Meta starts Iris custom AI chip production in September; plans 14-gigawatt datacenter by 2027
Meta will begin manufacturing its custom AI chip, code-named Iris, in September 2026, according to an internal memo reviewed by Reuters on July 9. The chip, developed under Meta's MTIA (Meta Training and Inference Accelerators) program with design partner Broadcom and manufacturer TSMC, cleared bug testing in just six weeks with no major issues. Iris is optimized for Meta's recommendation ranking and generative AI workloads on Facebook and Instagram, designed to reduce dependence on Nvidia and AMD GPUs rather than replace them entirely.
Meta's infrastructure scaling plan is massive: 7 gigawatts of computing capacity by end of 2026, doubling to 14 gigawatts by 2027. The company projects AI infrastructure spending of up to $145 billion in 2026 alone. To secure components, Meta has signed multi-year supply agreements with Samsung Electronics (memory chips), Sandisk (flash storage), and Sumitomo Electric (fiber-optic equipment), signaling aggressive vertical integration across the supply chain.
The MTIA program targets a six-month release cadence, roughly double the industry standard. Meta publicly unveiled the roadmap in March 2026 (MTIA 300, 400, 450, 500 variants), with MTIA 300 already in production for ranking and recommendations. This pace reflects Meta's conviction that custom silicon will be central to cost management as AI compute demands scale. Broadcom, which designs Iris, also designs Google's TPU and OpenAI's Jalapeno chip—a pattern showing consolidation in custom-chip design consulting.
For architects: Iris production signals that hyperscaler vertical integration in AI hardware is moving from R&D into manufacturing. The September timeline means field deployments of Meta-optimized chips on production workloads will begin within weeks, creating a real-world test of custom silicon ROI. The 14-gigawatt target by 2027 is roughly 17 times global renewable electricity capacity and underscores both the scale of the AI build-out and the power-cost pressure driving custom hardware investment.