Meta's new AI chips will begin production in September
TechCrunch
β’Thu, 09 Jul 2026 17:17:37 +0000
π° What Happened
Meta announced its latest MTIA (Meta Training and Inference Accelerator) AI chips will begin production in September 2026, aiming to reduce dependence on GPU suppliers like Nvidia and AMD. At least one chip passed testing in about six weeks, and Meta is working with Broadcom on design and TSMC on manufacturing.
π The Backstory
Meta has been developing its own AI chips since 2023 as part of a strategy to reduce the massive costs of AI compute, with the company expecting capital expenditures between $125-$145 billion in 2026. The MTIA program uses a modular chiplet approach to keep pace with rapidly evolving AI workloads. Meta previously detailed four new chips under this program in March 2026.
π― Why It Matters
If successful, Meta's custom silicon could reduce the AI industry's near-total dependence on Nvidia GPUs and reshape the AI hardware market. It also represents a significant step toward vertical integration for major tech companies seeking to control their AI infrastructure costs.
Meta announced its latest MTIA (Meta Training and Inference Accelerator) AI chips will begin production in September 2026, aiming to reduce dependence on GPU suppliers like Nvidia and AMD. At least one chip passed testing in about six weeks, and Meta is working with Broadcom on design and TSMC on manufacturing.
Meta has been developing its own AI chips since 2023 as part of a strategy to reduce the massive costs of AI compute, with the company expecting capital expenditures between $125-$145 billion in 2026. The MTIA program uses a modular chiplet approach to keep pace with rapidly evolving AI workloads. Meta previously detailed four new chips under this program in March 2026.
If successful, Meta's custom silicon could reduce the AI industry's near-total dependence on Nvidia GPUs and reshape the AI hardware market. It also represents a significant step toward vertical integration for major tech companies seeking to control their AI infrastructure costs.