TL,DR: the cost of intelligence is commoditizing via cheap open-source models. The Anthropic & OpenAI valuations are very reminiscent of the L1 premium fat protocol thesis. It’s difficult to justify a $2 trillion IPO valuation when Anthropic is operating at an $8.06b net loss ($42b if we don’t exclude non-cash charges*). And the constant advocacy for self-regulation and “pacing the frontier” is making their case even harder. At a $2 trillion implied valuation, you could make the case that Anthropic is simply overvalued.

Context:

Last week Anthropic’s IPO S-1 prospectus leaked. And the general sentiment was less than reassuring about its rumored $2 trillion target valuation. For context, Anthropic generated ~$4.6 billion of revenue in 2025, roughly 12x the prior year. But costs scaled even faster, with ~$12.65b in total operating expenses, of which $7.33b of that being compute and infrastructure related, netting ~ $8.06b in operating loss.

Slight caveat: A lot of headlines also quote a $42 billion net loss, but roughly $34 billion of that came from an accounting charge related to financing instruments that could convert into equity rather than operating cash burn, so we’ll assume the a