Anthropic's IPO: what changes for companies using AI

Anthropic investors expect the company to go public in October 2026 at a valuation of US$ 2 trillion or more. If it happens, it will be the largest initial public offering in history — surpassing SpaceX's June debut at US$ 1.77 trillion, and more than doubling Anthropic's own last private valuation of US$ 965 billion.

Before going further, one caveat almost every headline left out: that number is what investors expect, not a company target. According to the Financial Times, senior Anthropic executives have not set a valuation goal even in private conversations.

From the outside it reads like financial-market news. But there is a concrete consequence for any company that has put AI inside a process: the price you pay today is the price of a company that does not yet need to turn a profit. An IPO changes that yardstick.

The numbers behind the subsidy

Look at OpenAI, the direct competitor, in the first quarter of 2026: US$ 5.7 billion in revenue and US$ 3.7 billion burned in the same period. For the full year, the company projects revenue around US$ 30 billion against a cash burn of US$ 25 to 27 billion.

Zoom out and it gets clearer: OpenAI tells investors it is targeting roughly US$ 600 billion in compute through 2030 — and that figure is already revised downward, after Sam Altman spoke of US$ 1.4 trillion in infrastructure commitments. Projected cumulative cash burn through 2030 reaches US$ 665 billion.

None of that is mature-market pricing. It is the price of buying a market, funded by investment rounds. And investment rounds tolerate years of losses; public shareholders tend to be less patient.

Anthropic's side

The trajectory is remarkable and worth recording: annualised revenue passed US$ 47 billion in May, and investors project somewhere between US$ 100 and 120 billion by year-end — more than ten times where it stood at the start of 2026. The company filed a draft registration with the SEC in June, moving ahead of OpenAI in the race to list, days after closing a US$ 65 billion Series H.

But the same reporting points to three pressures the enthusiasm hides:

• Price. Anthropic's top model costs more than 2.5 times OpenAI's flagship.
• Open competition. Chinese open-weight alternatives deliver comparable capability at a fraction of the cost — and already lead benchmarks in specific areas, such as cybersecurity.
• Regulation. Revenue growth slowed in June after the U.S. Department of Commerce imposed temporary export controls on the company's best models.

OpenAI, meanwhile, appears to be moving the other way: New York Times reporting indicates the listing is being pushed to 2027, with Altman treating any valuation below US$ 1 trillion as unacceptable.

What changes inside your company

A publicly traded company answers to quarterly results. In practice, for anyone consuming AI, that usually shows up in four places:

1. Usage pricing. Cost per token and enterprise subscriptions stop being a land-grab instrument and become a margin line.
2. The free tier. Free plans exist to build habit. After an IPO they are cost without revenue — and they tend to shrink.
3. Older models. Keeping old versions running costs money. The pressure is to retire them faster, and whoever built on a specific model redoes the work.
4. Usage limits. Quotas and rate limits get stricter, because every call carries a real compute cost.

None of this is a catastrophe. It is the normal cycle of any technology that went through a subsidised phase — it happened with cloud computing, with ride-hailing apps, with food delivery. The mistake is planning as if today's price were permanent.

How to prepare without giving up AI

• Know what you spend and model double. If doubling the cost breaks the process, that process is fragile — and you want to find out now, not in the repricing email.
• Do not tie a critical process to a single vendor. The right question is: if this vendor triples its price tomorrow, how long would it take me to switch?
• Favour an architecture that allows swapping. Isolating the AI call behind your own layer costs little while building and pays off later. Changing the engine should be a configuration change, not a rewrite.
• Treat open models as a real plan B. They are no longer the poor alternative: on specific tasks the open ones already compete with closed models at far lower cost. Worth testing before you need it.
• Document what the AI does in your process. If the logic only exists inside a prompt someone wrote and nobody recorded, you do not have a process — you have a dependency.

Whoever has built AI applied to the business with portability in mind will barely feel it. Whoever tied everything to one vendor because it was cheap will discover the real cost all at once.

Update: the closed second-quarter numbers

Two days after we published this piece, the Wall Street Journal reported the closed second-quarter results — and they change part of the picture described above.

• OpenAI: revenue of US$ 6.7 billion, up 18% from US$ 5.7 billion in the first quarter. The operating loss, including stock-based compensation, rose from US$ 9.3 billion to US$ 12.3 billion — a 32% jump.
• Anthropic: revenue of US$ 11.6 billion, more than double the US$ 4.73 billion of the first quarter, up roughly 143%. And, for the first time, a frontier lab posted a quarterly operating profit on an adjusted basis.

Anthropic overtook OpenAI in quarterly revenue for the first time, and the two are no longer in the same financial position. The line saying "both burn billions" was true when this text was written and no longer is — the correction is on the record.

This adjusts the picture without knocking down the argument, but it demands an important nuance. OpenAI's 18% growth came in below what some investors expected, and the reasons analysts point to are precisely the cost of subsidising free ChatGPT users and recent price cuts on two models, at a moment when corporate clients are shifting part of their workloads to cheaper Chinese models.

Notice what that means for buyers: in the short term competition is pushing prices down, not up. The change of yardstick described above still holds — it just arrives with the listing and the quarterly scoreboard, not now. Whoever prepares while prices are falling is the one who will not feel it when they stop falling.

November update: what the prospectus shows

On November 2, 2026, Anthropic filed with the SEC, the US securities regulator, the amended prospectus for its IPO, asking to list on Nasdaq under the ticker ANT. For the first time, the numbers come from the official document rather than from investors or press reports.

• Revenue: about $17 billion in the third quarter, up from $11.6 billion in the second. Quarter-over-quarter growth slowed from about 144% to 47%.
• Result: an operating loss of $567 million, after $974 million in operating profit in the second quarter. According to the filing, the main reason was compute bought on short-term contracts, which costs significantly more than capacity secured under multi-year deals. On the company's own adjusted basis, operations remain profitable, with a 15% margin.
• Commitments: about $895 billion in purchase commitments, almost all of it compute and infrastructure.
• Price: the prospectus says price reductions, from new model versions and contractual discounts, took about $600 million off the quarter's revenue.

That last figure matters most to anyone paying for AI. It confirms what the previous update suggested: for now, competition is pushing prices down. At the same time, the loss shows the size of the compute bill behind those prices, and nearly $900 billion in commitments has to be paid for with revenue. The shift described above has not gone away; it has become easier to see.

The prospectus also carried a warning that is not financial. Among its risk factors, the company wrote that its estimates for future models, not current ones, indicate it may “imminently” meet its safety policy's definition of dramatically accelerated progress attributable to AI. Anthropic says no current model has crossed that threshold and that it is considering policy changes to keep humans in control of development. We explain what an AI that speeds up its own progress means in AI that improves itself.

For your business, the practical lesson stays the same: use the falling prices to organize how you use AI, measure the cost and keep a change of provider possible. Fast cycles of new models and price cuts favor those who can switch without rebuilding their processes.

The honest summary

A US$ 2 trillion IPO would be the largest debut in history — and it is still an investor expectation, neither a done deal nor a stated target. It may come in smaller, it may come later, it may not happen in October.

What is already fact: the two largest AI companies in the world are heading to the stock market, and holding current prices is expensive — OpenAI burned 32% more in the second quarter than in the first. While the race for market share lasts, price works as an instrument; once the yardstick becomes quarterly results, it becomes a margin line. That turn is the one worth getting ahead of.

Facts drawn from Financial Times and New York Times reporting, market filings and the technology press of August 2026. Anthropic's valuation and revenue projections come from investors, not from the company. Updated on 19 August 2026 with the closed second-quarter results.