Tech • AI • Robotics • Game
We Read Anthropic's Secret Accounts!
TL;DR
A leaked Anthropic IPO filing depicts an AI leader with explosive revenue growth, heavy infrastructure commitments and unusual risk disclosures, intensifying debate over whether frontier AI economics are sustainable.
KEY POINTS
A landmark IPO under scrutiny
Anthropic is preparing what could be the first major public listing of a frontier AI model company, with a target market debut before Thanksgiving and a listing on the Nasdaq. The filing is being treated on Wall Street as a rare full financial x-ray of a top AI lab, offering detail on revenues, customers, contracts and costs that private markets have largely obscured.
Losses of $42 billion require context
The most striking figure is a reported $42 billion loss last year, but most of it was not cash burn. Roughly $34 billion came from the accounting impact of convertible instruments granted to Google and Amazon, meaning earlier shareholders were diluted as the company’s valuation rose. The charge reflects a transfer of value to those investors rather than a direct cash payment by the company.
Revenue growth is extraordinary
The company’s revenue trajectory is unusually steep even by software standards. Revenue was cited at $386 million in 2024, then about $4.6 billion in 2025, with annualized recurring revenue near $65 billion by late summer and projected to reach $100 billion by year-end on a run-rate basis. Analysts cautioned that run rate is not the same as reported annual revenue, which could land closer to $50 billion to $60 billion.
Operating economics look better than the headline net loss
Beyond the accounting loss, the filing suggests Anthropic had reached operational profitability in the second quarter. That matters because the core question for public investors is whether the company can fund the immense costs of training and running models through real business activity rather than repeated fundraising. The company’s ability to offset compute costs with rapid sales growth is seen as one of the filing’s strongest signals.
Compute commitments are massive
The company has reportedly pre-committed about $518 billion in compute and infrastructure spending over roughly a decade, with major obligations including $111 billion with Google, $110 billion with Amazon, $31 billion with Microsoft, and more than $160 billion in equipment linked to Broadcom. Additional commitments tied to Nscale push the total close to $600 billion, underscoring how AI is becoming an infrastructure-heavy business rather than a conventional software story.
The debate is less about funding than durability
Some investors see those commitments as manageable if revenue keeps scaling, arguing compute is now financed like a new infrastructure asset class. The greater concern is whether growth can remain fast enough as enterprise customers become more price-sensitive, compare models more actively, and demand clearer return on investment. Pressure may fall not first on Anthropic itself but on more leveraged links in the chain such as neocloud providers.
Competition is intensifying on price and architecture
Anthropic and OpenAI still dominate the top tier of frontier models, especially for long-horizon agents that can work through complex tasks over extended periods. But competition is widening from open-source models, lower-cost alternatives and new architectures that can surge in adoption quickly. That is forcing routing strategies that shift simpler tasks to cheaper models and could compress pricing over time.
Growth may depend on vertical products, not just model access
Analysts increasingly argue that selling API tokens alone will not justify the valuations being discussed. Future upside may depend on deeper products for developers and industry-specific tools in areas such as healthcare, law and finance. The broader thesis is that value will accrue to AI platforms that can orchestrate enterprise workflows, not merely to whoever has the best model on a given day.
Risk disclosures are unusually expansive
Nearly a third of the filing reportedly discusses risks, including harms framed as potentially existential. Those warnings cover cyber, biological and misuse scenarios, but investors also highlighted a more immediate commercial risk: liability tied to the handling of proprietary client data and the possibility that AI providers could end up competing with their own customers in adjacent markets.
CONCLUSION
The leaked filing shows Anthropic as both a breakout growth company and a test case for whether frontier AI can become a durable public-market business. The IPO will likely shape how investors price not only model makers, but the entire infrastructure and software stack forming around artificial intelligence.
Ask a question