A draft IPO prospectus viewed by The Wall Street Journal shows Anthropic crossing into a different category of AI company: one with revenue exploding fast enough to excite equity markets, and infrastructure obligations large enough to make a public offering look as much like a funding vehicle for compute as a conventional growth-company listing. The draft reportedly shows $16.3 billion of revenue in the first half of 2026, up from $1.1 billion a year earlier, alongside an $80.7 billion net loss and at least $518 billion of commitments over a decade for cloud services, computing and infrastructure.
That combination is why this matters beyond one splashy AI filing. The real question is not whether Anthropic is growing. It plainly is. The question is what, exactly, public investors would be asked to underwrite: a software business with unusual capital intensity, or one of the largest private compute buildouts yet attempted. The answer will matter to cloud providers, chip and infrastructure suppliers, enterprise customers, lenders, employees and anyone trying to price the economics of frontier AI.
The reported loss is not the same as the cash burn
The $80.7 billion loss is the figure that will draw the biggest headline, but it is also the easiest number to misread. As Reuters reported, roughly $72 billion of that first-half loss came from an accounting charge tied to the rising fair value of convertible bonds. In other words, the draft does not describe an $80.7 billion operating cash hole in six months.
That distinction is not a technical footnote. Convertible-bond accounting can make a company’s reported net loss balloon as its valuation rises, which is almost the opposite of the distress signal that a giant loss would usually send. Anthropic showed a similar pattern before: Reuters previously reported a 2025 net loss of about $42 billion, mostly an accounting charge, on roughly $4.6 billion of revenue and an operating loss of about $8.1 billion.
But separating accounting effects from operating economics does not make the underlying business cheap. Reuters also reported that Anthropic spent nearly $9 billion on computing capacity and infrastructure in the first half of 2026, up from $2.3 billion a year earlier. That is real spending for real capacity. Unlike a fair-value mark, it reflects the hard economics of training and serving frontier models at scale.
Revenue is real, but the cost curve is chasing it
If the loss needs decoding, the growth figure does not. Going from $1.1 billion of first-half revenue in 2025 to $16.3 billion in the same period of 2026 is extraordinary by any software standard. Reuters separately reported that an October 14 investor meeting put third-quarter revenue at $16.8 billion, up 45% from the second quarter. Those numbers suggest real enterprise and developer demand, not just speculative enthusiasm around AI.
They also help explain why some investors are thinking about a valuation closer to $1.5 trillion than earlier expectations, even though no final valuation or timing is established. A business reaching that kind of revenue run rate this quickly would normally be judged as a rare breakout winner.
The complication is that frontier AI is not normal software. Sequential growth reportedly slowed in the third quarter from the more-than-doubling pace seen between the first and second quarters. Forty-five percent growth in a quarter is still blistering, but the change in trajectory matters because Anthropic’s costs are not only large; they are tied to capacity planning decisions that have to be made before future customer demand is fully known. Public markets may be less interested in whether Anthropic can grow than in whether revenue can keep outrunning the cost of securing and using compute.
That is where enterprise customers enter the story. If capital remains abundant and demand keeps compounding, long-term capacity bets can look prescient, even strategic. If capital gets more expensive or utilization disappoints, an AI vendor has fewer easy levers. It may have to push harder on pricing, steer customers toward higher-margin products, sign longer contracts or slow investments that do not pay back quickly.
The bigger IPO test is infrastructure finance
The most consequential figure in the draft may not be the loss or the revenue, but the commitments. The draft reportedly says Anthropic has committed to at least $518 billion over a decade for cloud services, computing and infrastructure. The Journal and Axios have also described a larger number, around $750 billion, when deals still under negotiation are included. That bigger figure should not be treated as a firm liability. Even so, the reported minimum alone is enough to reframe the prospective IPO.
These commitments are not the same as immediate spending, and public reporting still does not show the exact contract language, cancellation rights, delivery schedules, minimum-take terms or pricing. Those details will determine how rigid or flexible the obligations really are. But at this scale, the broad economic effect is already visible. Cloud providers and chip suppliers get a powerful long-term demand signal they can use to expand data centers, power arrangements and hardware supply. Anthropic, in turn, takes on utilization and financing risk in exchange for access to scarce capacity.
That is why the prospective listing looks less like a plain liquidity event and more like a stress test of the AI infrastructure financing model. If public investors are receptive, Anthropic gains access to a deeper capital pool and a market benchmark that could support years of compute spending. If investors are less forgiving than private backers have been, the company could face sharper scrutiny around margins, contract structure and how efficiently it is turning reserved capacity into revenue.
The unanswered questions are the ones that usually sit deeper in a prospectus: how much cash is on hand, how concentrated the customer base is, how product margins differ, how much of the $518 billion is firmly signed, and whether revenue can ultimately scale faster than the infrastructure it depends on. Until a final filing appears, those are the pages investors cannot read.
What the draft already makes clear is that Anthropic is no longer just an AI growth story. It is a public-markets test of whether extraordinary software demand can justify extraordinary pre-purchased compute capacity—and of who ends up holding the risk if it cannot.
By
By
By
By
By
By
