The Three People Who Knew How to Spend 100 Billion on AI Infrastructure Just Left OpenAI for Meta

The three people who knew best how to spend $100 billion on artificial intelligence infrastructure have left OpenAI in the same week. Now they work for Mark Zuckerberg.

Peter Hoeschele, Shamez Hemani, and Anuj Saharan are joining Meta's compute infrastructure division, Bloomberg reported Friday. All three were central to OpenAI's Stargate program, the company's ambitious plan to build hundreds of billions of dollars' worth of AI data centers in the United States. Hoeschele had been with OpenAI since 2022, AI Daily reported, specializing in strategy and operations for AI and digital infrastructure. Hemani focused on engineering execution. Saharan handled supply chain logistics. The three departed OpenAI on April 9, The Information reported Thursday, citing people with direct knowledge of the matter.

The departure matters beyond the usual executive attrition narrative. Hoeschele, Hemani, and Saharan did not just hold three separate jobs. Together they held the complete execution chain for a $100 billion infrastructure program: the strategy, the engineering, and the supply chain. OpenAI lost all three on the same day, to the same destination.

The irony is precise. OpenAI built Stargate as a "build-to-own" strategy, designing and operating its own data centers rather than renting cloud capacity from Microsoft or Amazon. That approach required exactly the expertise these three spent years developing. Now that expertise is Meta's. OpenAI, meanwhile, has pivoted from building its own data centers under the Stargate model to renting compute capacity from cloud providers, according to AI Daily, a posture better suited to its revised commercialization timeline ahead of a potential IPO.

Meta has committed $115 billion to $135 billion in capital expenditures for 2026 and 2027, much of it targeting AI infrastructure. The company has no existing cloud business to monetize that spend the way Microsoft or Amazon can. Buying the team that designed OpenAI's infrastructure playbook gives Meta something no press release can announce: the institutional knowledge of how to actually execute a buildout at that scale.

This is the second time in recent months that a major AI lab has shed infrastructure talent to a competitor. The question is whether the knowledge transfer stops at hiring or whether it accelerates a broader realignment of who builds what in the AI race.

For infrastructure investors and AI labs watching the Stargate program, the personnel shift is more significant than it might appear. The partners behind Stargate including SoftBank, Oracle, and the UAE-backed MGX committed up to $500 billion to the effort. The value of that commitment depends partly on who is executing it. That team is now at Meta.

OpenAI declined to comment.