Anthropic CEO Dario Amodei recently said bluntly: If AI companies want to survive, their annual revenue must reach "hundreds of billions of dollars" - otherwise, they may be swallowed up by this computing power competition at any time.This company's financial growth curve is a miracle in business history. By the end of 2024, Anthropic's annual revenue will be approximately US$1 billion; by the end of 2025, it will be close to US$10 billion; by March 2026, it will soar to US$30 billion; by May, annual recurring revenue has reached US$47 billion. In just over a year, revenue has increased approximately 47 times.
But the other side of the dazzling numbers is even more crazy capital expenditures. Anthropic's core appeal for this round of financing is to expand its computing infrastructure. It reached an agreement with Amazon to receive an investment of up to 25 billion U.S. dollars and locked in 5 gigawatts (GW) of computing power. It also cooperated with Google and Broadcom to launch another 5 GW of computing power next year. Google plans to invest up to an additional 40 billion U.S. dollars.
This is the underlying reason why Amodei issued a warning of "hundreds of billions in revenue":The costs of training cutting-edge large models—computing power, infrastructure, talent, and R&D—have converged into a bottomless financial black hole. Even with annualized revenue of US$47 billion, Anthropic still needs to continue to raise tens of billions of US dollars to maintain operations.
The primary purpose of the pre-IPO financing that the company is promoting is to satisfy this hunger for computing power. What is the concept of 5 gigawatts? The total installed capacity is equivalent to several large nuclear power units, all of which are invested in model training and deployment.
The logic of this strategy is very clear: the performance of large models is positively related to the scale of computing power - the stronger the computing power, the better the model, the more customers, the higher the income, which in turn feeds back the computing power, forming a self-reinforcing growth flywheel.
But the other side of the flywheel is a cliff: once growth slows down or competition loses its footing, huge fixed costs will instantly turn from a moat into a noose, dragging cash flow into a bottomless abyss.For Anthropic, the revenue myth is certainly shocking, but the real test of survival has just begun.