When a man with a possible $15 billion to $20 billion stake in a company tells you that company’s growth proves AI is not a bubble, you should listen carefully and then check his math. Eric Schmidt has done the first part. The second part is harder, because he did not explain how he made his comparison.
Schmidt, who ran Google from 2001 to 2011, said on the In Good Company podcast that Anthropic grew revenue more than twice as fast as Google did under his watch. “That’s real growth there. It’s not fake,” he told Nicolai Tangen, who runs Norway’s sovereign wealth fund. The catch is the disclosure. Bloomberg reported that Schmidt owns roughly 1% of Anthropic, with an IPO expected.
The revenue figures do back him up, at least partly. Reuters reported that Anthropic told investors its third-quarter revenue rose 45% to $16.8 billion, from $11.6 billion in the second quarter and $4.7 billion in the first. It also claimed adjusted profitability. But growth slowed after more than doubling between the first two quarters, and that deceleration is the part bulls tend to skip.
Then there is the cost side. A draft IPO prospectus reviewed by Reuters showed an $80.7 billion net loss for the first half of 2026. Most of that, $72 billion, was a noncash charge tied to convertible financing, so it looks scarier than the underlying operations. Still, Anthropic spent nearly $9 billion on computing in six months and has committed at least $518 billion over ten years. That is a bet on demand continuing to compound for a decade.
This matters because the bubble debate has mostly centered on OpenAI, Nvidia’s customers, and the circular financing between them. Anthropic is the cleanest test case for the opposing argument: enterprise and developer customers paying real money for Claude. If revenue keeps climbing against those commitments, Schmidt looks right. If it flattens, the infrastructure bills look reckless.
His more interesting point is about China. Schmidt thinks Chinese models trail American ones by about six months, but that free alternatives like DeepSeek and Qwen could win most of the world’s users. “If you define losing as users, the vast majority of the world will use the free Chinese models,” he said.
That is the real threat to the paid-model thesis. Anthropic, OpenAI, and Google can stay ahead on quality and still lose volume in price-sensitive markets. For developers choosing a stack, it means open and cheap models will keep pressuring API pricing. For investors, the question is whether premium customers alone can cover a $518 billion tab. Schmidt did not answer that, and given his stake, it is fair to ask him again.
