OpenAI plans massive compute pivot from Microsoft to SoftBank

For years, Microsoft has been OpenAI’s undisputed computing powerhouse. But over the next five years, that dynamic is set for a massive shakeup.

According to a report from The Information, OpenAI forecasts that by 2030, about three-quarters of its data center capacity will come from the Stargate project. And backing that gargantuan infrastructure push is none other than SoftBank, one of OpenAI’s newest heavyweight backers.

SoftBank steps in as Microsoft's share shrinks

Right now, Microsoft is OpenAI's biggest shareholder and supplies the vast majority of its computing power. Shifting 75% of capacity to Stargate represents a major pivot away from Redmond.

This won't happen overnight. OpenAI still plans to increase its spending on Microsoft-owned data centers over the next few years to keep up with immediate demand.

Yet the long-term blueprint clearly favors SoftBank, which is expected to heavily finance the Stargate buildout.

A $20B burn rate and the rise of inference

Fueling this computing reshuffle is an eye-watering surge in OpenAI’s expenses. The report notes that OpenAI projects burning through $20 billion in cash in 2027 alone, up sharply from the estimated $5 billion it burned in 2024.

Equally telling is where that money will actually go. By 2030, OpenAI expects its costs for running AI models—inference—to outpace the cost of training them.

Basically, keeping the models responding to millions of queries every second will soon be far pricier than creating them in the first place.

Why it matters

This shift shows OpenAI aggressively working to reduce its reliance on Microsoft. Microsoft rode its exclusive partnership to huge cloud wins, but OpenAI clearly doesn't want all its eggs in one basket. Bringing SoftBank and its deep pockets into the fold gives Sam Altman much-needed leverage and independence.

For developers and everyday users, the projection that inference will eclipse training costs by 2030 confirms that running AI at global scale remains brutally expensive. In our view, that massive $20 billion burn rate in 2027 means OpenAI will have to push monetization harder than ever, through either higher subscription prices or tighter API caps.

Strategically, diversification is the smart move. But reliance on megaprojects like Stargate carries immense execution risk. Betting billions on SoftBank-backed infrastructure while burning tens of billions of dollars looks like a high-wire act where OpenAI simply cannot afford a market downturn.