Meta's custom silicon effort appears to be moving from roadmap to production planning, which matters because hyperscalers are trying to lower their dependence on Nvidia GPUs while AI workloads keep expanding. Reuters reported on July 9 that Meta plans to put an in-house MTIA AI chip, code-named Iris in some coverage, into production in September and aims to roughly double data center computing capacity to 14 gigawatts by 2027. The report, based on a memo and people familiar with the matter, fits Meta's broader push to pair its AI model launches with more vertically controlled infrastructure. For practitioners, the signal is that custom accelerators are becoming a competitive constraint, not just a cost project: model availability, inference pricing, deployment regions, and vendor risk may increasingly depend on each lab's chip and power strategy.
Meta to put its own AI chip into production in September, aiming to double computing capacity