Dario Amodei has spent years telling the public that the technology he sells could be among the most dangerous in human history. He has said it could wipe out half of entry-level white-collar jobs and push unemployment into double digits. Now his company's IPO filing shows who gets to decide what happens next: seven co-founders, holding 50.1% of the vote through a special class of shares, answerable to no one who buys in.
The numbers first. Anthropic reports $16.29 billion in first-half revenue, up from $1.16 billion a year earlier. That is not a growth curve but a cliff face. Cost of revenue was just $2.38 billion, almost all of it compute. On paper, that looks like a software company with fat margins. Then comes the other figure, reported in a separate account of the prospectus: an $80 billion net loss for the same six months. That gap is the story. Whatever sits between those numbers, whether it is the cost of training the next model, the buildout of capacity, or something else the full filing will explain, it shows a business whose real expenses are the arms race itself. The company concedes that margins will swing as computing capacity comes online. In plainer terms, nobody outside the building knows what this business looks like at rest, or whether it ever rests.
Now ask who is being invited to absorb that uncertainty. Public investors, which means pension funds, index funds and eventually the 401(k)s of ordinary workers, will be asked to fund a capital burn that would make a 1990s telecom executive blush. In return they get shares with a minority say. The Class F structure guarantees that the people who set the pace of the race, the risk appetite and the spending are the same seven who set it before the money arrived.
The defense is predictable. Dual-class structures are common. Meta and Alphabet have them. Founders with a vision shouldn't be hostage to quarterly earnings. Anthropic is also a public benefit corporation, a mission-minded outfit and not just another profit machine. All true, and all beside the point. Mark Zuckerberg's control has not been a reassuring model of corporate conscience. And a public benefit charter is a promise the company makes to itself, enforced mostly by the people making it. Anthropic's pitch has always been that it is the responsible one, the lab that worries. But worry is not accountability. A conscience that cannot be outvoted is just a mood.
The broader public holds the largest stake and has no seat at all. Data-center buildouts strain local grids and water supplies. Communities give up tax revenue to land them. The labor disruption Amodei himself forecasts will land on workers who never bought a share. Anthropic also paid $1.5 billion to settle claims by authors whose books were used to train its models, a reminder that the industry's growth was built partly on other people's work. The costs are socialized, and now the upside is being privatized through a governance structure designed to keep outsiders out.
This is also a test for my own side. Democrats have been eager to court AI money, welcome the data centers and pose with the founders, while offering little beyond vague talk about guardrails. If the company that brands itself as safety-first can go public and lock in founder control without a serious fight over disclosure, labor protections, energy costs and real public oversight, then the Left will have mistaken a friendly tone for a regulatory regime.
Let the roadshow proceed. But the rest of us should ask the question the prospectus is built to avoid: when the people who warned us about the risks also hold the controlling vote on taking them, who exactly is the check?
How it may affect me
If Anthropic lists, its shares could eventually find their way into index funds and retirement accounts, meaning ordinary savers may hold a stake with almost no voting power. Check how your funds treat multi-class companies. Separately, data-center growth can affect local electricity costs, so watch utility-commission proceedings in your area.
