Meta's In-House AI Chip Enters Mass Production in September, Doubling Computing Power to 14 Gigawatts Next Year
Claire Weston
Meta will begin mass-producing its in-house AI chip codenamed Iris this September, with plans to double its compute infrastructure from 7 GW to 14 GW by 2027 — a signal that it is building its own silicon runway for the AI arms race ahead.
What is Iris, and why does it matter?
Iris is the fourth generation in Meta's in-house AI chip roadmap (MTIA) — designed by Meta, with Broadcom assisting on design and TSMC handling fabrication.
Chip testing took only about six weeks with no major issues found — significantly faster than industry norms.
This means → Meta's chip development pipeline has crossed from lab project to production-ready stage — the first time its five-year-plus in-house effort has reached scaled manufacturing.
Why does the iteration speed stand out?
Meta plans to release a new chip roughly every six months through 2027; the industry's typical cycle is over a year.
In plain terms = competitors update once a year; Meta aims for twice that pace. If sustained, its chip performance could catch up far faster than the market expects.
The memo also acknowledged that adopting the latest GPUs is "a heavy lift for a company of Meta's scale." The in-house chips are meant to supplement, not replace, GPUs from Nvidia and AMD.
What underpins the plan to double compute?
Meta targets 14 GW of total compute infrastructure by 2027, up from 7 GW this year.
To lock in supply, Meta has signed multi-year agreements: Samsung for memory, SanDisk for flash storage, Sumitomo Electric for fiber-optic equipment.
This reflects a strategy beyond chipmaking — Meta is locking down the entire supply chain, from chips to memory to fiber, to prevent bottlenecks during rapid expansion.
How much is being spent, and what does it mean for the industry?
Meta's AI infrastructure spending this year is expected to reach $145 billion, a substantial share of Big Tech's combined $700 billion-plus outlay.
Morgan Stanley analysts note that surging demand for memory and AI chips has triggered "chip inflation" — now a macroeconomic concern.
This means → Meta's purchasing volume alone is large enough to push prices higher across the chip supply chain. That is a tailwind for Nvidia and Samsung, but for smaller buyers, chips are getting more expensive.
What to watch next?
Whether Iris mass production proceeds smoothly is the first scaled validation milestone for Meta's in-house chip program.
The core question: how much can in-house chips actually reduce Meta's dependence on Nvidia and AMD.
Put simply = building the chip is step one. Running it at scale in Meta's own data centers — and genuinely cutting GPU procurement costs — is the real test of whether this bet pays off.
Content is for reference only, not financial advice.