Anthropic CEO Dario Amodei shared his thoughts on if the AI industry was in a bubble at The New York Times DealBook Summit on Wednesday. This was in addition to throwing shade on one particular unnamed competitor, which was clearly OpenAI.
Amodei declined to give a simple yes-or-no answer to the question of a bubble, saying it was a complex situation, but instead explained his thoughts about the economics of AI in more detail.
He described himself as bullish on the potential of the technology, but cautioned that there could be players in the ecosystem who might make a “timing error” or could see “bad things” happen when it comes to the economic payoffs.
“There’s an inherent risk when the timing of the economic value is uncertain,” Amodei explained. He said companies had to take risks to compete with each other and authoritarian adversaries — a reference to the threat from China — but added that some players were not “managing that risk well, who are taking unwise risks.”
The issue, he said, is the uncertainty around how quickly the economic value of AI will grow and properly mapping that to the lag times on building more data centers.
“There’s [a] genuine dilemma, which we as a company try to manage as responsibly as we can,” Amodei said. “And then I think there are some players who are ‘YOLO-ing,’ who pull the risk dial too far, and I’m very concerned,” he added, using the slang term for “you only live once,” which is often used to justify risk-taking.
Plus, he spoke to the question around AI chips’ deprecation timelines. That’s another hot-button topic and a factor that could negatively impact the industry’s economics if GPUs become obsolete and lose their value ahead of schedule.
“The issue isn’t the lifetime of the chips — chips keep working for a long time. The issue is new chips come out that are faster and cheaper…and so the value of old chips can go down somewhat,” Amodei said.
He said Anthropic was making conservative assumptions on this front and others as it planned for an uncertain future.
The AI company’s revenue has grown 10x per year over the past three years, the CEO said, going from zero to $100 million in 2023, then $100 million to $1 billion in 2024, and will land somewhere between $8-10 billion by the end of this year.
But Amodei said he would be “really dumb” to just assume that the pattern would continue. “I don’t know if a year from now, if it’s going to be 20 billion or if it’s going to be 50…it’s very uncertain. I try to plan conservatively. So I plan for the lower side of it, but that is very disconcerting,” he said.
AI companies like his have to plan how much compute they’ll need in the years ahead, and how much they should invest in data centers. If they don’t buy enough, they may not be able to serve their customers. And if they buy too much, they’ll struggle to keep up with costs or, in the worst-case scenario, they could go bankrupt.
Last month, OpenAI landed in a PR crises when its CFO said she wanted the U.S. government to “backstop” her company’s infrastructure loans, aka insure them so taxpayers would pick of the bill if OpenAI could not. After the furor, she walked back the comments.
Those who take more risks could overextend themselves, Amodei warned, especially if “you’re a person who just kind of, like constitutionally, just wants to ‘YOLO’ things, or just likes big numbers,” he said, in a veiled reference to OpenAI CEO Sam Altman.
“We think we’re going to be okay in, basically, almost all worlds…I can’t speak for other companies,” he said.
