Meta to Manufacture New AI Chip "Iris" in September 2026 to Reduce Nvidia Dependency
Over $600 billion. That is the projected global investment in AI infrastructure over the next few years, and Meta has no intention of leaving that revolution in someone else's hands. In an exclusive report published by Reuters on July 9, 2026, Meta revealed plans to begin manufacturing its newest artificial intelligence chip, codenamed "Iris," starting September 2026.
This move is not merely another product in Meta's portfolio. It is part of an ambitious strategy to redraw the dependency map of the AI chip market — a market that Nvidia and AMD currently dominate almost entirely.
What Is the Iris Chip?
"Iris" is the codename for Meta's latest chip in its Meta Training and Inference Accelerators (MTIA) family. This series is purpose-built to meet Meta's massive compute demands in training and running AI models, including large language models (LLMs) and the recommendation systems that power Facebook, Instagram, and WhatsApp.
According to Meta's official announcements on ai.meta.com/blog, the MTIA program aims to build chips designed from the ground up for Meta's specific AI workloads — rather than relying on general-purpose chips like those produced by Nvidia.
The key distinction here: Meta is not designing a chip to sell. It is designing a chip to serve itself, which gives the company complete control over performance, cost, and power efficiency.
The MTIA Program: Full Background
Meta first announced the MTIA program in 2023 with a clear vision: ship a new in-house chip every six months. This aggressive cadence puts Meta on an upward trajectory that matches — and in some cases exceeds — the traditional development cycle in the semiconductor industry.
Meta's announced roadmap so far:
| Generation | Expected Timeline | Status |
|---|---|---|
| MTIA v1 | 2024 | Deployed in data centers |
| MTIA v2 | 2025 | In production |
| Iris (next-gen) | September 2026 | Manufacturing imminent |
This schedule reflects a genuine commitment from Meta to build internal chip design capabilities, rather than relying exclusively on external suppliers.
Why Is Meta Building Its Own Chips?
The reasons are both strategic and economic:
1. Reducing Reliance on Nvidia and AMD
Nvidia controls more than 80% of the AI chip market today. This means any company seeking to build AI infrastructure — including Meta — pays premium prices and waits in supply queues. Building custom chips frees Meta from these constraints.
2. Improving Power and Cost Efficiency
Custom silicon is designed to run specific types of workloads. This translates to higher efficiency and lower power consumption compared to general-purpose chips. In data centers that consume electricity at the level of small cities, every watt saved equals millions of dollars annually.
3. Controlling the Supply Chain
The global chip shortage that struck between 2020 and 2023 remains fresh in the minds of industry leaders. Meta wants to control its own technological destiny, not be held hostage to the decisions of external suppliers.
4. Hardware-Software Integration
Meta develops its models (such as the Llama family) in parallel with its chip designs. This vertical integration between hardware and software opens performance possibilities that cannot be achieved with commercial off-the-shelf chips.
Comparison: Iris / MTIA vs Nvidia vs Google
To understand where Iris sits in the competitive landscape, here is a comparison with the major players:
| Feature | Meta MTIA (Iris) | Nvidia H100 | Nvidia B100 | Google TPU v5e |
|---|---|---|---|---|
| Type | Custom for Meta AI workloads | General-purpose AI | General-purpose AI | Custom for Google models |
| Target customer | Internal (Meta only) | Available to all | Available to all | Internal (Google Cloud) |
| Memory architecture | Optimized for Llama models | HBM3 (80GB) | HBM3e (192GB) | Optimized for TensorFlow/JAX |
| Power efficiency | Tuned for Meta workloads | High consumption | High consumption | Tuned for Google workloads |
| Availability | September 2026 | Available now | Available now | Via Google Cloud |
| Estimated cost | Unknown (internal) | ~$30,000 per chip | ~$40,000 per chip | Via Google Cloud subscription |
| Flexibility | Limited to Meta workloads | Very high | Very high | Limited to Google ecosystem |
Key takeaway: Iris is not a direct competitor to Nvidia in the open market. It is an internal tool that gives Meta greater independence and cost control. However, over the long term, it could reduce Meta's share of Nvidia purchases, affecting overall market dynamics.
What Does This Mean for You?
Whether you are a developer, a business owner, or an industry observer, Meta's chip moves affect you indirectly:
If You Use AI APIs
Iris means Meta will be able to run Llama and other models at lower cost. This could reflect in the pricing of AI-powered services across Meta's platforms and in the cost of using open-source models.
If You Build AI Infrastructure
The trend led by Meta (alongside Google, Microsoft, and Amazon) signals that major tech companies are moving toward custom silicon. If you are planning a large-scale AI project, consider whether buying from Nvidia is the only option, or whether cloud providers with proprietary chips offer a viable alternative.
If You Follow the Market
The Iris announcement is a strong signal that the AI chip market is undergoing a shift toward vertical integration. Large companies are no longer content being mere buyers — they want to be manufacturers too. This has implications for Nvidia and AMD stock prices over the medium term.
Limitations and Caveats: An Honest Assessment
Despite the excitement surrounding the Iris announcement, several realities must be considered:
First, designing chips is not enough. Meta does not own semiconductor fabrication facilities. It relies on foundries like TSMC to manufacture its designs. This means Meta is still dependent on an external supply chain, even if it has brought part of the value-add in-house.
Second, Nvidia is not standing still. Nvidia is investing billions in developing its next-generation chip architectures (Rubin and beyond), and its software ecosystem (CUDA) is something neither Meta nor any other competitor can match in the short term. CUDA is the native language for most AI developers worldwide.
Third, Iris's actual performance is unknown. To date, Meta has not released official benchmarks comparing Iris to Nvidia or Google chips. General claims of "improved efficiency" need verified data to substantiate them.
Fourth, the software challenge. Building a chip is half the battle. The other half is building a software ecosystem that allows developers to use it efficiently. Nvidia has dominated this space for years through CUDA, and breaking that hegemony requires enormous time and investment.
Predictions: Where Is the AI Chip Market Heading?
Based on current trends and the Iris announcement, we can sketch three scenarios for the next three years:
Scenario One — Vertical integration gains momentum: Major tech companies (Meta, Google, Microsoft, Amazon) continue building their own chips, reducing Nvidia's data center market share from ~80% to ~60% by 2028. Nvidia maintains its lead but with a narrower gap.
Scenario Two — Nvidia fights back hard: Nvidia releases chips with a significant performance gap over competitors, bolstered by CUDA improvements and new architectures, keeping major companies dependent despite internal alternatives. Nvidia's share drops only marginally.
Scenario Three — Strategic partnerships: Rather than building everything internally, companies adopt hybrid approaches — buying some chips and building others. This scenario is closest to the current reality, where Meta uses Nvidia chips alongside MTIA.
In all cases, the ultimate beneficiary is the consumer — competition drives prices down and accelerates innovation.
Frequently Asked Questions
What is the name of Meta's new AI chip?
The codename for the new chip is "Iris." It belongs to the Meta Training and Inference Accelerators (MTIA) family. Manufacturing is set to begin in September 2026 according to a Reuters exclusive report.
Will the Iris chip compete with Nvidia's H100?
Not directly. Iris is designed for internal use at Meta and is not a commercial product sold to customers. However, it reduces Meta's dependence on Nvidia chips, which indirectly affects competitive dynamics in the market.
When was Meta's Iris chip announced?
The plan was revealed in an exclusive report by Reuters on July 9, 2026, confirming that manufacturing would begin in September 2026. Technical details are available on the Meta AI blog.
Why is Meta investing in building its own chips?
The primary reasons are reducing reliance on Nvidia and AMD, lowering operational costs, improving power efficiency in data centers, and achieving tighter integration between hardware and software for running Meta's AI models like the Llama family.
Can I buy an MTIA chip from Meta?
No. MTIA chips, including Iris, are designed for internal use within Meta's data centers only. If you need AI infrastructure, available options include commercial Nvidia and AMD chips, or cloud services like AWS, Google Cloud, and Azure.
faq_schema_en: