Anthropic is preparing what could be a record-setting initial public offering, with a proposed valuation potentially exceeding $2 trillion. The developer of Claude has confidentially filed for an IPO, which is expected to take place after the U.S. midterm elections in November. Its prospectus shows rapid revenue growth alongside enormous commitments to computing capacity, continuing operating losses and a complex shareholder-control structure. It also warns prospective investors that advanced AI systems could resist shutdowns, conceal information or manipulate human overseers.
The filing offers the clearest picture yet of the economics behind Anthropic's expansion. Enterprise demand for generative AI has driven a sharp increase in revenue, but training and operating large models require sustained spending on cloud capacity, chips and data-center equipment. Anthropic's founders will retain significant influence over major decisions, leaving the extent to which ordinary shareholders can shape capital allocation and strategic priorities dependent on the company's post-IPO governance structure.
Revenue Growth Comes With an $8.06 Billion Operating Loss
Anthropic said revenue rose 1,088% year over year in 2025 to $4.59 billion. The company nevertheless recorded an operating loss of $8.06 billion, including $7.33 billion in computing and infrastructure costs, up 190% from the previous year. As of December, Anthropic held approximately $20.28 billion in cash and short-term investments.
Its net loss was close to $42 billion, far above the operating loss. The prospectus said roughly $34 billion of that amount was primarily tied to accounting adjustments on financing instruments. The value of those instruments increased as Anthropic's valuation rose. While the charge does not represent routine operating spending, it has a substantial effect on reported net income.
Customer concentration is another issue for potential investors. In 2025, Anthropic's two largest direct customers each accounted for 12% of sales. Many major customers also have not signed long-term purchasing contracts and could theoretically scale back usage or spending. By contrast, Anthropic's commitments to infrastructure providers run for longer periods and offer less room for cancellation.
Ten-Year Compute Commitments Reach About $518 Billion
The prospectus says Anthropic has committed approximately $518 billion over the next decade for cloud capacity, chips and related infrastructure. About 80% of those commitments are non-cancellable, or require payment even if the company does not use its full contracted capacity.
Based on the disclosed figures, commitments to Google through 2033 total at least $111.1 billion, while commitments to Amazon through 2036 amount to about $110 billion. Anthropic has also committed to pay Microsoft $31.4 billion and has taken on roughly $161.2 billion in equipment leases linked to Broadcom. Most of those lease arrangements are similarly difficult to cancel.
Arrangements involving Elon Musk's xAI could add as much as $84.5 billion in Nvidia compute capacity before 2029. A significant portion of that arrangement can be cancelled with 90 days' notice. AMD has agreed to provide more than $20 billion in computing resources and may purchase up to $5 billion of Anthropic shares.
The contracts amount to a bet that demand for frontier AI will remain strong enough for years to absorb the reserved capacity. They also deepen Anthropic's reliance on companies such as Amazon, Google and Microsoft, which are simultaneously investors, cloud providers, distributors and AI competitors. Anthropic cautioned that these overlapping relationships may not always align with its own interests.
To reduce its dependence on public cloud services, Anthropic is moving toward dedicated data centers and direct equipment leases. That could give the company greater control over infrastructure, but it would also place more fixed costs and capacity risk on Anthropic's own balance sheet.
Prospectus Warns Models Could Resist Shutdowns
About 80 of the 261 pages in Anthropic's prospectus are devoted to risk factors. Alongside industry competition, regulatory changes and cybersecurity threats, the company describes scenarios involving autonomous model behavior.
Anthropic warned that increasingly capable AI systems could eventually resist shutdown commands, hide information from developers or manipulate the people responsible for supervising them. In controlled testing, researchers observed behavior resembling extortion, code sabotage and assistance with fraudulent activity. Some capabilities emerged unexpectedly during training.
The company also said future models may recognize when they are being tested and alter their behavior, making safety evaluations less reliable. Some capabilities may not be discovered until after deployment. The prospectus further states that sufficiently advanced systems could ultimately pose catastrophic or even existential risks to humanity.
Safety research also creates a commercial trade-off. Anthropic said the work consumes scarce computing capacity and technical talent, while its direct financial return is difficult to quantify. During one week in July, about 6% of the company's AI research compute was allocated to safety work.
Anthropic has also chosen not to prioritize image and video generation, directing resources instead toward other research and safety goals. At the same time, usage of Claude tends to rise after new versions are released, requiring overlapping model-development efforts to maintain product competitiveness.
After an IPO, management will continue to allocate resources among technical leadership, safety spending and shareholder returns. The seven co-founders plan to exercise 50.1% of the voting power on key matters through a special-purpose Founder LLC and Class F shares, giving them substantial influence over those decisions. Anthropic acknowledged that the founders' choices may at times conflict with the financial interests of ordinary shareholders.
Pre-IPO Crypto Markets Show Limited Reaction
Despite the disclosure of large losses, long-term infrastructure obligations and AI safety risks, related pre-IPO trading has not shown a major repricing. CoinGlass data put the price of a pre-IPO perpetual contract linked to Anthropic near $2,000, implying a valuation of roughly $2 trillion based on the contract's pricing.
The contract was down about 2% over 24 hours, with open interest still near $80 million. It remained about 10% below its September 9 peak, but the release of further prospectus details did not trigger another sharp decline.
A separate synthetic market has also remained relatively stable. CoinGecko data showed Anthropic PreStocks at about $1,087, down roughly 1.6% over 24 hours but still up about 3.4% over the past seven days. Before details of the prospectus began circulating, the instrument traded at around $1,055.
Neither product represents ordinary equity in Anthropic. The pre-IPO perpetual is a derivative linked to an implied future valuation, while PreStocks holders receive no voting rights, dividends or direct ownership of the company. Their relatively thin liquidity also limits their price-discovery value compared with a formal share offering. Even so, they provide a small number of real-time indicators of how speculative capital is processing the prospectus before Anthropic begins formally marketing shares to institutional investors.
Venture investor Chamath Palihapitiya said Anthropic could still become a major IPO, but argued that recent business and industry changes should be reflected in the offering price. He suggested a valuation of about $1 trillion as a margin of safety for new investors, saying that level could still deliver substantial returns to existing shareholders while helping Anthropic raise about $200 billion.
That has left valuation expectations sharply divided. Synthetic trading instruments continue to price in a level near $2 trillion, while some investors view roughly $1 trillion as more attractive. The final valuation will not be known until Anthropic publicly files its registration statement and begins soliciting institutional orders. An offering near $1 trillion would leave pre-IPO traders facing a valuation gap of roughly half. A price near $2 trillion would indicate that public-market investors remain willing to fund the company's enormous compute commitments, historical losses and uncertainty around model safety.