A Hacker News essay argues there is no AI bubble, using Anthropic's S-1

Most online talk about AI treats a crash as a given. Matteo Raso wrote the other side on his blog. The post reached 136 points and 193 comments on Hacker News. Raso is a software engineer. He notes at the end of the post that he is also seeking a Liberal Party of Canada nomination.

His main point is about the shape of the spending. The widely quoted five trillion dollar figure covers data centre build. He calls this a one-off capital cost that drops away once the data centres are running. He also cites Anthropic's S-1 filing for a ratio: the company makes two dollars and eight cents of revenue for every one dollar of compute.

The numbers Raso reads from the S-1

Raso quotes Anthropic's revenue path from the S-1. He cites 41 million in 2023 and 386 million in 2024. Then 4.585 billion in 2025. Then 16.3 billion in the first six months of 2026.

He notes the 72 billion dollars of loss in that same half year. He says this loss is shares that are due to lenders. The lenders are repaid in equity rather than cash. The Verge and the WSJ both reported these S-1 figures when Anthropic filed in late October.

For Alphabet, he cites the published prospectus. Google Services grew by 37.8 billion dollars, a 12 percent rise. Google Cloud grew by 15.5 billion dollars, a 36 percent rise. Google Cloud is the data centre business. Google Services is the product business. The product business added more money in absolute dollars.

Where the essay stops short

Raso describes OpenAI as "the only real unknown". OpenAI is the largest buyer of compute in the industry. A bubble often starts with the firm that has the hardest numbers to defend. Leaving the largest buyer out of the post is a gap.

He also raises two risks and then moves past them. One is that revenue growth may slow down. He says this would change his model. The other is that an autonomous AI hacker could cause serious damage. He says public demand would then drop fast. He does not say what evidence would make him change his view.

His chicken sandwich comparison between Popeyes and model vendors runs thin. Both firms, he says, have no durable advantage and still make money. Fast food margins tell you nothing about how a lender treats an eighty billion dollar loss on an AI book.

What this changes for a team picking an AI vendor

The essay is one person's view on a personal blog. It does not settle anything on its own. What it does give is the figures a buyer should actually read. Anthropic's S-1 is the only AI lab filing audited numbers right now. The ratio of revenue to compute cost is the number that matters to a vendor's survival.

A founder choosing which model API to build on should read both sides. The read is worth an hour before signing a one-year contract. If the figures hold up, the strong AI labs can carry their debt. If the same ratio breaks at OpenAI, the picture gets worse for every vendor downstream.

Source

I'm Not Convinced The AI Bubble Exists, Matteo Raso. Discussion: Hacker News. Primary data: Anthropic S-1 filing.