OpenAI Is About to Spend an Entire Country’s GDP

Here is a number that reframes the entire AI race: $750 billion. That is what OpenAI now says it will spend on infrastructure through 2030 — the equivalent of a mid-sized nation’s yearly economic output, committed by a single company to build the machines that run its models. And the electricity bill behind that spending is about to test the limits of the American power grid.

A $750 Billion Bet, 25% Bigger Than Planned

Project Camellia: the first $20 billion salvo

OpenAI announced on Wednesday that it will spend $750 billion on infrastructure through 2030, some 25% more than it estimated earlier this year, according to a Wall Street Journal report. To put that figure in perspective, OpenAI’s own framing is blunt: it is the equivalent of Sweden’s entire GDP, funneled into a single company’s compute buildout. The renewed blitz arrives even as its earlier Stargate data center project appeared to stall.

The opening move is a $20 billion data center campus in Georgia known as Project Camellia. The development spans 1,400 acres northwest of Savannah and will draw at least 3.2 gigawatts of power from Georgia Power, the region’s utility, with generating capacity expected to come online between 2028 and 2032. OpenAI says it will “pay the full cost of the infrastructure and electric-service costs” for the site, and it is receiving a 50% property tax abatement for 15 years from Effingham County, according to the Effingham Herald.

Speed appears to be the priority. OpenAI recently hired Brett Mayo to lead data center construction; Mayo previously oversaw xAI’s Colossus facility in Memphis, which was built in record time but drew a lawsuit from the NAACP and the Southern Environmental Law Center over dozens of allegedly unpermitted natural gas turbines. Georgia Power’s own filings show most of the new capacity feeding these campuses will come from natural gas — more than doubling the utility’s gas fleet — with the rest from grid-scale batteries and solar.

The Power Grid Can’t Keep Up

Data centers could devour one-fifth of U.S. electricity

OpenAI’s spending is one company’s slice of a far larger surge. A new report from BloombergNEF projects that data centers will use one-fifth of all electricity generated in the U.S. by 2035 — four times today’s share — as a wave of AI compute pushes capacity toward nearly 200 gigawatts over the next decade. Nearly half of that will go to training and inference, and by 2033 the U.S. will host 64% of the world’s AI chips by power demand.

The forecasts keep climbing. BloombergNEF’s 2035 estimate is 83% higher than what the same consultancy predicted in December. EPRI has more than doubled its 2024 estimate, and S&P’s forecast rose by more than a third between October and April. The revisions reflect what analysts describe as the fevered pace of data center development across the country.

The strain is concentrated. The PJM Interconnection, spanning Virginia to Illinois, is projected to send 34% of its electricity to data centers, while Texas’s ERCOT grid will devote 22% of its generating capacity. PJM has already struggled to cope, pausing new generation-connection applications for four years — and wholesale power prices on the grid have climbed 76% over the past year. Globally, if aggressive AI adoption holds, data centers will create 1,935 terawatt-hours of new electricity demand by 2033, nearly as much as India uses in a year.

What It Means for the AI Business

Capital intensity is becoming the new moat

For years, the defensible advantage in AI was assumed to be talent and data. The $750 billion figure suggests a third moat is hardening fast: the sheer ability to finance and physically build compute at national scale. A startup can rent GPUs, but it cannot conjure 3.2 gigawatts of firm power or negotiate a 15-year tax abatement with a county. That gap increasingly separates the handful of players who can operate at the frontier from everyone else.

It also reshapes where value accrues. Google spent much of the past week justifying its own massive AI outlays by pointing to a booming cloud business, and firms like Anthropic and Blackstone have argued the next trillion-dollar opportunity lies in AI implementation, not just model training. In other words, the buildout is a bet that demand for inference — the day-to-day running of AI in real products — will be large and durable enough to justify infrastructure priced like a sovereign economy.

The risk is symmetry: if adoption or pricing disappoints, hundreds of billions in gas turbines, transformers, and concrete do not un-build themselves. For business leaders watching from outside the hyperscaler club, the practical takeaway is less about matching the spend and more about positioning around it — building products, workflows, and services that ride on top of this compute rather than competing to own it.

Related

• Qualcomm Just Spent $4 Billion to Break Nvidia’s Grip
• Together AI’s $800M Bet on Cheaper Inference
• Pfizer Just Handed This AI Startup Its Secret Recipe
• Google Scrapped a Finished AI Model — Here’s Why

Sources

1. TechCrunch — OpenAI’s AI spending spree has ballooned to $750B (July 22, 2026)
2. TechCrunch — Data centers expected to use 4x more electricity by 2035 (July 21, 2026)
3. The Wall Street Journal — OpenAI’s Planned Cloud Spending Hits $750 Billion