Meta to Start AI Chip Production to Boost Compute Capacity

Meta Platforms Inc. is set to initiate the production of its proprietary artificial intelligence chip, code-named "Iris," in September 2026. According to an internal memo reviewed by Reuters, the move is a central component of the social media giant's effort to expand its computing infrastructure to 14 gigawatts by next year, effectively doubling its current capacity.

The "Iris" chip is part of the Meta Training and Inference Accelerators (MTIA) project, a multi-generational effort to design custom silicon in-house. By transitioning to custom-built hardware, Meta aims to decrease its heavy reliance on third-party graphics processing units (GPUs) supplied by companies such as Nvidia and Advanced Micro Devices (AMD).

In-House Silicon Strategy

The development of "Iris" marks a significant milestone for Meta’s hardware team. While previous attempts at in-house silicon faced performance challenges, recent internal testing of the "Iris" chip reportedly concluded in just six weeks with no major issues, signaling a shift in momentum for the project.

Meta is collaborating with Broadcom for chip design and Taiwan Semiconductor Manufacturing Co. (TSMC) for fabrication. This partnership allows Meta to tailor its silicon specifically for the recommendation engines that power Facebook and Instagram, optimizing both power efficiency and processing speeds for its massive internal workloads.

Aggressive Infrastructure Scaling

The shift toward proprietary hardware occurs against a backdrop of historic capital investment. Meta plans to deploy seven gigawatts of computing infrastructure in 2026, with an internal goal to double that figure in 2027.

This expansion reflects the company's broader pivot toward AI leadership. Alongside its hardware initiatives, reports indicate that Meta is exploring a cloud business model to monetize its excess computing capacity, allowing third-party developers to access its AI infrastructure. This dual-track strategy—designing custom chips to lower costs while building a cloud service to generate revenue—is designed to offset the high financial burden of its AI investments, which are projected to reach as high as $145 billion in 2026.

Strategic Implications

For the tech industry, Meta's move toward vertical integration illustrates the increasing necessity for hyperscalers to control their own semiconductor supply chains. By accelerating the cadence of its chip releases—targeting a new iteration roughly every six months—Meta aims to remain at the cutting edge of AI development while insulating itself from the supply chain volatility and rising costs associated with third-party GPU markets.

Key Facts at a Glance

• New Chip: Internal "Iris" data center chip enters production in September 2026.

• Infrastructure Goal: Meta targets 14 gigawatts of total computing power by 2027.

• Projected Spending: The company expects 2026 capital expenditures on AI infrastructure to reach up to $145 billion.

• Partnerships: Meta is partnering with Broadcom for chip design and TSMC for manufacturing.

• Strategic Pivot: The initiative seeks to reduce reliance on external suppliers like Nvidia and AMD.

FAQ Section

1. What is the "Iris" chip?

"Iris" is the code name for Meta’s latest in-house data center chip, part of its Meta Training and Inference Accelerators (MTIA) project, designed to handle AI training and inference tasks.

2. Why is Meta moving to in-house chip production?

The company aims to improve power efficiency, lower costs, and gain independence from external GPU suppliers, whose products have become increasingly expensive and difficult to source at scale.

3. Does this mean Meta will stop buying Nvidia chips?

No. Meta will continue to purchase high-end GPUs from Nvidia and AMD. The in-house chips are intended to augment—not entirely replace—these external supplies for specific internal workloads like social media feed ranking.

Source: Reuters, Investing.com, AlphaMatch.