OpenAI Stargate staff move to Meta
Key engineers from OpenAI’s Stargate data‑centre project have left for Meta, highlighting the intensifying talent fight over large‑scale AI infrastructure work. The departures were reported this week and add to a pattern of staff moves that can reshape who builds hyperscale compute backbones (x.com).
Three people who helped OpenAI line up the giant buildings, chips, and power deals behind Stargate are now heading to Meta instead. Bloomberg reported on April 10 that Peter Hoeschele, Shamez Hemani, and Anuj Saharan are joining Meta after working on OpenAI’s data-center push. (bloomberg.com) These were not chatbot product jobs. They were infrastructure jobs, which means finding land, electricity, networking, servers, and construction partners so an artificial intelligence model has somewhere physical to run. (bloomberg.com) Stargate is OpenAI’s umbrella for the huge computing buildout it announced with SoftBank, Oracle, and MGX in January 2025. The plan said it intended to invest $500 billion over four years and start with $100 billion “immediately” in the United States. (openai.com, group.softbank) That kind of project is less like buying more laptops and more like building a private power station for a steel mill. A modern artificial intelligence data center needs scarce Nvidia graphics processing units, long-term electricity contracts, fiber connections, and local permits that can take months. (openai.com, datacenterdynamics.com) OpenAI’s first flagship Stargate site is in Abilene, Texas, where developer Crusoe said the campus would scale to about 1.2 gigawatts across multiple buildings. Data Center Dynamics reported last month that Oracle and OpenAI had dropped plans to expand that site further to 2 gigawatts. (datacenterdynamics.com) This week also brought another sign that OpenAI’s infrastructure plan is being reshaped. Bloomberg and CNBC reported that OpenAI paused its Stargate effort in the United Kingdom over energy costs and regulation. (bloomberg.com, cnbc.com) Meta is hiring these people because it is trying to build its own giant computing backbone for what Mark Zuckerberg calls Superintelligence Labs. Bloomberg said the new hires will work on the computing capacity behind that effort. (bloomberg.com) The fight here is over a bottleneck most users never see. The company that controls enough buildings, power, cooling, and chips can train bigger models faster and run them for more people at lower cost. (openai.com, bloomberg.com) That is why three infrastructure departures can matter as much as a flashy research hire. In artificial intelligence right now, the people who know how to assemble a one-gigawatt campus are closer to shipyard builders than software managers, and there are not many of them. (bloomberg.com, datacenterdynamics.com)
