Anthropic's Prospectus: The Triple Books of 4.6 Billion Revenue, 42 Billion Losses and 2 Trillion Valuation

A company with annual revenue of 4.6 billion U.S. dollars and operating losses exceeding 8 billion U.S. dollars is seeking a 2 trillion U.S. dollar valuation from the capital market — this is not science fiction, but the prospectus that Anthropic is submitting to Wall Street. How to make sense of these figures? Three numbers, three layers of logic, and a narrative chain stretched to its extreme.

The Truth Behind the $42 Billion Loss: "Paper Losses" in Accounting Entries

First, let's break down the most striking figure: a net loss of 42 billion U.S. dollars. This number is enough to make the CFO of any traditional industry resign on the spot, but it includes approximately 34 billion U.S. dollars in non-cash accounting expenses, which reflects the increase in fair value of equity-convertible instruments in previous financing arrangements, rather than operating cash outflows. What truly reflects operating quality is the operating loss — after excluding the impact of this liability impairment, the GAAP operating-level loss stands at 8.06 billion U.S. dollars, compared with 2.98 billion U.S. dollars in 2024.

An operating loss of 8.06 billion U.S. dollars corresponding to revenue of 4.59 billion U.S. dollars means that for every 1 dollar earned, about 1.75 dollars is burned. In 2025, computing power and infrastructure expenditures reached 7.33 billion U.S. dollars, three times the figure in 2024, accounting for more than half of the total operating expenditures of 12.65 billion U.S. dollars. By the end of 2025, the company held a total of 20.28 billion U.S. dollars in cash and short-term investments — at the current cash burn rate, this "reserve" is enough to sustain operations for about two and a half years.

The problem is that the cash burn rate is still accelerating. The prospectus discloses that the company plans to undertake 5.18 trillion U.S. dollars of performance obligations in cloud computing, computing power and infrastructure in the coming year. This figure is 112 times the current annual revenue, and even for an AI company labeled with "exponential growth", the size of this commitment is enough to make prudent public market investors gasp.

Growth Quality of $4.6 Billion Revenue: Fast but Unstable

Revenue surged 12 times to 4.59 billion U.S. dollars, an astonishing growth rate. But if we look into the revenue structure, hidden risks emerge.

About 75% to 85% of Anthropic's revenue comes from enterprise-level API calls, which is completely different from OpenAI's early model dominated by consumer subscriptions. The stickiness of enterprise-level APIs is usually stronger than that of consumer subscriptions, which is an important support for Anthropic's valuation narrative. From the perspective of the competitive landscape of the enterprise market, Anthropic's customer stickiness is indeed increasing. The Ramp AI index shows that as of April 2026, 34.4% of U.S. enterprises pay for Anthropic, surpassing OpenAI's 32.3% for the first time, and its share in the enterprise LLM market has also reached 32%.

However, it is necessary to distinguish that this is an indicator of enterprise adoption rate, which measures how many enterprises are paying for a certain AI vendor, rather than the proportion of a company's revenue coming from enterprises. In terms of the proportion of enterprise revenue, OpenAI's revenue structure has reversed rapidly in 2026 — at the beginning of the year, enterprise revenue only accounted for about 40%, and by mid-year, it had surpassed the consumer business, making enterprise revenue the main source of the company's income. Therefore, Anthropic's lead in enterprise adoption rate is still two separate issues from whether it can convert this advantage into sustained revenue quality and profits.

What is more alarming is revenue concentration. The prospectus discloses that nearly a quarter of the revenue comes from two customers, and many of the largest customers have not signed long-term contracts, so they may reduce or stop spending in the future. A more specific risk is that about 45% of the revenue comes from Amazon — which is not only Anthropic's largest computing power supplier, but also its largest source of revenue.

This means there is a structural paradox in Anthropic's revenue structure: it pays Amazon for computing power, and Amazon returns the money to it in the form of API calls. The prospectus does not provide a clear accounting split to show whether this part of the revenue is real end-user demand, or just "circular cash flow" within the cloud vendor's ecosystem. When Amazon is simultaneously the landlord, supplier and largest tenant, the "purity" of this revenue has to be discounted significantly.

The $5.18 Trillion Computing Power Bet: Locking in the Future, as well as Risks

Anthropic's computing power commitment is not just an expenditure, but a strategic binding. The company promises to invest more than 1 trillion U.S. dollars on AWS in the next ten years, and procure up to 5 gigawatts of AWS Trainium chip computing power. At the same time, Anthropic also promises to invest about 2 trillion U.S. dollars in Google's cloud services and chips.

Adding up Anthropic and OpenAI's cloud expenditure commitments, the two companies account for nearly half of the approximately 2 trillion U.S. dollars of "backlogged revenue" of the four major cloud service providers: Amazon AWS, Microsoft Azure, Google Cloud and Oracle Cloud. This figure reveals a neglected reality: deep symbiosis has been formed between AI start-ups and cloud giants. Cloud vendors lock in large customers through strategic investment, and AI companies obtain preferential pricing and priority supply through long-term computing power commitments, and the two sides jointly push up each other's book value.

However, the fragility of this symbiosis is equally obvious. If the actual commercial returns of AI models fall short of expectations, Anthropic will face dual pressures: on the one hand, the huge computing power performance obligations cannot be easily reduced, and on the other hand, customers may reduce API calls due to budget cuts. The prospectus expands the risk factor section to about 80 pages, almost twice the 48-page business description section, and the company frankly admits that its models may exhibit behaviors such as "refusing to shut down", "concealing or manipulating information". This candor itself is respectable, but it also indicates from the side that the company does not have 100% certainty of control over the direction of its own technology.

The $2 Trillion Valuation: Explaining Today's Price with the 2028 Narrative

If calculated simply using current revenue, the 2 trillion U.S. dollar valuation corresponds to an annualized revenue run rate of about 31 times. Even by the standards of fast-growing tech companies, this multiple is in an extreme range.

The key to supporting this valuation is the forward forecast in the prospectus: the company expects its 2028 revenue to reach approximately 190 billion to 200 billion U.S. dollars. If calculated based on 200 billion U.S. dollars in revenue, the forward revenue multiple corresponding to the 2 trillion U.S. dollar valuation will drop to about 10 times — turning a "crazy" figure into a "discussable" one.

However, this growth curve from 4.6 billion to 200 billion requires adding two more current Anthropic entities in less than three years. During this period, it also has to face head-on competition from OpenAI — the latter plans to raise a new round of financing at a valuation of 1.2 trillion U.S. dollars, and has made it clear that it will not go public in 2026. OpenRouter data shows that recently, user spending on OpenAI's models has for the first time exceeded spending on Anthropic's models, and user growth for its flagship models has shown signs of stagnation.

A more macro test comes from the market environment. Anthropic has postponed its IPO to after the U.S. mid-term elections in November, and AI and chip stocks have recently seen sell-offs. Some analysis points out that if AI liquidity tightens, the valuation of AI-related stocks in the S&P 500 may face a 20% to 45% downside risk. After SpaceX's stock price surged 19% on the first day of listing, it has fallen back to near the issue price, providing a less reassuring reference for the market performance of high-valuation tech stocks.

Essentially, Anthropic's prospectus is a statement of belief about the future of AI: it believes that its 2028 revenue can grow 40 times compared to today, that the moat of enterprise-level APIs is deep enough, and that the scale effect of computing power investment will eventually bring a profit inflection point. These beliefs may come true, or they may not. But one thing is certain — the public market will not pay for today's price with "forward narratives" the way the private market does. When the 2 trillion U.S. dollar valuation needs to be cashed out, what Anthropic needs to present is not just the next version of Claude, but a truly viable profit model.

This article is from the WeChat official account "Tech Business", and is republished with authorization from 36Kr.