Anthropic S-1 IPO filing: compute costs fall to 55% of revenue as operating loss narrows

Anthropic filed its S-1 on October 26, 2026, and the headline number is doing the loudest talking. The company reported a net loss of $80.7 billion for the six months ended June 30, 2026. That figure is real, audited, and in the filing. But it is not the number that matters most for investors watching the ANT listing, because nearly all of it is an accounting artifact of how the company priced its convertible notes, not the cash reality of the underlying business.

The number worth leading with is the compute bill. Anthropic spent $8.9 billion on compute and infrastructure in the first half of 2026, which works out to about 55 cents of every revenue dollar. A year earlier that ratio was roughly 201 cents. The company did not spend less on compute in dollar terms, it grew revenue 14 times faster than it grew compute and infrastructure costs, and the operating result turned decisively toward breakeven.

Related reading: Anthropic Revenue Hit $16.8 Billion in Q3, First Profit Sets Up IPO.

The operating story under the accounting loss

Strip out the fair-value swings and Anthropic's income statement reads like a startup that crossed an inflection. Revenue came in at $16.3 billion for the six months ended June 30, 2026, against $1.2 billion in the prior-year period, a roughly 14 times jump. Operating expenses grew, but not at that pace, so the operating loss shrank from $2.1 billion to $693 million. On an adjusted basis the company reports it reached positive operating income of $1.6 billion in the first half, and 19% adjusted operating margin in the second quarter on its own definitions, up from a negative 129% in the same quarter a year prior.

The contrast is most visible inside the loss line. The net loss of $80.7 billion includes a change in fair value of convertible notes, derivative liabilities and SAFEs of $72.4 billion, up from $10.9 billion in the first half of 2025. As the underlying valuation of the company has climbed ahead of an IPO, the mark on those instruments goes against the income statement. The cash the company consumed is far smaller than the accounting loss implies.

Related reading: Meta Files $95 Billion AI Raise, Anthropic IPO Shows $80.7 Billion Loss.

Source: Anthropic PBC S-1, Registration Statement on Form S-1, filed with the SEC on October 26, 2026, Consolidated Statements of Operations and Management's Discussion and Analysis.

What the compute ratio tells the market

For the AI trade, the single most consequential number in this filing is not the loss. It is that Anthropic is generating more revenue per unit of compute than it did a year ago, by a wide margin. Revenue rose roughly 14 times while compute and infrastructure spend rose about 3.8 times. That is the efficiency claim the hyperscaler capex debate has been testing, and the company is now asserting it in a place it must defend under SEC liability.

Related reading: Lam Research Posts Record Quarter but Supply Bottlenecks Cap the AI Chip Equipment Ceiling.

The company says compute efficiency shows in revenue that grew about 1,300% (roughly 14 times) year over year, and in a net revenue retention of about 900% as of September 30, 2026. It also points to model pricing: Claude Opus 5.5 cut customer cost per task by about 40% against Opus 5, which the company uses to argue that lower cost per task drives usage that more than offsets the pricing pressure. Whether that holds at agentic scale is the open question. Agentic workloads are structurally more compute intensive than non-agentic ones, and the company itself flags in the risk factors that if it cannot price agentic use to cover its compute cost, its efficiency could deteriorate.

The commitments reach far past this quarter

The efficiency trend runs headlong into the scale of locked-in compute. The S-1 details non-cancelable compute commitments that dwarf the current spend. With Google, Anthropic committed to spend at least about $114.5 billion between April 2026 and July 2033 regardless of usage, with a true-up if it falls short, plus an estimated additional $125 billion to $150 billion for TPU compute through the early 2030s. With Amazon, it committed to at least $110 billion between May 2026 and April 2033. Separate deals add roughly $41.3 billion with Lambda, non-cancelable unless Lambda misses specified delivery terms, and a March 2026 equipment lease with Broadcom for specialized AI accelerators that Broadcom intends to assign to a financial partner.

Those are the numbers that determine whether the efficiency ratio is durable. Anthropic is paying these take-or-pay bills regardless of demand, so its operating leverage over the next several years is, in part, a function of whether the revenue line can keep growing faster than a fixed compute burden it has already signed away. The bull case says the commitments secure capacity the company needs and convert to revenue efficiently. The bear case is the risk factor the company itself printed: if customer demand underwhelms, the fixed obligations still come due, and the company is on the hook for the difference.

The dollar figures also explain why the company keeps raising. As of June 30, 2026, it reports having raised an aggregate of $141.1 billion of capital, holding $112.4 billion in cash plus $19.3 billion in short-term investments. Financing activity in the first half alone produced $110.8 billion of cash inflow, against $4.7 billion a year prior, mostly redeemable convertible preferred stock. Total indebtedness stood at $6.2 billion as of June 30, 2026, under the Amazon convertible notes.

Two other disclosures shape how the market will price ANT. First, concentration: no single customer represented more than 10% of revenue for the nine months ended September 30, 2026, but the top five customers accounted for about 25%, and several of those are not tied to long-term contracts and can cut spend on short notice. Second, governance: Anthropic is a public benefit corporation with a Long-Term Benefit Trust, and a share structure that lets founders exercise control through a Founder LLC. That structure balances return-seeking with AI safety obligations, and it is novel enough that the company devotes a full risk factor to it.

What to watch in the next print

The S-1 is a preliminary prospectus, so the offering price range is still blank. The two signatures that set the next leg are the price range Anthropic and its underwriters print, and the revenue efficiency figures in the first quarterly report after listing, which is when the market sees whether the 55% compute ratio holds against the fixed commitments. The other watch item is the fair value line: as the offering prices and the convertible instruments mark closer to the IPO valuation, that non-cash charge should begin to fade, which is partly why the operating-loss trajectory, not the headline net loss, is the metric to follow.

The listing is led by Goldman Sachs and Morgan Stanley, with J.P. Morgan, Citigroup, Barclays, Wells Fargo Securities, BofA Securities, Deutsche Bank Securities and RBC Capital Markets on the syndicate, and the stock is set to trade on the Nasdaq Global Select Market under ANT. Anthropic has stated in the filing that the public markets are the realization of years of capital raising and compute building; the market's job now is to work out whether the efficiency curve it printed can outrun the commitments it signed.

This content is for informational purposes only and does not constitute financial advice. Past performance is not indicative of future results. Consult a qualified financial advisor before making investment decisions.