Anthropic’s proposed TAM is unhinged, like, a lot

Sometimes, companies are like children.

They say the darnedest things to make their case.

But unlike with children, when you simply jot down whatever hilarious one-liner they blurt out and only get back to it 15 years later, finding the dusty notebook while cleaning the attic after you’ve kicked them they’ve moved out, things companies say can have… consequences.

Monetary consequences.

Your wallet consequences.

In late August the Wall Street Journal, citing the usual “people in the know”, reported that Anthropic is expected to tell IPO investors its TAMSidenote: Total Addressable Market.A popular investor pitch estimate basically saying what amount of revenue the company could theoretically generate if it captured 100% of its potential market.I'd stress the words "theoretically" and "potential".Some companies read that as "every single human alive". End of sidenote. exceeds 30 trillion dollars. In reaction, I, as a potential future investor, raised my eyebrows and whispered excitedly: “Singularity now!”

That lasted about two seconds.

Once I gave it some thought, it swiftly unraveled. So swiftly, in fact, that I actually started having second thoughts about whether I should really buy some stock once Anthropic goes public.

Because the number is untethered from any kind of non-drug-addled realitySidenote: Anthropic's own internal forecast for 2028 revenue is reportedly around 190 to 200 bn USD.That's circa 0.65% of the market it is (loosely) claiming.Latest self-reported run-rate revenue (July 2026) stands at $65 bn. Roughly 0.22% of its own claimed TAM. End of sidenote.. If that’s the, I assume, serious estimate they give potential multi-billion whales, what are they going to tell the rest of us plebs?

Or, as Claude would have put it: “It is wrongSidenote: WeWork's prospectus claimed a "total opportunity" market of $3 trillion in 2019.Long story short, it hilariously filed for bankruptcy in 2023.Of companies that fared a bit better (but far off their TAMs), Uber claimed a $6 trillion market, also in 2019.In May of this year SpaceX pitched the largest actionable TAM in history: $28.5 trillion, of which AI comprises $26.5 trillion.I guess Anthropic saw that and decided to slap some more sugar on top. End of sidenote., but the reason why is half the story, and it is in its re-telling that the answer lies.”

So, let’s get on with it.

Dream big, quantify bigger

There’s a lot we don’t know.

Since Anthropic’s draft prospectus to the SEC is confidential, all we’ve got about how the TAM was calculated is second-hand. Apparently, the company put a number to the full scope of work that could be completed with AI models.

Ugh… okay?

My head hurts when I try to figure out ways to quantify that.

What exactly is included?

Only mundane, already automation-prone tasks or also stuff where context, taste and expertiseSidenote: In other words, areas where AI is lagging significantly, because you can't solve everything using statistics, prediction and machine learning.Quote me on this: AI and LLMs will NOT reliably overcome these hurdles in the next 3 to 7 years. If ever. End of sidenote. are required?

Completed, as in completed good, or completed however?

Does the figure represent costs of services sold or productivity gains or savings?

Do the numbers account for needed and paid-for human oversight?

Or is this a future where we are just needless meatbags, subservient to the almighty machines?

Your guesses are as good as mine, Johnny.

When something is incredibly complex to forecast or estimate, we have a tendency to overshootSidenote: It has a pretty name, too: optimism bias. End of sidenote.. It’s terra incognita, land of the unknown, land of unlimited possibilities, and boy oh boy, that’s exciting.

I mean, why not be bold, who’s gonna argue with what is, basically, a thought experiment.

The investors. The investors definitely should argueSidenote: Unless they ALREADY plunged billions of dollars into AI, in which case... they kinda have to just nod and pray that at least 0.5% of that becomes true. End of sidenote. with that.

Because this is Anthropic kinda pitching them its future business, or, more specifically, the potential returns on their investments. So if I were on the other side of the table, I’d want to hear numbers that are plausible, grounded and with serious thought behind them.

I don’t think Anthropic’s TAM meets any of these criteria.

Let me offer a counter-pitch:

“Would you commit billions of dollars to get my zero-calorie, non-GMO, extra-whatever lemonade e-shop up and running? Y’know, my TAM is 41.5 billion USD, because, theoretically, I can serve everyone on Earth for just five bucks a year. And everyone needs to drink and everyone wants to be healthy. Including birds, maybe I can even push it up to 50 bil!”

Of course, I stretch for effect, but this strikes at the heart of overly optimistic TAM estimates. It’s not that hard to smuggle something defensible into them. Doesn’t mean you should, though.

A world where everyone, or even the vast majority of consumers, use your product, is a world which every company wants to live in. TAM is a great platform for that.

But, funny thing is, it’s absolutely no wonder that Anthropic came up with such a wild number.

It kinda needed to because of the sheer scale of investments alreadySidenote: Per Goldman Sachs: $1 trillion of AI-related investment in 2026 alone... and running. End of sidenote. made and committed to AI in general.

And, who knows, maybe the math adds up.

The math doesn’t add up

Now-now, I know companies jawbone their book constantly, because you’re simply not gonna attract big money with small promises. Carve it on a startup’s wall somewhere.

However, as I’ve said, when talking to potential investors, even your wishful thinking should have some basis in market and business realities. Because in the end it is real money we’re talking about, and nothing makes numbers go red faster than unfulfilled expectations.

And while no one will fault you for being optimistic (to an extent), even optimism hits limits. Especially the mathematical ones.

Disclaimer: it’s almost impossible to plausibly reverse-engineer Anthropic’s TAM number based on the “work that can be completed with AI models” promise. Thus, to illustrate my point, I’ve taken some liberties hereSidenote: As, apparently, Anthropic did when coming up with the TAM number.The circle became whole. End of sidenote..

So, based on the 30-tril figure, what could we assume about AI’s potential market size?

Calculators out.

There are, give or take, 8.3 billion people on Earth. If, theoretically, every single one of them generated the same amount of revenue just for Anthropic, each one would have to fork out approximately 3,600 USD. Every year.

If every single one of them used Claude Pro (priced, as of the time of writing, at 20 USD/month), that’s a 15-year subscription. Bought annually.

But, oops, not everyone has a computer. And an internet connection. Or can read.

So let’s shrink that number to 6.1 billionSidenote: Roughly 6.12 billion active internet users as of April 2026 or 73.8% of the world. Per DataReportal and Kepios. End of sidenote.. Now everyone needs to spend 4,900 USD. Which would buy 20 years of Pro usage.

What’s that? Some people deliberately limit their use of or completely avoid AI? Others binge on FREE models only?

Let’s shave another billion off that. Bill’s gone up to 5,880 USD. About 24.5 years of Pro. Per year.

I know, this isn’t what Anthropic had in mind when making up calculating their estimates. Institutional users cough up far more dough than retail ones. And processes/workSidenote: Fun fact: global wages are approximately $66 trillion annually (derived from data by ILO and World Bank).If it is possible to somehow automate or augment $30 trillion of that, that would make close to half of every buck of global labor income an eventual victim of AI.Not sure about you, but I am excited about all the houses large language models are going to build and crops they're gonna plant. End of sidenote. are not the same as users.

But I think it illustrates the massive scale of adoption needed to even come close to fulfilling the TAM.

The China Effect

They say that the US and China are in an AI war.

I’ll leave it to more competent people to decide who’s winning on the technological frontier, but when it comes to naming their modelsSidenote: China: Qwen, Kimi, GLM, DeepSeekUS: ChatGPT, Claude, Gemini, Grok End of sidenote.… I’d say they are pretty much tied in the “lack-of-imagination” tier.

But if China’s got one thing working in its favor, it is the fact that it is able to develop and train its models far more cheaply than its American counterparts can.

What does that mean?

Primarily that it is also able to sell them far more cheaply, or even offer them for free.

Now, it is important to note here that based on various benchmarks the Chinese models are still somewhat behind in terms of performance and quality of outputs. But, let me tell you, there are at least two reasons that’s not a problem:

• the gap is rapidly narrowing,
• the vast majority of users simply won’t care.

We can also re-word the second point to: cheap is good.

This, my brothers and sisters, is a story that’s at least… 25 years old.

You see, back in 2001 China became a member of the World Trade Organization and happily primed its manufacturing capacities to flood the global markets with low-cost goods. And, boyyy, they did.

How did everyday consumers react?

They embraced it.

Clothes, household items, later on tech gadgets and even cars; when faced with a choice between a more expensive domestic/established alternative and a cheaper Chinese one, a significant portion of buyers chose the latter.

It’s all part of the so-called China Shock which is, in my opinion, THE most significant economic event of this century.

But opinions aside, you can probably see where I’m heading with this.

There’s currently no reason to think that LLMs and AI will be any different. In fact, a migration to Chinese open-weight models has already begunSidenote: By May 2026 Chinese models were around 60% of all tokens consumed on OpenRouter, an LLM marketplace allowing you to use hundreds of models from various providers.Also in 2026, Zhipu's GLM 5.2 saw the fastest adoption of any model tracked by the American deployment platform Vercel, with customers up roughly 80x in its first full week. End of sidenote..

Led by companies, the biggest spending segment of Anthropic’s customers.

I’ll be the first to admit that this doesn’t specifically address the TAM part, rather it points out the hurdles on the way towards it.

And this is a big (and simultaneously cheap!) one.

Better than Windows?

The best way to tell if something is big or small, crazy or sane, is to simply compare it to something. Like a peer. Because as Albert Einstein definitely did not say: “It’s all relative.”

So let’s do just that.

Do you have a computer at home?

If so, chances are it’s running on Windows. You know, only the most widespread operating system for over 30 years and counting. It is assumed that nowadays roughly 3 out of 5 personal computersSidenote: I wanted to do a side-by-side here, but it's hard to get any vetted customer numbers for Anthropic.By its own past disclosures it counted 300,000 business customers.Stale number (October 2025), so treat it so. End of sidenote. worldwide have Windows installed.

There’s hardly a more pronounced technological (and even cultural) phenomenon, bar a few, so permeated into the everyday lives of even “non-tech” people.

Wherever you look, folks use computers running on Windows. From that Nigerian fella that’s constantly trying to send you his 2 mil USD, to the bored 72-year-old grandma from Kissimmee, Florida, who just found a mysterious and generous lover on the internet.

What does Microsoft say its all-encompassing TAM is?

It… doesn’t say.

More than 50 years up and running, hundreds of billionsSidenote: $331 bn for 2026. End of sidenote. in real revenue, an unimaginable amount of market data, and it still won’t put a total number on its own playing field.

And remember, Microsoft, that’s not only Windows, but Copilot, Xbox, Azure, MS Office, LinkedIn, etc.

Now let’s sit with those facts for a while and then meditate on the answer to the following question:

How plausible is it for a 5-year-old company to dwarf a giantSidenote: For the purposes of this article, comparing Microsoft and Anthropic is a bit stretchy, I admit.Even if they both are primarily software companies.The reason being that Anthropic's $30 trillion wet dream is basically measured against labor, not software itself.If Microsoft did the same, it could say that its TAM is the sum of all cost savings made by using a computer and not an abacus. End of sidenote. (even if only in projections) that’s been dominating the personal computer and adjacent markets for more than three decades, that ALSO kinda competes in the same space?

Since we’re meditating, you should answer that in silence.

What to take from it?

Friends, listen.

I hope AI and LLMs are the breakthrough they are made out to be. Maybe they won’t be (at least not to the full extent of the most optimistic fantasies), but that’s alright, since they already are useful enough, and undeniably will be more so.

And I say that as a heavy Claude userSidenote: It proof-read my first drafts of this article and took everything in stride.Hats off.And don't forget that I always said hello and thank you once you take over this planet! End of sidenote..

But the technology being good doesn’t make those numbers sane.

Why is that important?

Because institutional whales will look at the TAM number, file it under “vision and bravado, treat with skepticism” and price the company off next year’s projected revenue and capital already put in.

But retail investors and traders…

Well, we sometimes read a number like that and open a brokerage app.

Buying into a company based on vibes is a far too common occurrence. Just go read some retail forums and look for reasons people purchased stocks. It’s crazy.

Anthropic could be a magnificent company and the AI buildout could pay off big, and you could still lose rent money buying the IPO at the wrong size or the wrong moment.

The quoted TAM is not a rationally deduced number. It’s EXACTLY what a vibe is.

And vibes make people oversize.

So will I buy some stock after all?

You know, I might, because $30 trillion is a sexy number.