Anthropic posted an $80 billion net loss in the first half of 2026, a number that sounds like a company on fire. It is not. Roughly $72 billion of that loss — 90 percent — comes from non-cash accounting charges related to convertible notes and other financial instruments revaluing as the company's implied worth shifted. This is mark-to-market bookkeeping, not operational hemorrhaging. Strip the accounting noise and you are looking at a company burning roughly $8 billion on actual operations in six months, which is still enormous but structurally different from what the headline implies. The number that matters more is $750 billion. That is the figure Anthropic has indicated it could spend in coming years on computing capacity, including deals currently under discussion. This is not a forecast of revenue or profit. It is a statement of intent to lock up a massive share of global AI compute infrastructure — chips, data centers, power — in a bet that whoever controls the most compute wins. AI infrastructure spending already nearly quadrupled to $9 billion in the first half of 2026 alone. Revenue, meanwhile, is genuinely impressive on a growth basis: $16.3 billion in H1 2026, up from $1.1 billion a year earlier. That is a roughly 15x year-over-year jump, driven largely by Claude Code's adoption among developers. But revenue of $16.3 billion against $9 billion in infrastructure spend and the operational burn means the company is still deeply cash-flow negative on its actual business, before you even get to the accounting charges. Investors are noticing the gap. At Anthropic's investor day in San Francisco last week, attendees discussed a valuation of around $1.5 trillion — a significant haircut from the $2 trillion figure previously expected for the November IPO. Four sources told Reuters that the latest financial disclosures prompted the reassessment. The market is doing what markets do when confronted with extreme capital intensity and uncertain unit economics: it is demanding a discount. The risk disclosures in the draft prospectus deserve attention. Anthropic itself flagged uncertainty over preventing its products from enabling sophisticated cyberattacks or operating autonomously in ways the company cannot control. It also reported last week that AI could be months away from learning to self-replicate on everyday home devices. A company warning regulators about its own product's potential to go rogue, while simultaneously asking those same markets for a trillion-dollar-plus valuation, is a peculiar kind of transparency. The structural question is whether the AI infrastructure race is generative or extractive at this scale. Anthropic, OpenAI, Google, and others are collectively committing trillions to compute buildouts. If the technology delivers broad productivity gains, this capital formation could be the most generative investment cycle in decades. If it does not — if returns concentrate among a handful of model providers while costs socialize through energy consumption, resource extraction, and labor displacement — then this is the most capital-intensive rent-seeking operation in history. The answer is not yet knowable, but the magnitude of the bet is. The IPO will test whether public markets share private investors' faith. A $1.5 trillion valuation for a company with $16.3 billion in half-year revenue and no path to near-term profitability prices in a future where Anthropic captures an extraordinary share of global economic output. The market is being asked to fund the build-out first and find out later whether the returns justify it.
