OpenAI's compute bill hits $750B — and the first shovel lands in Georgia

OpenAI raised its 2030 infrastructure forecast to $750 billion and, for the first time, will design and build a data center itself: a $20B campus near Savannah called Project Camellia.

✓ Verified Source Independent reporting (TechCrunch, Bloomberg, WSJ, AIbase) ⚑ AI infrastructure

The 60-second version

OpenAI raised its 2030 compute-spending forecast ~25% to $750B and launched its first self-built data center — a $20B, 1,400-acre, 3.2 GW campus in Georgia called Project Camellia.

Key points

  • $750B projection is up from ~$600B; the trend keeps rising after Altman's $1.4T musing and Friar's walk-back.
  • Project Camellia in Effingham County, GA is the first data center OpenAI designs and builds itself — tenant to landlord.
  • ≥3.2 GW from Georgia Power (2028–2032); full-scale cost could top $30B; OpenAI is seeking financing partners.
  • OpenAI covers all infrastructure + power cost and will shed up to 1 GW at peak, but >half the new grid supply is gas.
  • The spend is a projection: the CFO has privately doubted OpenAI can pay, after a ~$38.5B loss in 2025.

Verdict. A genuine strategic shift — OpenAI is building, not just renting — but the financing is a bet against its own revenue, and the power is mostly gas.

The number$750 billion by 2030

On 22 July 2026, OpenAI told investors it now plans to spend roughly $750 billion on computing infrastructure through 2030 — about a 25% jump from the ~$600 billion it had projected earlier this year. The revision caps a confusing few months: CEO Sam Altman publicly floated a $1.4 trillion figure, CFO Sarah Friar walked it back to ~$600 billion to reassure investors, and now the number has climbed again.

$750Bcompute spend through 2030
~25%increase over the earlier ~$600B
$20Bcommitted to the first self-built campus
3.2 GWpower for Project Camellia

The first shovelProject Camellia — from tenant to landlord

The headline shift isn't the dollar figure — it's the model. Until now OpenAI has mostly rented capacity from Microsoft, Oracle and Amazon. Its first project designed and built in-house is Project Camellia, a data-center campus in Effingham County, Georgia, about 45 minutes from Savannah. OpenAI is moving from tenant to landlord.

To run construction, OpenAI hired Brent Mayo, who previously led the build of xAI's Colossus supercomputer campus in Memphis — a signal that OpenAI wants first GPUs online quickly.

The dealLand, power, and the grid

LocationEffingham County, GA (~45 min from Savannah)
Land1,400 acres, northwest of Savannah
Power≥3.2 GW from Georgia Power, phased 2028–2032
Demand responseCut up to 1 GW at peak (25-year agreement)
Community pledge$80M in benefits over the project's life
Tax break15-year, 50% property-tax abatement (Effingham County)

OpenAI says it will pay the full cost of the new infrastructure and the electricity to run it — and under Georgia Public Service Commission rules, those costs can't be passed on to existing ratepayers. On paper, that makes it a careful grid citizen.

The catchGas, and whether the bill gets paid

To feed data centers at this scale, Georgia Power plans to add nearly 9.9 GW of new generation — and more than half (~5.8 GW) is slated to be natural gas, with solar and storage covering the rest. Neither side has disclosed a concrete clean-energy transition plan, raising real concern that the compute build-out deepens fossil-fuel dependence.

Why it mattersThe arms race went physical

The $750B still rides on partner deals — 6 GW from Oracle, a $138B eight-year AWS arrangement, $250B in incremental Azure spend — but Project Camellia shows the frontier labs now compete on megawatts, land and transformers, not just models. When one company's data-center plan reshapes a regional power grid and revives gas plants, AI becomes an infrastructure-and-energy story.

OpenAI is moving from tenant to landlord — betting three-quarters of a trillion dollars that compute is destiny.