Meta forces AI regroup

Meta has begun ordering selected engineers to move into a new Applied Artificial Intelligence Engineering unit as it reshapes the company around artificial intelligence work. (reuters.com) Reuters reported on April 9 that the transfers were no longer voluntary for some engineers, according to an internal memo by Maher Saba, the Reality Labs vice president running the new group. The memo said Meta started notifying selected staff that week. (reuters.com) The unit was set up in March and reports to Chief Technology Officer Andrew Bosworth. Earlier reporting said the group was built with an ultra-flat structure, with as many as 50 employees per manager. (wsj.com) The job of “applied” artificial intelligence is not basic research. It is the engineering layer that turns models into tools people inside Meta can actually use, including coding agents and systems that automate routine software work. (reuters.com) That push sits on top of a much larger infrastructure buildout. Meta said in January that 2026 capital spending would reach $115 billion to $135 billion, with much of that going to servers, data centers and other computing capacity for artificial intelligence. (cnbc.com) Meta has also been moving money and people away from other parts of the company. Bloomberg reported in late March that Meta cut several hundred jobs across sales, recruiting and Reality Labs, and multiple outlets put the total near 700 starting March 25. (bloomberg.com) (businessinsider.com) At the same time, Meta is recruiting infrastructure executives from OpenAI’s Stargate project, the effort to line up hundreds of billions of dollars in data-center capacity for artificial intelligence. Bloomberg reported on April 10 that Peter Hoeschele, Shamez Hemani and Anuj Saharan plan to join Meta. (bloomberg.com) Those hires point to the bottleneck Meta is trying to solve. Training and running advanced artificial intelligence systems requires chips, electricity, buildings and networking gear at a scale that now matters as much as model design. (about.fb.com) (cnbc.com) Meta has been making that case publicly for months. In March, the company said four new generations of its Meta Training and Inference Accelerator chips were planned over the next two years as part of its artificial intelligence infrastructure strategy. (about.fb.com) Meta declined to comment to Reuters on the forced transfers, and Bloomberg said Meta declined to comment on the Stargate hires. The picture from both reports is the same: Meta is treating artificial intelligence staffing and computing capacity as the same reorganization, not two separate ones. (reuters.com) (bloomberg.com)