Anthropic loses $80.7 billion in half a year, compute commitments up to $750 billion, Polymarket gives 17% bankruptcy probability

Prediction markets assign Anthropic a 17% probability of bankruptcy, reflecting the market's complex judgment of its viability under the high-capex model of AI model companies.

According to the latest data from prediction market platform Polymarket, the probability that Anthropic will declare bankruptcy before the end of 2028 is only 17%. This pricing stands in sharp tension with the draft IPO prospectus that surfaced this week—the draft shows that Anthropic's net loss in the first half of 2026 exceeded $80.7 billion, and its committed spending on computing infrastructure could reach as high as $750 billion.

A previous Wall Street CN article noted that, according to The Wall Street Journal citing Anthropic's latest version of its S-1 draft prospectus, about $72 billion of the $80.7 billion net loss consists of non-cash accounting adjustments, with the actual operating loss at about $8.7 billion—but that figure already exceeds the more than $8 billion in total operating losses for all of 2025. At the same time, the company's revenue in the first half of 2026 reached $16.3 billion, and third-quarter revenue grew 45% quarter over quarter to $16.8 billion, showing some commercialization momentum.

However, the slowdown in revenue growth has already triggered valuation disagreements among potential IPO buyers. Management disclosed at investor day that quarter-over-quarter revenue growth slowed sharply from about 140% between the first and second quarters to 45% to 46% in the third quarter, prompting some institutional investors to cut their reference valuation range from $2 trillion to about $1.5 trillion. Polymarket's 17% bankruptcy probability pricing is essentially the market betting that Anthropic's revenue growth trajectory will be enough to achieve profitability at scale before its funds run out—but the preconditions for that bet are becoming increasingly demanding.

Hidden concerns in revenue growth: concentration risk and slowing growth

The core logic behind the market assigning Anthropic a relatively low bankruptcy probability lies in its rapid revenue growth. But that growth itself has structural vulnerabilities.

The prospectus shows that of the $16.3 billion in revenue in the first half of 2026, about 20%, or roughly $3.26 billion, came from just two customers. The company also warned investors that "for the foreseeable future," revenue will remain highly concentrated among a "limited number of customers," and that many large enterprise customers have not signed long-term contracts and could in theory switch suppliers at any time.

At the same time, the slowdown in revenue growth has triggered repricing by institutional investors. Management disclosed at investor day that quarter-over-quarter growth slowed sharply from about 140% between the first and second quarters to 45% to 46% in the third quarter. Although 45% quarterly growth is still strong in absolute terms, the gap from earlier growth rates has been enough to prompt some institutions to cut their reference valuation range from $2 trillion to about $1.5 trillion.

Anthropic said it achieved adjusted positive profit in the second quarter of 2026, but before fully audited financial data is released, the specific definition of "adjusted" profit remains to be clarified.

The pricing logic behind 17%: what the market is betting on

Polymarket's 17% bankruptcy probability pricing essentially reflects the market's probability-weighted judgment on the following core question: whether Anthropic's revenue growth trajectory can cover its fixed computing liabilities before the end of 2028 while maintaining sufficient financing capacity to bridge its operating gap.

Arguments supporting the low bankruptcy probability include: revenue growth remains high; strategic investors such as Google and Amazon have strong incentives to keep the company operating; the IPO itself will provide a new channel for capital replenishment; and the vast majority of the $80.7 billion net loss consists of non-cash items that do not create immediate cash pressure.

But the risk implied by this pricing is that it may underestimate the liquidity pressure from fixed computing liabilities in a scenario of continued slowing revenue growth, as well as the revenue fragility caused by high customer concentration. The sustainability of the high-capex model of AI model companies ultimately depends on an assumption that has not yet been fully validated by historical data: that computing investment can be converted quickly enough into defensible, long-term locked-in revenue.

Anthropic's IPO will be the first real stress test of that assumption in public markets.

Astronomical losses: the two-layer structure of $80.7 billion

A Wall Street CN article noted that Anthropic's $80.7 billion half-year net loss consists, at the accounting level, of two parts of very different nature.

About $72 billion of it consists of non-cash accounting adjustments. As the company's private valuation has continued to climb across successive financing rounds, U.S. generally accepted accounting principles (GAAP) require the fair value liability of convertible notes and other complex financing instruments issued to early investors to be marked up accordingly, thereby creating large-scale "paper losses" on the books. This accounting treatment is not uncommon among fast-financing tech startups, but the absolute scale of $80.7 billion is still rare.

Stripping out this accounting adjustment, the company's actual operating loss in the first half still reached about $8.7 billion, already exceeding the more than $8 billion in total operating losses for all of 2025. The core driver is the explosive growth in computing spending: in the first half of 2026, Anthropic's direct spending on computing resources and infrastructure approached $9 billion, nearly four times the $2.3 billion in the first half of 2025 and even exceeding total computing spending of $7.3 billion for all of 2025.

This means that before paying any compensation to researchers, management, or sales staff, more than 55 cents of every dollar of revenue already flows to data center operators and chip suppliers. This cost structure is the key starting point for understanding Polymarket's pricing logic.

A $750 billion computing bill: from flexible spending to irrevocable liabilities

The real risk of Anthropic's computing investment is not the scale itself, but the fact that its contract structure has evolved from a flexible pay-as-you-go model into fixed, irrevocable long-term liabilities.

A Wall Street CN article mentioned that, according to The Wall Street Journal, the company's locked-in computing infrastructure spending commitments have exceeded $600 billion, and if an additional $150 billion agreement under negotiation is finalized, the total will reach $750 billion. According to the latest draft prospectus, the company must bear about $450 billion in cloud computing contractual obligations over the next seven to ten years, and the vast majority of the agreements explicitly state that payment is required "regardless of actual usage."

Key contract terms include: a $111 billion contract with Google Cloud; an $110 billion contract with Amazon Web Services (AWS); a $31 billion contract with Microsoft; chip leasing agreements with AI cloud service providers Nscale and Lambda totaling up to $86 billion, with lease terms extending to 2033-2034. In addition, through a separate arrangement with Broadcom, the company has assumed more than $160 billion in Google TPU chip leasing obligations.

The prospectus also discloses a supply-side dilemma: Anthropic has sought more computing capacity than AWS, Google, and Microsoft are willing to provide, partly because the three hyperscalers increasingly view it as a direct competitor. Forced to turn to alternative suppliers, the company began building its own data centers and partnering with third-party facility operators, and the prospectus acknowledges that these operators have a "limited operating history" and that their reliability is uncertain. The prospectus also mentions that some computing arrangements are financed off-balance-sheet through special purpose vehicles (SPVs) and private credit instruments, further adding to uncertainty about financial transparency. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment based on this article is at the user's own risk.