Meta poaching infra talent
Meta has hired multiple senior infrastructure leaders from OpenAI as it pours resources into a Superintelligence Labs buildout, a move that signals the company is prioritizing compute operations and cost-aware systems engineering. The reporting frames this as part of a broader push—Meta is increasing AI spending and looking to win the battle for people who can procure, schedule and operate large compute estates, not just model researchers. (Outlook Business)
Meta is not just hiring more artificial intelligence researchers. It has pulled in three OpenAI infrastructure leaders, including Peter Hoeschele from the Stargate project, to help build out Meta Superintelligence Labs. (outlookbusiness.com) That job is less about inventing a new model and more about running a giant industrial system: buying chips, finding power, placing workloads, and keeping expensive clusters busy instead of idle. (ft.com) Meta is spending at that industrial scale now. Outlook Business says Mark Zuckerberg is preparing for as much as $135 billion in capital expenditure in 2026 to expand the company’s artificial intelligence data center footprint. (outlookbusiness.com) Meta has been signaling the same thing in public. In February 2026, it announced a long-term artificial intelligence infrastructure agreement with Advanced Micro Devices and said the deal was part of a Meta Compute effort to “massively scale” infrastructure for personal superintelligence. (about.fb.com) The phrase “superintelligence lab” can sound like a research slogan, but the bottleneck is often plumbing. A frontier model needs thousands of graphics processors, steady electricity, networking that moves data without traffic jams, and software that decides which job runs on which machine. (ft.com) That is why infrastructure talent has become poachable star talent. A company that knows how to keep a large compute estate full and efficient can get more model training out of the same chips than a rival that simply buys hardware and hopes for the best. (ft.com) OpenAI has been building its own giant supply chain through Stargate. In January 2025, it said Stargate planned to invest $500 billion over four years in United States artificial intelligence infrastructure, with $100 billion to be deployed immediately. (openai.com) By September 2025, OpenAI said Stargate had nearly 7 gigawatts of planned capacity across Abilene, Texas, new sites in Texas and New Mexico, a Midwest location later identified as Wisconsin, and other projects with Oracle and CoreWeave. (openai.com) So when Meta hires people from that operation, it is hiring from one of the few teams that has already tried to assemble artificial intelligence infrastructure at national-grid scale. The value is not a paper or a benchmark; it is knowing which delays kill a build and which inefficiencies quietly burn millions of dollars. (openai.com) (ft.com) This also fits Meta’s wider recruiting pattern. Semafor reported in July 2025 that Meta’s superintelligence unit was already pulling in specialists from areas like perception, synthetic data, and reasoning, and Outlook later described hires from Apple, Google DeepMind, Anthropic, and OpenAI around the same push. (semafor.com) (outlookbusiness.com) Meta’s own product language now ties those hires to a concrete buildout. Three days ago, it said Meta Superintelligence Labs had rebuilt its artificial intelligence stack over the last nine months and introduced Muse Spark as the first model in a new series. (about.fb.com) The simplest read is that the artificial intelligence race is splitting into two markets at once. One market pays for researchers who design better models, and the other pays for operators who can turn power, chips, buildings, and scheduling software into a machine that actually runs them. (ft.com) (about.fb.com)
