Anthropic CEO Dario Amodei sees a disconnect between technological progress and economic reality. In a clear swipe at competitors like OpenAI, he suggests some companies are taking reckless financial risks.
During an interview with Andrew Ross Sorkin, Amodei expressed concern about how some industry players manage risk. While he remains optimistic about the technology itself, he sees economic danger in the massive upfront capital required for infrastructure combined with the risk of companies overextending themselves.
According to Amodei, the core problem is a "cone of uncertainty." Building data centers requires capital commitments one to two years in advance, while future sales remain a question mark.
In this environment, Amodei says some players are "YOLOing" - using the slang for "You Only Live Once" to describe an all-or-nothing investment strategy.
The target of the comment seemed clear. Amodei's warning followed a discussion about Sam Altman's aggressive spending plans at OpenAI. When pressed on the specific target, however, Amodei refused to name names: "That's a question I'm not going to answer."
Playing it safe while rivals gamble
Amodei warns that some companies are operating on assumptions that are “very far inflected on the optimistic.” Anthropic, by contrast, plans conservatively to ensure it stays solvent even in worst‑case scenarios.
“We want to buy enough compute that we’re confident … even in the 10th percentile … scenario … we think we can pay for it,” he said. “We think we’re going to be okay in almost all worlds.”
The CEO also distinguished Anthropic’s business model from its rivals. While Google and OpenAI focus primarily on the consumer market, Anthropic targets enterprise customers. Amodei notes that the consumer sector faces uncertain margins and immense pressure.
The strategy appears to be working. Revenue climbed from zero to $100 million in 2023, hit $1 billion in 2024, and is projected to reach between $8 billion and $10 billion this year.
Despite his economic caution, Amodei trusts the so‑called “scaling laws,” the observation that adding more compute and data yields smarter models. He likens future models to a “country of geniuses in a data center,” AI systems that “can outsmart us in every way.” Rather than pinpointing a specific date for general artificial intelligence (AGI), Amodei describes an exponential curve of continuous improvement.
Amodei's criticism carries some weight given his history. Before launching Anthropic with his sister Daniela, he worked at Google and was an early employee at OpenAI. He helped develop the GPT-2 and GPT-3 models before leaving, reportedly due to disagreements over safety.
