while I believe in the plausibility of my valuation, there are multiple dimensions on which I can (and will) be wrong. In my view, the key assumption that drives almost of the others is my trillion-dollar revenue projection in a decade for that company, insofar that if that does not come to fruition, it will ripple through my other assumptions, with lower profit margins and higher failure risk following. That should make it the central question that we should be debating, and my extended discourse on why making cost of capital your central discussion point is misguided is designed as much to bring attention back to things that matter as it is to restoring balance to discussions of what a reasonable or high cost of capital for a company can be. 

    You are, of course, welcome to quote any number you want as the cost of equity and capital (and they are not interchangeable), but while you are entitled to your opinions, you are not entitled to make up facts. When you claim that the cost of equity is 12%, 15% or 20% for a typical firm, you are either valuing a company in Turkish lira or Zambian kwacha (where inflation is 15%+) or a company in Afghanistan or South Sudan (where country risk is daunting) or you are just making up stuff! Would I like to make 20% returns on the stocks I invest in? Of course, but I would also like to have a milk shake with every meal, without affecting my glucose and cholesterol scores, and get older without any of the symptoms of aging. 

YouTube Video

Data

1. Costs of equity and capital (in US $), by Industry (US, Global)
2. Historical returns on US stocks. bonds and bills: 1928 - 2025
3. Implied Expected Returns for US equities: 1960 -2025

Blog Posts on AI in 2026

1. To a Trillion Dollars and Beyond: A SpaceX Odyssey!
2. Revisiting SpaceX: A post-Prospectus valuation
3. OpenAI, Anthropic and SpaceX: The S&P 500 Inclusion Question and Investment Consequences
4. AI Scenarios: From Doomsday to Do-nothing!
5. AI's Bar Mitzvah: From hope & hype to business questions!
6. Stock Prices, Earnings and Cashflows: The AI Effect!
7. From Eyeballs to ARR: Buzzy Pricing Metrics and the Anthropic Pricing
8. The Anthropic Prospectus Arrives: My (but not the) valuation of Anthropic