OpenAI’s president Greg Brockman dropped a bombshell in court this week. The company plans to spend $50 billion on computing power in 2026 alone. That’s up from $30 million in 2017. A 1,666-fold jump in nine years.
The disclosure came during testimony in OpenAI’s legal clash with Elon Musk. Brockman laid bare the scale of the AI lab’s hunger for chips and servers. Compute costs now hit tens of billions annually, he said. And the trajectory points higher. OpenAI eyes $600 billion total through 2030, according to sources familiar with the plans reported by Reuters.
But here’s the catch. OpenAI isn’t bankrolling this from profits. The firm missed revenue and user targets, as detailed by The Wall Street Journal. Much of the cash flows from big-tech partners through conditional deals. Think rebates disguised as investments.
In February, Amazon, Nvidia, and SoftBank touted a $110 billion package. At least $80 billion hinges on OpenAI renting their gear. Amazon’s $50 billion slice demands two gigawatts of Trainium accelerators plus GPT models on AWS. Nvidia’s $30 billion ties to five gigawatts of training and inference capacity—potentially costing $300 billion overall. These aren’t gifts. They’re prepayments for capacity OpenAI craves, as unpacked by The Register.
Stargate: The $500 Billion Infrastructure Push Accelerates, Then Shifts
That spending frenzy powers Stargate. OpenAI’s massive buildout for data centers, chips, cloud, and power. Launched in January 2025 with Oracle and SoftBank, it targets 10 gigawatts of U.S. AI compute by 2029. OpenAI smashed that early. It secured the capacity in just over a year, adding 3 gigawatts in the last 90 days alone.
First site? Abilene, Texas. More coming in Texas, New Mexico, Wisconsin, Michigan. Nvidia signed a letter of intent for at least 10 gigawatts on its Vera Rubin platform, with up to $100 billion phased in. New Mexico even built an on-site gas plant for one facility—its grid couldn’t keep up, despite 38% wind and 11% solar, per Voronoiapp.
SoftBank and OpenAI just pumped $1 billion into SB Energy for a 1.2 gigawatt site in Milam County, Texas. “The result is a fast, reliable way to scale compute through large, highly optimized AI data centers,” Brockman said in the announcement via Investing.com.
Oracle locked a five-year, $300 billion deal to supply Stargate compute. The stock jumped then—now investors fret over circular risks, with OpenAI as both customer and backer, noted by The Next Web. AWS chipped in $50 billion, including a $138 billion lease for two gigawatts of Trainium chips.
Yet momentum falters in spots. OpenAI pulled from a Norway data center deal—Microsoft scooped the 230-megawatt capacity at Nscale’s Stargate Norway site, per MSN. Halted UK plans. No expansion at Abilene flagship. Senior Stargate execs jumped to Meta. A strategic reset, as The New York Times described, shifting to flexible compute pacts over rigid joint ventures.
Scaling back some projects. Pausing Texas expansions. OpenAI adapts to surging demand, per reports from Intellectia.ai and its own updates.
Power Crunch and Global Ripples Test the Model
Stargate demands dwarf norms. Nine-point-six gigawatts at one site matches New York City’s peak load. Energy bottlenecks force gas plants. Compute scarcity turns it into a tradable asset, BlackRock’s Larry Fink mused recently.
OpenAI’s burn rate draws scrutiny. It projects $121 billion annual compute costs by 2028, with $85 billion losses—$1.69 spent per dollar earned. CFO Sarah Fryer warned contracts could strain if revenue lags, via social media buzz echoing filings.
Europe watches warily. Its €20 billion AI gigafactory plan pales against Stargate. OpenAI’s UK pullback stings. Hyperscalers earmark $180 billion each for 2026 infra, leaving the EU scrambling, as Techzine reports.
Anthropic mirrors the frenzy, committing $200 billion to Google’s cloud and chips over years, per The Information. The arms race intensifies.
OpenAI races toward AGI—or superintelligence, goalposts shifting. Partners foot much of the bill. But viability hinges on revenue ramps and execution. Stargate sites activate soon. Nvidia systems hit in late 2026. Bets are on. Tokens burn cheap now. Prices will climb.
WebProNews is an iEntry Publication