Back in February, OpenAI put its computing budget through 2030 at around $600 billion. Less than six months later, that number has climbed to $750 billion. According to The Wall Street Journal, the company widened its spending plan because demand for its services is outpacing how fast data centers can be built.
The money is flowing through several channels at once. There are cloud contracts with Microsoft, Amazon Web Services and CoreWeave — back in February, OpenAI already tacked on an extra $100 billion over eight years to its original $38 billion AWS deal. There's a separate five-year, $300 billion contract with Oracle starting in 2027, requiring 4.5 gigawatts of computing capacity. And there's OpenAI's own build: the Project Camellia data center complex in Effingham County, Georgia, where the company is committing $20 billion and locking in 3.2 gigawatts of power from 2028 through 2032 — roughly what three nuclear plants produce.
For context, CEO Sam Altman floated a much bigger number earlier this year — $1.4 trillion in combined infrastructure commitments — and that figure spooked investors badly enough that the company had to walk the tone back. The current $750 billion looks more modest by comparison, but a 25% jump in just a few months still signals that OpenAI's appetite for compute isn't slowing down.
There's a flip side to all this. The gigawatts that AI data centers need are the same gigawatts that bitcoin miners compete for in Texas and other power-rich regions. The more aggressively AI companies lock up energy contracts, the harder and pricier electricity gets for other industries that depend on cheap power.
The swings in the forecast itself are telling too — from $600 billion to $1.4 trillion and back down to $750 billion within a year. That suggests even OpenAI can't pin down exactly how much compute it will need three or four years out. The AI market is moving faster than any budget plan can keep up with.
