Meta is doubling down on its custom silicon ambitions. According to a Reuters report cited by The Verge, the company plans to begin manufacturing its next-generation AI chip — codenamed “Iris” — as early as September 2026. The chip will join Meta’s growing family of MTIA (Meta Training and Inference Accelerators), marking another major step in the social media giant’s quest to reduce its dependence on Nvidia and AMD.
What Is the Iris Chip?
The Iris chip is the latest addition to Meta’s MTIA lineup, a family of custom-built silicon designed specifically to power Meta’s AI workloads — from ranking and recommendation systems to generative AI inference at scale. While Meta hasn’t released full technical specifications, the company previously announced plans to ship a new in-house chip every six months, and Iris fits squarely into that accelerated roadmap.
Meta’s MTIA strategy is built on three pillars:
- Rapid iteration: A new chip generation every six months, far outpacing the industry-standard one-to-two-year cycle.
- Inference-first focus: Unlike mainstream chips optimized for training, MTIA chips like Iris are built first for GenAI inference, then adapted for other workloads.
- Industry standards: Built on PyTorch, vLLM, Triton, and Open Compute Project (OCP) standards for frictionless deployment.
Why It Matters
The move is part of Meta’s broader portfolio approach to AI infrastructure. By developing its own chips alongside sourcing from industry leaders, Meta aims to:
- Cut costs: Custom chips tailored to Meta’s specific workloads deliver better compute efficiency per dollar than general-purpose GPUs.
- Reduce dependency: Less reliance on Nvidia and AMD gives Meta more leverage and supply chain stability.
- Scale faster: With hundreds of thousands of MTIA chips already deployed for inference, new generations can be dropped into existing rack infrastructure.
“There is no single chip that can meet all the demands across our varying needs, which is why we’re working to deploy a variety of chips optimized for each workload.” — Meta AI Blog
The Bigger Picture: Meta’s AI Chip Roadmap
Meta’s custom silicon push extends well beyond Iris. The company has revealed plans for four new chip generations within two years:
- MTIA 300: Already in production, used for ranking and recommendations training.
- MTIA 400, 450, 500: Capable of handling all workloads, with a primary focus on GenAI inference through 2027.
This aggressive cadence puts Meta in direct competition with other tech giants building custom AI chips — including Google (TPU), Amazon (Trainium/Inferentia), and Microsoft (Maia). The race to build cheaper, faster, and more efficient AI hardware is heating up, and Meta is making sure it stays in the game.
Frequently Asked Questions
When will the Meta Iris chip be available?
Manufacturing is expected to begin in September 2026, with deployment in Meta’s data centers following shortly after.
Will Meta sell the Iris chip to other companies?
No. Like Google’s TPU and Amazon’s Trainium, Meta’s MTIA chips are designed exclusively for internal use to power Meta’s own AI workloads across Facebook, Instagram, WhatsApp, and other platforms.
How does Iris compare to Nvidia’s H100 or B200?
Meta hasn’t released direct performance comparisons. However, MTIA chips are designed for specific Meta workloads rather than general-purpose AI computing, so they’re optimized for efficiency within Meta’s ecosystem rather than competing head-to-head on raw specs.
Looking Ahead
With Iris entering production and the MTIA roadmap accelerating, Meta is positioning itself as a serious player in the AI silicon space. For the tech industry, this means more competition, more innovation, and potentially lower costs for AI infrastructure across the board. The September manufacturing launch will be a key milestone to watch.
Source: Reuters, The Verge (July 9, 2026), Meta AI Blog