Meta Pushes Ahead With Custom “Iris” AI Chip to Cut Nvidia Dependence

Meta Platforms is preparing to begin manufacturing its own custom data-center AI chip, internally codenamed “Iris,” starting in September, as part of a broader four-generation roadmap for its in-house silicon program. The move is aimed at reducing the company’s reliance on external GPU suppliers as it races to expand computing capacity for its artificial intelligence workloads.

Internal planning documents suggest the company intends to scale its overall computing infrastructure to roughly 14 gigawatts by 2027, up from a 2026 target of seven gigawatts, reflecting the scale of investment major technology firms are pouring into AI infrastructure. The Iris chip itself has reportedly been developed in partnership with Broadcom and manufactured by TSMC, and is said to have completed testing in only a matter of weeks without major issues surfacing.

The chip is designed specifically to support Meta’s own training and inference needs across its platforms, including recommendation systems, content moderation tools, and generative AI features embedded across Facebook and Instagram. By building tailored silicon rather than relying solely on third-party GPUs from Nvidia and AMD, Meta hopes to gain more control over both cost and supply as demand for AI compute keeps climbing industry-wide.

The announcement lands amid a broader wave of custom silicon investment across the technology sector, as major platform companies look to insulate themselves from GPU shortages and pricing pressure. Industry watchers say the success or failure of programs like Iris could shape how dependent the next generation of AI infrastructure remains on a handful of chipmakers.