While AI hyperscalers are committing hundreds of billions of dollar per year on capital expenditures, Anthropic’s spending plans are more cautious by comparison.

But cofounder and CEO Dario Amodei said the reason for his more measured approach is because even a slight miscalculation could sink the company.

In an interview with Dwarkesh Patel on Friday, the podcaster asked why Anthropic, the developer of the Claude chatbot, doesn’t spend more aggressively, given Amodei’s earlier prediction that an AI data center could one day be a “country of geniuses.”

Amodei replied that while he is confident the technical milestone is achievable soon, he’s less certain about the timing of the economic returns.

“I really do believe that we could have models that are a country of geniuses in the data center in one to two years,” he added. “One question is: How many years after that do the trillions in revenue start rolling in? I don’t think it’s guaranteed that it’s going to be immediate. It think it could be one year. It could be two years. I could even stretch it to five years, although I’m skeptical of that.”

Because of this uncertainty on how fast revenue will grow, spending massive amounts of money now to quickly build data centers could be “ruinous” if estimates are off even slightly, Amodei warned.

In November, Anthropic said it will spend $50 billion on AI infrastructure in the U.S., starting with data centers in Texas and New York.

Meanwhile, the top hyperscalers stunned Wall Street in recent weeks with plans to boost capital expenditures by much more than expected.

For example, Amazon plans to spend $200 billion this year alone, while Alphabet projected up to $185 billion, and Meta sees capex as high as $135 billion.

To illustrate his point about the timing of returns from AI investments, Amodei highlighted the potential for medical breakthroughs, which would drive enormous economic value.

There’s the question of how much of the gains pharmaceuticals companies receive versus AI companies. The research, manufacturing and regulatory processes also take time. Amodei noted that after the first COVID-19 vaccines were developed, it took about a year and a half to achieve widespread distribution.

When it comes to buying data centers, he looks at Anthropic’s 10-fold growth in revenue each year with 2026 tracking around $10 billion. At the same time, building and reserving a data center takes one to two years. By then, revenue could top $1 trillion if it follows its current trajectory, allowing the company in theory to commit a similar amount to data centers.

“If my revenue is not $1 trillion, if it’s even $800 billion, there’s no force on Earth, there’s no hedge on Earth that could stop me from going bankrupt if I buy that much compute,” Amodei said. “Even though a part of my brain wonders if it’s going to keep growing 10x, I can’t buy $1 trillion a year of compute in 2027. If I’m just off by a year in that rate of growth, or if the growth rate is 5x a year instead of 10x a year, then you go bankrupt.”

As a result, he instead accepts the risk that the company may not be able to meet all the demand for AI, acknowledging that Anthropic spends less than some of its competitors.

But without naming names, Amodei criticized rivals for “YOLOing” on spending, failing to fully comprehend the risks, and “just doing stuff because it sounds cool.”

He also pointed out that Anthropic’s AI is geared toward enterprise customers rather than fickle consumers, allowing them to rely more on revenue. Overall, Anthropic’s spending on computing capacity is still substantial.

“We’re buying an amount that’s comparable to what the biggest players in the game are buying,” Amodei said. But if you’re asking me, ‘Why haven’t we signed $10 trillion of compute starting in mid-2027?’ First of all, it can’t be produced. There isn’t that much in the world. But second, what if the country of geniuses comes, but it comes in mid-2028 instead of mid-2027? You go bankrupt.”

This story was originally featured on Fortune.com