Meta Fires Up Its Own AI Chip Factory, Challenging Nvidia's Grip

Meta is taking control of its own AI hardware destiny by starting mass production of its Iris chip, which will accelerate AI workloads within its data centers. The company's internal memo, reviewed by Reuters, confirms that Iris will begin shipping in September 2026.

The MTIA program, from which the Iris chip originates, has been shipping inference-focused chips since 2023, but this represents a significant jump to training-class products, where Nvidia's GPUs have held sway until now. The four-generation roadmap suggests Meta is committed to reducing its dependence on external suppliers for workloads where custom silicon can be more efficient.

The partnership with Broadcom on Iris design and TSMC on fabrication is notable, as it puts Meta's custom silicon in direct competition with Nvidia's next-generation Rubin chips for leading-edge manufacturing capacity at the world's most important chip foundry. The 7-gigawatt target by the end of 2026 and doubling to 14 gigawatts in 2027 are tied directly to Iris and the rest of the MTIA lineup coming online.