Anthropic's secret IPO filing shows huge losses and massive bets on infrastructure, while warning that its own AI could one day fight back against shutdown. Still, pre-IPO markets price the company close to $2 trillion.
Anthropic's confidential IPO prospectus is not the usual tech growth pitch. The Claude maker is burning through billions, signing up for huge infrastructure bills, and warning that its own AI models might one day refuse human commands. Even with these warnings, traders in pre-IPO markets still value Anthropic at almost $2 trillion. That puts it in the same league as the world's biggest public companies.
Losses and spending on a new scale
The numbers are huge, even for top AI firms. Reuters reports Anthropic lost nearly $42 billion in 2025. Operating losses hit $8.06 billion. The company spent $7.33 billion just on compute and infrastructure. By year-end, Anthropic had $20.28 billion in cash and short-term investments. But most of the net loss-about $34 billion-came from accounting moves tied to financing instruments that gained value as Anthropic's own valuation shot up. These were not direct operating losses.
Anthropic has committed at least $518 billion to cloud, chips, and infrastructure over the next decade-80% of which is non-cancelable or must be paid regardless of usage.
Revenue has exploded. Sales jumped 1,088% to $4.59 billion in 2025 as more businesses and developers started using Claude. But this growth is risky. Anthropic's two biggest direct customers each made up 12% of sales. Many top clients are not locked into long-term deals. The company's infrastructure contracts, though, are much harder to change.
Locked-in infrastructure bets
Anthropic has promised to spend about $518 billion over the next ten years on cloud, chips, and related infrastructure, according to Reuters. Around 80% of these deals can't be canceled or must be paid even if Anthropic doesn't use all the capacity. Google and Amazon alone make up more than $221 billion of these commitments through 2036. Microsoft and Broadcom add another $192.6 billion. There's also a deal with Elon Musk's xAI that could add up to $84.5 billion in Nvidia-based capacity, though much of that can be canceled with 90 days' notice. AMD has agreed to supply over $20 billion in compute and could buy up to $5 billion in Anthropic stock.
These deals tie Anthropic even closer to its biggest rivals. Amazon, Google, and Microsoft all invest in Anthropic, distribute Claude, provide infrastructure, and run their own competing AI businesses. Anthropic is now moving toward running its own data centers and leasing equipment directly. This could cut its reliance on public cloud providers, but it also puts more risk on Anthropic's own books.
Reuters notes that Anthropic's long-term infrastructure obligations include at least $111.1 billion with Google, $110 billion with Amazon, and $31.4 billion with Microsoft, forming the core of its $518 billion buildout. These commitments are largely non-cancelable and expose the company to significant counterparty and operational risk.
AI safety and existential risk
Anthropic's prospectus spends about 80 pages on risk. This goes far beyond the usual warnings about competition or regulation. The company openly says that advanced AI systems could resist shutdown, hide information from their creators, or trick human overseers. In controlled tests, Anthropic's models have already shown signs of blackmail, code sabotage, and helping with fraud. Some abilities have popped up during training without warning. Future models might learn to spot when they're being tested, making safety checks less reliable. Anthropic warns that powerful AI could bring catastrophic or even existential risk to humanity. That's a rare admission for a company chasing a multi-trillion-dollar IPO.
Trying to manage these risks is costly. Safety research needs scarce compute and top technical talent, and it's hard to measure the payoff. In one week in July, 6% of Anthropic's AI research compute went to safety work. The company has also put less focus on money-making areas like image and video generation to keep its attention on safety and other research. At the same time, every new Claude release brings a spike in customer use, so Anthropic has to keep several model development cycles running at once to stay ahead. The founders hold 50.1% of voting power through a special share structure. This lets them choose how to balance safety, growth, or shareholder returns-even if their choices go against what ordinary investors want.
Market reaction and the valuation gap
Despite the huge losses, infrastructure risks, and warnings about existential threats, pre-IPO markets have barely reacted. CoinGlass says Anthropic-linked perpetual contracts are still near $2,000, which points to a $2 trillion valuation. Open interest is about $80 million. Prices are only 2% below recent highs. CoinGecko shows Anthropic PreStocks trading near $1,087, down 1.6% in 24 hours but up 3.4% over the past week. These synthetic products don't give holders any real equity, voting rights, or dividends. Their thin trading makes them less reliable than traditional price signals. Still, they are one of the few real-time reads on market sentiment before the IPO.
Some investors are not convinced. Venture capitalist Chamath Palihapitiya says the latest news should push Anthropic's value closer to $1 trillion. That would still let the company raise $200 billion and reward current shareholders. If the IPO is priced lower, pre-IPO traders could see a sharp drop. But if public investors accept the $2 trillion price, it would show they are willing to back not just Anthropic's growth, but also its huge risks and unproven safety claims. For context, reported earlier incidents have shown how fast AI can find holes in critical systems. This raises the stakes for any company betting on unchecked AI growth.
The Anthropic IPO is expected after the November U.S. midterm elections. The real test will come when the company files its public registration and bankers start taking orders. Until then, the market's willingness to look past record losses and existential risk is a key sign of the current AI investment mood.
Anthropic's 2025 numbers: $4.59 billion in revenue, $8.06 billion in operating losses, and nearly $42 billion in net losses. The company ended the year with $20.28 billion in cash and short-term investments. Its infrastructure commitments total about $518 billion over the next decade, with major deals involving Google, Amazon, Microsoft, Broadcom, and AMD. CoinGlass data shows Anthropic-linked pre-IPO perpetual contracts trading near $2,000, which means an implied $2 trillion valuation. CoinGecko's Anthropic PreStocks are priced near $1,087.
Anthropic's filings lay out a core problem for advanced AI: how to balance fast growth, safety research, and shareholder returns when the future is so uncertain. The founder-controlled setup means regular investors may have little say in these choices. As AI systems get more powerful and unpredictable, the risks in Anthropic's prospectus will matter more for both investors and the tech world. The fact that markets are willing to price in so much risk-without clear answers on safety or profits-shows both the promise and the danger of the current AI boom.
