Meta Doubles Down on Custom AI Silicon While Managing Semiconductor Manufacturing Risks

Meta is rolling out four new generations of its homegrown AI chips over two years, dramatically escalating its competition with Nvidia and AMD while wagering billions on custom silicon development. The company announced the Meta Training and Inference Accelerator (MTIA) roadmap on March 11, 2026, committing to deploy MTIA 300, 400, 450, and 500 chips across its data centers starting immediately, with new generations arriving every six months—a pace that defies industry norms. Meta claims the MTIA 400 is the first iteration to deliver both cost savings and performance competitive with commercial alternatives, specifically targeting GenAI inference at the massive scale required to power Facebook, Instagram, and Threads infrastructure.