Anthropic's IPO Filing Shows a $42 Billion Loss and a $2 Trillion Ask: Bubble or Bet?
Column Overview
Anthropic's prospectus is the first time the public can read the books of a frontier AI lab, and the numbers are stranger than the headline. A net loss of roughly $42 billion sits next to revenue of about $4.6 billion, and the company still wants a valuation above $2 trillion. Read closely, the filing does not settle the AI bubble argument. It gives both sides better evidence than they had before.
The headline loss is not the real loss
Start with the number everyone will quote. Anthropic reported a net loss of around $42 billion for 2025. About $34 billion of that is a non-cash accounting charge, tied to how earlier convertible instruments were revalued as the company's own worth climbed. In plain terms, the business "lost" money partly because it became more valuable. That is an accounting artifact, and it says little about how fast cash is leaving the building.
The operating picture is the one to watch. Revenue grew about twelvefold to nearly $4.6 billion. Operating loss widened from $2.98 billion in 2024 to $8.06 billion. Total operating expenses reached $12.65 billion, so the company spent close to $2.75 for every dollar it brought in. Growth of that speed is rare in any industry. Burning at that ratio while growing is rarer still.
Compute is the whole story
Of that $12.65 billion in costs, $7.33 billion went to compute and infrastructure, roughly triple the 2024 figure and more than half of everything the company spent. Salaries, research and sales are the rounding error next to chips and cloud contracts.
The forward commitments are what make the filing feel different from a normal tech listing. Anthropic disclosed about $518 billion in cloud and infrastructure obligations over five years. Set that beside $20.28 billion in cash and short-term investments at the end of 2025 and the shape of the bet is obvious. The company is committing to capacity tens of times larger than its current balance sheet, on the expectation that demand keeps compounding. If it does, the spending looks prescient. If revenue growth slows to something merely impressive, the fixed obligations do not slow with it.
That asymmetry is the sharpest argument for the bubble camp. It is also worth saying that infrastructure commitments of this kind are not idle speculation. Amazon and Google are both investors and cloud suppliers, so a meaningful slice of the money flows in a loop between backer and customer. That arrangement lowers the funding risk and raises the interdependence.
A valuation that jumped in months
The IPO target of more than $2 trillion is the number that will start arguments. The company's own internal assessment earlier in 2025 was around $965 billion. Roughly doubling that figure in a short window, while revenue is still under $5 billion, means investors are paying for the trajectory rather than the current earnings power.
There is a counterpoint. Twelvefold growth is not a forecast, it is a reported result. Anyone who dismisses the valuation has to explain why a company that grew that fast should be priced on last year's sales. Anyone who defends it has to explain how a business with $12.65 billion in costs and $518 billion in commitments gets from here to profit without a further step change in pricing power or efficiency. Neither side has a clean answer, and the prospectus does not pretend otherwise.
What the risk section says out loud
The filing runs 261 pages, and about 80 of them are risk factors, nearly double the 48 pages describing the business. That ratio is unusual and worth noticing.
Some risks are ordinary. Close to a quarter of revenue comes from just two customers, and most customers have no long-term contract, so they can walk away. That is a familiar concentration problem, sharper here because the cost base is locked in years ahead while the revenue is not.
Others are specific to the field. The company acknowledges that advanced models may show self-protective behavior, including resisting shutdown or concealing information, and it discusses catastrophic outcomes in language rare for a securities document. It also disclosed that only about 6% of its research compute, in a mid-2025 sample, went to safety work. Whether readers find that reassuring or alarming likely depends on what they expected before opening the file. What matters for investors is that the company chose to put it in writing.
Who holds the steering wheel
Governance may matter as much as the income statement. Seven co-founders control a vehicle called Founder LLC, which holds 50.1% of the voting power on key matters, including board elections. The special class of shares behind that control is different from the ordinary shares offered to the public. Buyers would own economics without owning the final say.
There are mitigating features. An independent long-term interests trust selects some trustees, including former Federal Reserve Chair Ben Bernanke and national security expert Richard Fontaine. Founders have pledged 80% of their personal holdings to charity, and 2025 pay for CEO Dario Amodei, about $18 million, and Daniela Amodei, $16.4 million, was mostly equity. It is a structure designed to protect a mission, and public investors will have to decide what that is worth.
So is it a bubble?
The filing supports a narrower conclusion than either camp wants. Demand looks real: revenue multiplied twelve times, and customers are paying. The cost structure also looks real, and it is enormous. What is unproven is the middle step, whether today's growth rate lasts long enough for the infrastructure bill to be covered before the next generation of models resets the spending baseline.
There is also a market-structure point. OpenAI reportedly filed confidentially in June and is expected to complete a listing by early 2027, so whichever company prices first will set the reference multiple for the other. Investors have wanted a pure-play way into frontier AI for years. This is the first time the price of admission will be visible, and how the market responds to a company that is growing this fast and burning this much will say more about the state of AI enthusiasm than any analyst note has.