In a recent podcast appearance, Dario Amodei, chief executive of Anthropic, offered a clear warning about the race to build massive AI infrastructure.
The core problem, he said, is simple to describe but hard to solve: nobody knows when powerful AI capabilities will turn into reliable revenue.
Amodei believes AI systems that perform at the level of Nobel Prize winners could arrive within a few years, or even sooner.
If that prediction holds, the logic behind huge spending on computing power seems obvious. More compute means faster progress, stronger models, and a better chance of leading the market.
Yet the business reality looks less certain.
Why Timing Is the Ultimate Risk for Anthropic and OpenAI
Even when AI produces major scientific breakthroughs, money does not appear overnight. Amodei used disease cures as an example. An AI system might discover a treatment quickly, but the path from discovery to profit remains slow.
Researchers must test the biology, companies must manufacture drugs, and regulators must approve them. Each step takes time. Revenue follows years after the original insight.
That gap between capability and income creates risk for companies investing billions in infrastructure today.
Anthropic’s own growth shows how fast the industry can move. The company’s revenue rose from zero to $100 million in 2023, reached $1 billion in 2024, and climbed to roughly $9–10 billion in 2025. Annualized revenue stands near $14 billion in early 2026. Those numbers suggest explosive demand for AI systems.
Still, Amodei warns against assuming that growth will continue at the same pace. Forecasts that look safe on paper can collapse if timelines shift even slightly.
He described a stark example. A company could commit to buying $1 trillion worth of compute capacity scheduled to come online in 2027. If revenue reaches only $800 billion instead of $1 trillion, the mismatch could trigger bankruptcy. No financial hedge can protect against spending far ahead of income. In capital-heavy industries, timing matters as much as innovation.
Being wrong by a single year could prove fatal.
Ambition, Economics, and the Battle for Gigawatts
Amodei suggested that some competitors underestimate this danger. Without naming firms directly, he said some companies appear to chase ambitious projects because they sound exciting rather than because the economics make sense. Many observers read this as a quiet criticism of OpenAI, which has announced aggressive infrastructure partnerships and expansion plans.
The scale of spending across the industry is hard to grasp. Anthropic plans to invest in at least ten gigawatts of compute capacity over the next few years. That level already ranks among the largest technology infrastructure efforts ever attempted.
Other players aim even higher. OpenAI has announced partnerships tied to more than 30 gigawatts of computing capacity through collaborations with chipmakers and cloud providers, including Nvidia, Broadcom, Oracle, and AMD.
Many details of those agreements remain unclear, which adds to uncertainty about how quickly that capacity will generate returns.
Why Timing Trumps Speed?
Amodei stressed that Anthropic is still buying enormous amounts of compute. The company does not reject large investments. Instead, it tries to match spending with realistic expectations about adoption and revenue timing.
There is also a physical limit to growth. According to Amodei, the world cannot produce unlimited computing infrastructure on demand. Supply chains, energy availability, and construction timelines constrain expansion. Even if companies wanted to sign deals worth trillions, the hardware and power simply do not exist yet.
His broader message reflects a tension at the center of the AI boom. Leaders believe transformative systems are close. Investors see historic opportunity. At the same time, the economic payoff may lag behind technical progress.
If AI creates what Amodei calls a “country of geniuses,” the breakthrough could reshape science and industry. But if that moment arrives later than expected, companies that spent too far ahead of reality may struggle to survive long enough to benefit from it.
In the AI race, speed matters. Timing matters more.
