Meta Platforms is moving forward with its in-house artificial intelligence chip strategy, planning to begin manufacturing its custom "Iris" processor in September 2026 as part of a broader effort to reduce dependence on external chip suppliers like Nvidia and AMD. The chip, developed in partnership with Broadcom and manufactured by Taiwan Semiconductor Manufacturing Company (TSMC), represents a significant milestone for Meta's four-generation Meta Training and Inference Accelerators (MTIA) program, which aims to power the artificial intelligence systems behind Facebook and Instagram.

According to an internal memo reviewed by Reuters, Meta has completed testing of the Iris chip in just six weeks with no major issues reported, signaling genuine momentum for an in-house effort that has struggled since its launch more than five years ago. This rapid testing cycle is notable because it demonstrates that Meta's custom silicon approach is finally gaining traction after years of development challenges.

Why Are Tech Giants Building Their Own AI Chips?

The shift toward custom silicon reflects a fundamental change in how the world's largest technology companies approach artificial intelligence infrastructure. Rather than relying exclusively on general-purpose graphics processing units (GPUs) from established suppliers, companies like Meta, OpenAI, and others are designing chips tailored to their specific workloads and proprietary models. This vertical integration strategy allows these companies to control more of their supply chain and reduce costs at scale.

"You can't become an AI titan if you are dependent on another company for chips. The hyperscalers and even SpaceX all plan chips because it will be the only way to compete on price for model usage," said Mike Gualtieri, Vice President and Principal Analyst at Forrester.

Mike Gualtieri, Vice President and Principal Analyst at Forrester

Meta's aggressive expansion plans underscore the urgency of this shift. The company plans to deploy seven gigawatts of computing infrastructure by the end of 2026, then double that capacity to 14 gigawatts in 2027. To put this in perspective, one gigawatt of energy is enough to power approximately 800,000 homes. Meta has increased its 2026 capital expenditure guidance to between $125 billion and $145 billion, a significant portion of Big Tech's projected $700 billion outlay on artificial intelligence infrastructure this year.

How Is Meta Planning to Scale Its Custom Chip Production?

• Production Timeline: Meta plans to launch a new chip approximately every six months through 2027, a faster cadence than the typical one-year or longer intervals between AI chip releases from other manufacturers.
• Supply Chain Partnerships: Beyond its work with Broadcom and TSMC, Meta has secured long-term, multi-year supply agreements with Samsung Electronics for memory chips, SanDisk for flash storage, and Sumitomo Electric for fiber-optic equipment to support its data center expansion.
• Complementary GPU Strategy: The Iris chip is designed to augment, not replace, the large quantities of GPUs Meta purchases from Nvidia and AMD, allowing the company to optimize its infrastructure for different types of workloads.

Meta's decision to pursue custom silicon comes at a time when the semiconductor industry is experiencing what analysts call "chipflation," a surge in component prices driven by intense demand from data center buildouts. Memory and other chip prices have risen rapidly enough to become a macroeconomic concern, according to Morgan Stanley analysts. By designing its own chips, Meta can potentially reduce these cost pressures and gain more predictable pricing for critical components.

What Does This Mean for the Broader AI Hardware Landscape?

Meta is not alone in this strategy. OpenAI has unveiled the "Jalapeño" inference chip, also developed in partnership with Broadcom, which targets roughly 50 percent lower inference costs per token compared with current-generation GPUs. IBM, Microsoft, and Anthropic are pursuing parallel strategies, with IBM introducing the first sub-1 nanometer chip designed to lower AI running costs and Microsoft developing its Maia family of accelerators for cloud inference.

The implications of this shift extend beyond individual companies. As companies like Meta and OpenAI secure compute resources through custom silicon, they insulate themselves from supply constraints and competitor pricing pressure. This vertical integration is reshaping competitive dynamics across the technology sector and has profound geopolitical implications, particularly regarding the critical role of Taiwan and TSMC in the global technology ecosystem.

For Broadcom, Meta's aggressive expansion represents a significant growth opportunity. The semiconductor company announced a "multi-year, multi-generation strategic partnership" with Meta in April to help design custom silicon for AI data centers. With Meta's capital expenditure guidance now reaching as high as $145 billion and its computing capacity targets doubling year-over-year, Broadcom could see substantially accelerated revenue growth in the second half of 2026 and beyond.

The race for custom silicon also signals a fundamental shift in what artificial intelligence products become economically viable to build. When inference costs collapse, the constraint on AI product design moves from compute budget to product imagination, enabling continuous, always-on intelligence embedded throughout enterprise workflows rather than occasional, prompt-based interactions. This transformation could reshape how businesses deploy artificial intelligence across their operations in the coming years.