Meta to Begin Production of Custom 'Iris' AI Chip in September

What's happening

Meta Platforms is set to begin production of its custom artificial intelligence chip, codenamed 'Iris,' in September 2026, according to an internal memo reviewed by Reuters on July 9, 2026. Iris represents the latest iteration of Meta's MTIA program — Meta Training and Inference Accelerators — a four-generation in-house chip development initiative designed to power the company's AI training and inference workloads. Broadcom serves as Meta's design partner on the Iris chip, and Meta has secured additional supply chain agreements covering memory, flash storage, and fiber-optics to support the broader deployment.

The Iris production launch sits within a significantly larger infrastructure buildout. Meta has stated it expects to spend up to $145 billion on AI infrastructure in 2026, and the company is targeting an expansion of its overall computing capacity to 14 gigawatts by next year. By developing and manufacturing its own silicon, Meta aims to reduce its reliance on third-party accelerator suppliers, including Nvidia and AMD, for the AI workloads that underpin its advertising-driven platforms and generative AI products across Facebook, Instagram, and WhatsApp.

Why it matters for markets

Meta's vertical integration into custom silicon carries direct financial implications for its cost structure and supplier relationships. With up to $145 billion earmarked for AI infrastructure spending in 2026 alone, even a partial substitution of third-party accelerators with in-house chips could represent a meaningful reallocation of capital away from external vendors. Meta's market capitalization stands at $1.60 trillion, and its annual revenue reached $214.96 billion, giving the company the financial scale to sustain a multi-generational chip program — the MTIA roadmap now spans four generations — while continuing to operate its core advertising business.

For Nvidia, whose data center and AI accelerator products have been a primary source of demand from hyperscale customers, Meta's stated intent to reduce reliance on external suppliers introduces a longer-term demand variable. Nvidia reported revenue of $253.49 billion and carries a market capitalization of $4.91 trillion, reflecting its current dominance in AI accelerator supply. The degree to which Iris can displace Nvidia hardware at Meta's scale — particularly given Meta's 14-gigawatt compute capacity target — will determine the practical magnitude of any shift in procurement. Custom silicon typically addresses specific workloads rather than replacing general-purpose accelerators entirely, meaning the two approaches may coexist within Meta's infrastructure.

TSMC, as the world's largest dedicated semiconductor foundry and the manufacturer of advanced chips for clients including Nvidia and AMD, occupies a central position in this supply chain regardless of which vendor ultimately supplies Meta's compute. Should Meta's Iris chip be fabricated at TSMC — as is common for advanced AI accelerators given TSMC's leadership in 3nm and 5nm process nodes — the foundry could see sustained or increased demand from Meta directly, even as Meta reduces purchases from Nvidia. TSMC's revenue stands at $4.10 trillion and its market cap at $2.27 trillion, reflecting its entrenched role across the AI chip ecosystem.

Sectors and assets to watch

The primary tickers directly implicated by this development are Meta Platforms (META), Nvidia (NVDA), and Taiwan Semiconductor Manufacturing Company (TSM). Meta is the central actor, committing up to $145 billion in AI infrastructure spending and targeting 14 gigawatts of compute capacity as it brings Iris to production. Nvidia faces the prospect of reduced long-term procurement from one of the largest hyperscale AI spenders, as Meta's MTIA program is explicitly designed to internalize workloads currently served by external accelerator vendors. Broadcom, as Meta's named design partner on Iris, is also directly involved in the chip's development, though it is not among the primary tickers in this brief.

The semiconductor supply chain more broadly warrants attention. Meta's deals for memory, flash storage, and fiber-optics alongside the Iris chip suggest a wide set of component suppliers are engaged in the buildout. TSMC's position as the leading foundry for advanced-node AI chips means it remains a critical infrastructure node whether Meta sources chips internally or externally — and a ramp of Iris production in September 2026 would require foundry capacity at the leading edge.

What to watch next

Key developments to monitor include confirmation of the September 2026 Iris production start and any disclosed details on initial deployment volumes or the specific AI workloads targeted. Meta's quarterly earnings disclosures will provide the clearest visibility into how capital expenditure is being allocated between third-party accelerators and in-house silicon as the MTIA program scales. Progress toward the 14-gigawatt compute capacity target — and whether that buildout accelerates or moderates purchases of Nvidia hardware — will be a closely watched indicator of the practical impact on external suppliers. Any announcements regarding the foundry partner for Iris manufacturing would also clarify TSMC's direct role in the production ramp.