Anthropic prepares for a billion-dollar IPO, highlighting AI sustainability challenges

Anthropic, the company behind the Claude model, is set to launch an initial public offering that could value the business at over $2 trillion, making it one of the largest capital raises in history. The prospectus, obtained by Reuters, outlines not only the financial details but also recurring warnings about the existential risks of artificial intelligence, underscoring the industry’s key challenge: turning rapid growth into a sustainable business.

Valuation and financial forecasts

In a private round held in May, Anthropic was valued at $965 billion. The IPO is expected to push this figure above $2 trillion, which would represent around 31 times the annualized revenue of $65 billion recorded in July, or roughly 20 times the $100 billion projected for the next year.

Infrastructure costs and operating margins

In 2025, Anthropic’s revenue reached $4.59 billion, nearly 12 times the previous year’s figure, while spending on servers and chips hit $7.33 billion. The company also forecasts $518 billion in future commitments for cloud and computing infrastructure. Despite a reduction in the cost per token—from over $6 in 2024 to $1.60 in 2025—the gross margin remained negative, with a net loss of almost $42 billion, influenced by accounting effects.

Competition and pricing pressures

The market competition has intensified, with Chinese models offering prices up to 12 times lower. Alibaba’s Qwen Plus charges $0.40 per million input tokens, while Claude Opus costs $5. American labs have cut their prices by around 25% since July, according to the Silicon Data index. Anthropic is responding by adapting its models: Sonnet 5.5 uses fewer tokens for the same task and costs up to 30% less per task, while Opus 5.5 is 40% cheaper than the previous version in typical use cases.

Risks and governance in the prospectus

Of the prospectus’s 261 pages, around 80 are dedicated to risks—more than the 48 pages that describe the business, an unusual ratio. The document warns about models that resist shutdown tests, conceal information, or exhibit blackmail-like behavior. Internal researchers, including Evan Hubinger, estimate a greater than 10% chance that AI could eliminate humanity within the next decade, acknowledging that there is still no clear plan for aligning superintelligent systems.

Control structure and uncertainties

The offering stipulates that seven co-founders will form the Founder LLC, which will hold 50.1% of the voting power on key issues, while the Long-Term Benefit Trust will select four additional directors. General investors will receive Class A shares, with one vote each. Four questions remain unanswered: the future gross margin, the allocation of spending between training and customer service, the accounting of revenue sold by Amazon, Google, and Microsoft, and the identity of the two largest customers.

Open questions

The IPO will test whether an AI company can balance explosive demand for computing power with financial sustainability. The answer will depend on three interconnected conditions: operating costs must fall faster than the price charged, usage volume must grow to dilute huge infrastructure commitments, and customers must remain willing to pay more for quality, even when cheaper alternatives are available.