Meta Platforms plans to begin manufacturing a new in-house artificial intelligence chip in September as it dramatically expands its AI computing infrastructure, according to an internal memo reviewed by Reuters.
The move marks a significant step in Meta’s plan to build more of its own AI hardware, reduce reliance on external chip suppliers and lower the enormous cost of training and deploying increasingly sophisticated artificial intelligence models across its platforms.
The custom chip, code-named “Iris,” forms part of Meta’s multi-year Meta Training and Inference Accelerator (MTIA) program, a four-generation family of processors designed specifically for the company’s AI workloads. The effort is intended to optimize the AI systems powering Facebook, Instagram, WhatsApp and Meta’s growing suite of generative AI products while giving the company greater control over one of the most critical components of the AI technology stack.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
According to the internal memo, testing of the Iris chip was completed in just six weeks, with engineers finding no major issues before production.
The unusually short validation period represents a notable breakthrough for Meta’s custom silicon initiative, which has faced repeated setbacks since it was launched more than five years ago.
A Shift Away From Dependence on Nvidia
As companies race to build powerful AI models, they are becoming less willing to depend entirely on external suppliers for the chips that power those systems.
Today, Nvidia dominates the market for AI accelerators, while Advanced Micro Devices has emerged as another major supplier.
However, demand for AI chips has consistently outpaced supply, leading to long waiting periods, rising prices and intense competition among cloud providers and technology companies.
Meta hopes to reduce those bottlenecks while tailoring hardware specifically to its own software and AI models by developing proprietary processors. The company is designing Iris in collaboration with Broadcom, while manufacturing will be handled by Taiwan Semiconductor Manufacturing Company (TSMC), the world’s largest contract chipmaker.
Rather than replacing Nvidia’s processors entirely, Iris is expected to complement the large numbers of graphics processing units (GPUs) Meta continues to purchase from Nvidia and AMD.
The internal memo acknowledged the operational challenges associated with deploying the latest commercially available AI processors across an organization of Meta’s scale.
“Adopting the latest GPUs at a firm as large as Meta has been a heavy lift, and it has cost us time,” the memo said.
Custom chips offer an opportunity to reduce that complexity by building processors specifically optimized for Meta’s infrastructure, workloads and software ecosystem.
The chip program is only one part of a much larger expansion in Meta’s AI infrastructure. According to the memo, the company expects to operate approximately seven gigawatts of computing capacity by the end of this year.
To reach that target, Meta added roughly one gigawatt during the first half of the year and expects to deploy another 5.5 gigawatts before year-end. For perspective, a single gigawatt of electricity is sufficient to power roughly 800,000 homes.
The expansion does not stop there.
Meta plans to double its AI computing capacity again, reaching approximately 14 gigawatts next year, underscoring the extraordinary scale of investment taking place across the artificial intelligence industry.
Computing power, rather than software alone, has become one of the defining competitive advantages in AI. Larger computing clusters allow companies to train bigger models more quickly, process more user requests and deploy increasingly capable AI services across billions of users.
Supporting that expansion will require enormous capital investment. Meta expects to spend as much as $145 billion on AI infrastructure this year, making it one of the largest investors in artificial intelligence globally.
That spending forms part of an industry-wide AI investment wave expected to exceed $700 billion among major technology companies as firms compete to build next-generation AI models and data centers.
The scale of spending underpins an industry consensus that access to computing capacity has become as strategically important as access to software talent. As AI models continue growing in size and complexity, shortages of chips, memory and networking equipment increasingly determine how quickly companies can deploy new products.
To support its aggressive buildout, Meta has signed long-term supply agreements with several key hardware manufacturers. According to the memo, the company has secured multi-year agreements with Samsung Electronics for memory chips, Sandisk for flash storage and Sumitomo Electric for fiber-optic infrastructure.
The agreements reflect growing competition for critical data center components as AI demand stretches global semiconductor supply chains. Shortages of high-bandwidth memory (HBM), networking equipment and advanced storage devices have become major constraints on AI expansion, forcing many technology companies to secure production capacity years in advance.
Apple and several other technology companies have already increased product prices to offset rising memory costs linked to the AI boom.
Meta’s custom silicon strategy places it alongside other hyperscale technology companies that are increasingly designing proprietary processors.
According to Mike Gualtieri, Vice President and Principal Analyst at Forrester, controlling chip development has become essential for companies seeking leadership in artificial intelligence.
“You can’t become an AI titan if you are dependent on another company for chips,” he said. “The hyperscalers and even SpaceX all plan chips because it will be the only way to compete on price for model usage.”
Meta first unveiled Iris, under its technical MTIA designation, in March alongside three other AI processors. The company plans to introduce a new AI chip approximately every six months through 2027, a significantly faster cadence than the traditional semiconductor industry, where major AI accelerators are typically refreshed annually or even less frequently.
The rapid expansion of AI infrastructure is also contributing to rising semiconductor prices across the industry. Demand for GPUs, memory chips, networking hardware, and storage has grown so rapidly that Morgan Stanley analysts have warned of “chipflation,” describing accelerating semiconductor prices as an emerging macroeconomic concern.