Anthropic, the company behind Claude, lost $80.7 billion in six months, according to its draft IPO filing. About $72 billion of that happened because the company got more valuable, and if its value had crashed instead, the same accounting rule would have booked it a gain on paper. That sounds made up, so here's how it works. When Anthropic raised money, a lot of it came in through convertible notes and similar deals. Think of them as IOUs that turn into stock later. Accounting rules make the company list those IOUs as something it owes, and price them at whatever that future stock is worth today. Anthropic started 2026 valued at around $350 billion. It raised at $380 billion in February, and by late May it closed a $65 billion round at $965 billion, passing OpenAI. Every jump made those IOUs worth more on paper, and every time that happened the books recorded it as a loss, even though no actual cash went anywhere. The same thing happened last year. Anthropic's 2025 net loss was $42 billion, and more than $34 billion of it came from this exact kind of adjustment, in a year when its value went from $61.5 billion in March to $350 billion by November. Take the $72 billion out and you're left with a loss of about $8.7 billion over six months, on $16.3 billion of revenue. A year earlier, first-half revenue was $1.1 billion. Then Q3 by itself brought in $16.8 billion, more than the whole first half, and Anthropic told investors it was profitable on an adjusted basis that quarter. The number I'd actually watch is somewhere else in the filing. Anthropic said it could spend as much as $750 billion on computing power over the next few years, and two customers made up about a fifth of its 2026 revenue. At last week's investor event, people reportedly talked about a valuation around $1.5 trillion, down from the roughly $2 trillion that was floating around for the November IPO. Anthropic's IPO is going to come down to whether revenue growing 45% a quarter can keep up with a $750 billion computing bill.