OpenAI’s $600B AI Infrastructure Bet: A Race Fueled by Compute Power, Not Innovation

AI’s Infrastructure Boom Hits a Wall

Despite pledging hundreds of billions of dollars to build massive data centers, the AI industry is struggling to meet its ambitious goals. Recent reporting from Bloomberg and Ed Zitron reveals that nearly half of the data centers planned for the United States are either delayed or canceled outright. Shortages of critical electric components and skyrocketing costs have slowed the infrastructure expansion to a crawl, leaving tech leaders frustrated.

OpenAI’s Compute Ambitions: A Numbers Game

In this high-stakes environment, every AI player is making bold claims to sustain the hype. OpenAI, however, has elevated bragging to an art form. A Bloomberg-obtained memo outlines the company’s plan to achieve 30 gigawatts of computing capacity by 2030—enough to power over 22 million US households. By comparison, rival Anthropic aims for just 7 to 8 gigawatts by the end of 2027.

To put these figures into context, OpenAI had only 1.9 gigawatts of computing capacity in 2025, while Anthropic had 1.4 gigawatts. The memo declares,

“Even at the high end of that range, our ramp is materially ahead and widening.”

OpenAI further claims it is outpacing Anthropic by “rapidly and consistently” adding computing capacity. The memo states,

“That gap matters because compute is now a product constraint.”

Anthropic Fires Back at OpenAI’s Claims

The timing of OpenAI’s memo is significant. It was released shortly after Anthropic unveiled its latest AI model, Claude Mythos, which its staffers argue is too powerful—and thus too risky for cybersecurity—to release in full. In response, Anthropic pushed back on OpenAI’s assertions, highlighting a recent deal with Broadcom and Google as evidence of its “disciplined approach to scaling infrastructure.”