Anthropic CEO weighs in on AI bubble talk and risk among competitors

Anthropic CEO Dario Amodei shared his thoughts on whether the AI ​​industry is in a bubble at the New York Times DealBook Summit on Wednesday. This was in addition to throwing shade at a particular unnamed competitor, which was clearly OpenAI.

Amodei declined to give a simple yes or no answer to the question about the bubble, saying it was a complex case, but instead explained his thoughts on the economics of AI in more detail.

He described himself as optimistic about the potential of the technology, but warned that there may be players in the ecosystem who might make a “timing error” or might see “bad things” happen when it comes to economic rewards.

“There is an inherent risk when the timing of economic value is uncertain,” Amodei explained. He said companies have to take risks to compete with each other and with authoritarian rivals – a reference to the threat posed by China – but added that some players “are not managing those risks well, and are taking unwise risks.”

The problem, he said, is uncertainty about how quickly the economic value of AI will grow and properly mapping that to lags in building more data centers.

“there [a] “It’s a real dilemma, and we as a company are trying to manage it as responsibly as possible. I think there are some players who are taking too much risk, and I’m very concerned,” Amodei said, using the colloquial term “you only live once,” which is often used to justify risk-taking.

Additionally, he raised a question about timelines for decommissioning AI chips. This is another hot topic and a factor that could negatively impact the economics of the industry if GPUs become obsolete and lose value prematurely.

“The problem is not the lifespan of the chips, the chips keep working for a long time,” Amodei said. “The problem is that new chips come out faster and cheaper, so the value of old chips can depreciate somewhat.”

He said Anthobik was making conservative assumptions on this and other fronts as she planned for an uncertain future.

The CEO said that the artificial intelligence company’s revenues have increased 10 times annually over the past three years, rising from zero to $100 million in 2023, then from $100 million to $1 billion in 2024, and will reach between $8 to $10 billion by the end of this year.

But Amodei said it would be “really stupid” to assume that pattern would continue. “I don’t know if it’s a year from now, or if it’s going to be $20 billion or if it’s going to be $50 billion,” he said. “It’s very uncertain. I’m trying to plan conservatively. So I’m planning on the lower side of it, but that’s very concerning.”

AI companies like his have to plan how much computing they will need in the coming years, and how much to invest in data centers. If they don’t buy enough, they may not be able to serve their customers. If they buy too much, they will find it difficult to keep up with costs, or in a worst-case scenario, they may go bankrupt.

Last month, OpenAI was hit with a PR crunch when its CFO said she wanted the US government to “back” her company’s infrastructure loans, also known as underwriting them so taxpayers can foot the bill if OpenAI can’t. After the uproar, she retracted the comments.

Amodei warned that those who take more risks could overextend themselves, especially if “you’re someone who, like Constitution, just wants to micromanage things, or just likes big numbers,” a veiled reference to OpenAI CEO Sam Altman.

“We think we’ll be fine in almost all worlds… I can’t speak for other companies,” he said.