On September 29, 2026, a draft of Anthropic's IPO prospectus was disclosed by Reuters and quickly spread. The document reveals the truest picture of the company's finances: 2025 revenue was nearly $4.6 billion, nearly twelve times the roughly $400 million in 2024; at the same time, however, operating losses exceeded $8 billion, and the net loss reached $42 billion.
The $42 billion net loss is a startling figure, but it needs to be broken down to be understood. According to investment news website Investment News, about $34 billion of it is non-cash accounting charges—"losses" created under accounting rules after valuation increases in early financing arrangements, rather than actual cash outflows. Excluding this portion, Anthropic's actual operating loss is about $8 billion.
$518 Billion: The Structure of This Bill Is More Dangerous Than the Total
The line in the prospectus most worth reading over and over is this: the company has committed to spending about $518 billion over the next several years on cloud services, compute, and infrastructure. As of the end of 2025, the company held about $20.3 billion in cash and short-term investments.
There is an order-of-magnitude gap between the two figures, but the more critical detail lies in the composition of the $518 billion. According to Investment News, about 80% of it consists of irrevocable minimum payment commitments—regardless of how much compute Anthropic actually uses, the money must be paid out.
The breakdown: Google will receive at least $111.1 billion (agreement period April 2026 to July 2033), Amazon will receive $110 billion (May 2026 to April 2036), and Microsoft will receive $31.4 billion (November 2026 to May 2033). The original prospectus text states: "If actual spending does not reach the committed amount, we must pay Google the difference," and Amazon's agreement contains a similar provision. Microsoft's agreement may be canceled only in the event of an uncured material breach by Microsoft. In addition, Broadcom-related equipment lease obligations are about $161.2 billion, and SpaceX's compute agreement is worth up to $84.5 billion (but is relatively flexible and can be terminated with 90 days' notice).
This means Anthropic has in effect locked in its procurement list for the coming seven to ten years in advance, and cannot adjust it substantially as the market changes. This is not a budget plan but a legal contract.
Can 12x Revenue Growth Support This Bill?
To turn the $518 billion obligation into a manageable liability, Anthropic needs sustained, large-scale revenue growth while depending on this IPO to raise sufficient funds.
In terms of revenue trajectory, it went from about $400 million in 2024 to about $4.6 billion in 2025. But there are several structural risks behind this: First, the prospectus discloses that Anthropic's two largest customers together accounted for about 25% of 2025 revenue. Second, most of these large customers have not signed long-term contracts and can reduce or stop purchasing at any time. Single-customer dependence plus contract gaps means that once revenue growth slows, the pressure to pay subsequent compute bills will rise sharply.
Another figure is the growth rate of compute costs. In 2024, Anthropic spent about $2.5 billion on compute and infrastructure; in 2025, that figure jumped to $7.3 billion, nearly tripling and accounting for 58% of total operating expenses. Revenue is growing, but compute cost growth is equally steep, and a considerable portion consists of "pay-first, use-later" contractual obligations.
Valuation Logic: A Half-Year Leap from $965 Billion to $2 Trillion
In May 2026, Anthropic completed its Series H financing at a post-money valuation of $965 billion. According to SmartAsset, the potential valuation reported for this IPO exceeds $2 trillion—in less than six months, valuation expectations have more than doubled.
What does $2 trillion mean? Currently, only a handful of companies worldwide have a market value above $2 trillion, and all are enterprises that have achieved large-scale, stable profitability. Anthropic's operating loss still exceeds $8 billion, and compute costs are still climbing rapidly. The logic supporting this valuation is the line explicitly stated in the prospectus: the impact of AI will exceed that of industrialization, electrification, and the internet combined.
This is an unfalsifiable premise. The problem is that being unfalsifiable does not mean no risk must be borne when investing.
The Signal Behind the SpaceX Partnership
In the $518 billion compute map, the appearance of SpaceX is a signal. According to multiple technology media reports, Anthropic signed a compute agreement with SpaceX worth up to $84.5 billion, running through 2029, providing compute infrastructure based on NVIDIA hardware.
Unlike the irrevocable commitments to cloud giants such as Google and Amazon, the SpaceX agreement allows termination with 90 days' notice, a rare flexible provision on this contract list. Allocating nearly one-sixth of compute procurement to SpaceX reflects, on the one hand, backup demand when traditional cloud providers face supply tightness, and on the other hand shows that Anthropic intends to build some diversity in its compute supplier structure.
However, even if the $84.5 billion can be terminated early, termination itself entails decision costs. More importantly, nearly three-quarters of the total is concentrated in Google and Amazon, which are also important distribution channels for the Claude model. The overlap between the roles of compute supplier and business partner creates an inherent tension of interest in commercial negotiations.
Why Go Public, and Why Now
Back to the most fundamental question: Why does Anthropic want to go public?
The answer is implicit in the numbers. The company holds about $20.3 billion in cash but has signed contractual obligations exceeding $500 billion. This gap cannot be filled by revenue alone, especially since a large portion of the contractual obligations are minimum payment commitments. Going public is fundraising; it is converting future growth expectations into real cash available now.
Anthropic officially confidentially submitted its draft S-1 to the SEC on June 1, 2026; prospectus details were revealed on September 29; and the IPO is reportedly possibly scheduled after the U.S. midterm elections in November 2026. The choice of this time window reflects both market-timing considerations and the need to wait for the SEC to complete its review.
The company's prospectus candidly mentions that AI may pose an existential risk to humanity—this is a company publicly admitting that it is creating "what may be the most dangerous technology ever." This unusual self-disclosure is a true reflection of the founders' belief system and one of the risk factors for future regulatory friction.
Conclusion: This Is Not an Ordinary Tech Company IPO
Anthropic's IPO is highly unconventional in its financial structure: revenue is growing rapidly but remains relatively limited in scale; losses are large but mostly accounting-related; compute commitments far exceed cash on hand, and a considerable portion is irrevocable.
This is not a profitable company choosing to go public to cash out, but a company still in its cash-burning stage that needs the capital markets to support a bill it has already signed. Whether the IPO succeeds and whether the valuation holds depends on whether the market is willing to pay in advance for the proposition that "AI will rewrite human civilization."
Over the past two years, the technology market has paid high premiums for several similar propositions. But $518 billion in locked-in obligations, two major customers contributing a quarter of revenue, and no long-term contract guarantees—these three factors combined constitute the risk variable in Anthropic's IPO story that most needs continuous tracking. The value of the prospectus is not that it confirms an optimistic narrative, but that for the first time it allows outside observers to directly read the company's real books.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接