Loss of $80.7 billion in half a year, a $750 billion computing power gamble—why is Anthropic's bankruptcy probability priced at only 17%?

Prediction markets assign a 17% probability of bankruptcy to Anthropic, reflecting the market’s complex judgment of the survival ability of AI model companies under high capital expenditure models.

According to the latest data from prediciton market platform Polymarket, Anthropic has only a 17% chance of declaring bankruptcy before the end of 2028. This pricing stands in sharp contrast to the company’s IPO prospectus, which was revealed this week—the prospectus shows Anthropic’s net loss in the first half of 2026 exceeded $80.7 billion, and its commitments to computing infrastructure spending could reach up to $750 billion.

As a Wallstreet Insights article previously noted, according to the Wall Street Journal referencing Anthropic's latest S-1 filing, about $72 billion out of the $80.7 billion net loss was due to non-cash accounting adjustments, with actual operating losses around $8.7 billion. However, this number already exceeds the total operating loss of more than $8 billion for all of 2025. Meanwhile, the company’s revenue in the first half of 2026 reached $16.3 billion, with a 45% quarter-on-quarter increase to $16.8 billion in the third quarter, suggesting some commercial momentum.

However, the slowdown in revenue growth has led to valuation disputes among potential IPO buyers. Management disclosed at Investor Day that quarter-on-quarter growth slowed dramatically from about 140% between Q1 and Q2 to 45%-46% in Q3, prompting some institutional investors to lower their reference valuation range from $2 trillion to about $1.5 trillion. Polymarket’s 17% bankruptcy probability pricing is essentially a market bet: that Anthropic’s revenue growth trajectory is sufficient to achieve large-scale profitability before funds run out—but the prerequisites for this bet are becoming increasingly stringent.

Revenue Growth Concerns: Concentration Risks and Decelerating Growth

The core logic behind the low bankruptcy probability that the market gives Anthropic is its rapid revenue growth. However, this growth itself carries structural risks.

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

At the same time, the slowdown in revenue growth has already triggered a repricing by institutional investors. Management disclosed at Investor Day that quarter-on-quarter growth slowed from about 140% between the first and second quarters to 45%-46% in the third quarter. Although a 45% quarterly growth rate is still strong in absolute terms, the sharp deceleration has been enough for some institutions to lower the valuation range from $2 trillion to about $1.5 trillion.

Anthropic stated that it achieved adjusted positive profits in Q2 2026, but the specific basis for "adjusted" profits still needs clarification until fully audited financial data are released.

The Logic of the 17% Pricing: What Is the Market Betting On?

Polymarket's 17% bankruptcy probability pricing essentially reflects the probability-weighted judgment of the following core issue: Can Anthropic’s revenue growth trajectory cover its fixed compute debt by the end of 2028, while maintaining sufficient financing capacity to make up for operational shortfalls.

Arguments supporting a low probability of bankruptcy include: sustained high revenue growth, strong motivation from strategic investors such as Google and Amazon to keep the company operational, the IPO itself providing a new capital channel, and the fact that most of the $80.7 billion net loss is non-cash and does not represent immediate cash pressure.

However, the risk implied in this pricing is that it may underestimate the liquidity stress of fixed compute liabilities if revenue growth slows further, as well as the vulnerability of revenue resulting from high client concentration. The sustainability of the high capital expenditure model for AI model companies ultimately depends on an assumption that historical data have not fully verified: that compute investments can be converted into defensible, long-term locked-in revenue quickly enough.

Anthropic’s IPO will be the first real stress test of this assumption in the public market.

Astronomical Losses: The Two-Tier Structure of the $80.7 Billion

As Wallstreet Insights writes, Anthropic’s $80.7 billion net loss for half a year is comprised of two fundamentally different accounting components.

Of this, about $72 billion was due to non-cash accounting adjustments. As the company’s private valuation continued to soar in successive funding rounds, U.S. GAAP required the fair value of convertible notes issued to early investors and other complex financing instruments to be marked up accordingly, resulting in large-scale "paper losses" on the books. This accounting approach is not uncommon for high-growth tech startups raising large rounds of funding, though an absolute figure like $80.7 billion is still highly unusual.

Stripping out this accounting adjustment, the company’s actual operating loss in the first half of the year was still an enormous $8.7 billion—already exceeding the total annual operating loss of over $8 billion for all of 2025. The core driving factor is the explosive growth in compute spending: in the first half of 2026, Anthropic’s direct spending on compute resources and infrastructure approached $9 billion, nearly four times the $2.3 billion spent in the first half of 2025, and even exceeding the total $7.3 billion spent on compute in all of 2025.

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

The $750 Billion Compute Bill: From Flexible Spend to Irrevocable Liabilities

The real risk of Anthropic’s compute investments lies not in the sheer scale, but in the shift in contract structure from flexible, pay-as-you-go arrangements to fixed, irrevocable long-term liabilities.

As Wallstreet Insights mentions, according to the Wall Street Journal, the company’s commitments to compute infrastructure spending have surpassed $600 billion. If an additional $150 billion deal being negotiated materializes, the total will reach $750 billion. According to the latest prospectus, the company will be obliged to honor about $450 billion in cloud computing contracts over the next seven to ten years, with most agreements explicitly stating payment is due "regardless of actual usage."

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

The prospectus also disclosed a supply-side dilemma: Anthropic has sought more computing power than AWS, Google, and Microsoft are willing to provide, partly because the three hyperscale cloud service providers increasingly see it as a direct competitor. After being forced to turn to alternative suppliers, the company began building its own data centers and partnering with third-party facility operators. The prospectus admits these operators have "limited operational history," and their reliability is uncertain. It also mentions that some computing power arrangements are financed off-balance-sheet through special purpose vehicles (SPVs) and private credit instruments, further increasing uncertainty around financial transparency.