Anthropic IPO prospectus reveals $42B 2025 loss and $518B planned infrastructure spend

Anthropic's prospectus gives the market its first audited look at frontier-lab economics. The company lost $42 billion in 2025. It plans $518 billion in infrastructure spending. Enterprise customers account for about 80% of its business, which confirms that Anthropic's strategy centers on corporate and developer workloads, not consumer chat.

The spending figure is the headline. $518B in planned infrastructure places Anthropic among the largest compute buyers in the world, comparable to hyperscaler capex programs. It implies long-term commitments with cloud partners including AWS and Google. The loss reflects training and inference costs outpacing revenue even as enterprise adoption grows.

The filing sharpens a debate already live in developer communities. HN threads argue that closed frontier labs cannot raise API prices without pushing teams to open alternatives, because current infrastructure burn looks unsustainable. The contrast with DeepSeek, which just passed a $1B run rate on ultra-sparse models priced far below rivals, is stark. An FT report that corporate America is moving toward cheaper open models added fuel.

Skeptics on HN also ask whether Anthropic's 'safety' positioning works partly as gatekeeping, especially amid complaints about Claude 5.5's restrictive cyber safeguards. Investors will focus on the path to margins: whether enterprise pricing power holds as Sonnet 5.5, GPT-6.1 Sol and DeepSeek compete on cost. Watch for valuation details, the listing timeline and underwriter commentary.