Anthropic CEO Dario Amodei told podcaster Dwarkesh Patel this week that a single year of miscalculated revenue could destroy any major AI company, including his own, even as Anthropic's annualized revenue hit $14 billion in early 2026.
The math is simple and brutal. Data centers take one to two years to build. Revenue projections compound at rates nobody has ever sustained. If Anthropic's roughly 10x annual growth rate drops to 5x, or arrives 12 months late, the company is finished. "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 on the Dwarkesh Podcast.
The spending gap nobody talks about
That $800 billion figure sounds absurd until you trace Anthropic's actual trajectory: zero to $100 million in 2023, $1 billion in 2024, $9 to $10 billion in 2025. Amodei said the company added "another few billion" in January 2026 alone. If that 10x curve somehow holds, $1 trillion in annual revenue by late 2027 is the logical endpoint.
Amodei clearly does not believe the curve will hold, which is the interesting part. He accused unnamed competitors of "YOLOing" on infrastructure spending and "just doing stuff because it sounds cool." Nobody in the room needed a decoder ring. OpenAI has signed compute partnerships totaling over 30 gigawatts with Nvidia, Broadcom, Oracle, and AMD. Anthropic is planning roughly 10 gigawatts.
According to Fortune's reporting from late 2025, leaked investor projections show Anthropic budgeting $78 billion on compute through 2028, against OpenAI's $235 billion. Anthropic claims it will generate 2.1 times more revenue per dollar of compute than OpenAI over that period. Whether that reflects genuine efficiency or a convenient excuse for spending less is a question Amodei does not really answer.
So when does the money actually come?
Here is where Amodei's story gets complicated, maybe deliberately so. He remains confident that AI models matching Nobel Prize winners could exist within one to two years. "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 told Patel. But he quickly separated the technical milestone from the financial one.
"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. I 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."
Patel pushed back, pointing out that if Amodei truly believes genius-level AI is imminent, Anthropic should be spending more aggressively on compute. Amodei's response was a version of: I believe it, but not enough to bet the company. "If the country of geniuses comes, but it comes in mid-2028 instead of mid-2027? You go bankrupt."
The profitability mirage
Investor documents from late 2025, previously reported by The Information, had Anthropic projecting breakeven by 2028 and potential revenue of $70 billion that year. Amodei complicated that picture considerably on the podcast. He pushed back against the idea that AI labs follow a simple invest-then-profit arc.
"We could be profitable in 2026 if the revenue grows fast enough," he said. "And then if we overestimate or underestimate the next year, that could swing wildly." In other words, profitability is not a destination. It is a moving target driven by how accurately you guess demand two years from now. If you guess right, you are profitable. If you guess wrong by 20%, you might be dead.
Amodei described a model where roughly 50% of compute goes to research, gross margins run above 50%, and correct demand prediction yields profit. "That's the profitable business model that I think is kind of there, but obscured by these building ahead and prediction errors."
What he said about software engineers
Amodei also revisited his prediction from mid-2025 that AI would write 90% of code within months. He said it happened, at least inside Anthropic and among many of their customers. But then he did something unusual for a CEO making capability claims: he walked it back.
"People thought I was saying that we won't need 90% of the software engineers. Those things are worlds apart." He laid out a spectrum, from 90% of code written by models (already happening) to 90% of end-to-end software engineering tasks done by models (soon) to 90% less demand for SWEs (further out, but coming). The distinction matters more than the headline number.
Patel asked a sharper question: if AI tools are this capable, where is the renaissance of new software? Where are the products that would not exist otherwise? Amodei conceded the point but argued adoption is moving faster than any previous technology, just not infinitely fast. Enterprise procurement cycles, security reviews, change management: all real friction, even for products that clearly work.
Anthropic announced its Series G funding this week at a $380 billion valuation, having raised $30 billion led by GIC and Coatue. The company now has over 300,000 business customers and roughly 4,000 employees.
The full podcast runs over two hours and covers scaling laws, China competition, and whether regulation will slow AI's economic impact. Anthropic's next model release has not been announced.
