Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-07-13 · high
Built with Broadcom and TSMC, the chip backs a buildout targeting 7GW by the end of 2026 against up to 145 billion dollars in infrastructure spend.
Why now: The September production date lands inside the same 2026 cycle where Meta's AI infrastructure spending is already set to hit as much as 145 billion dollars.
Meta's custom AI chip, code-named Iris, moves into production in September, per Reuters. The stakes are the buildout: Meta's targeting 7 gigawatts of AI compute by the end of 2026, and 14 gigawatts in 2027. That sits against up to 145 billion dollars in 2026 AI infrastructure spending alone. That's the scale every hyperscaler is chasing. It's also why the Nvidia tax matters: shaving even a slice of GPU spend off a budget that size is real money.
Iris is Meta's fourth-generation MTIA design, developed with Broadcom and manufactured by TSMC. A six-week test run turned up no major issues, clearing the chip for production. Meta and Broadcom have also formalized their partnership through 2029, so this isn't a one-off project.
Iris supplements Meta's existing Nvidia and AMD GPUs. It doesn't replace them. Every hyperscaler wants to cut its Nvidia bill. Shipping working silicon after a clean test run is real progress, even if the GPU fleet isn't going anywhere yet.
Each link below shares sources, entities, or timing with this story.
A leaked memo (Meta stock rose ~7% on it) shows growth from 7 GW in 2026 to 14 GW in 2027 at roughly $145B annual capex, with the in-house "Iris" inference chip clearing testing and entering mass production this September (Reuters via U.S. News). Long-term memory and materials...
AMD unveiled its first rack-scale system to directly contest Nvidia at the rack level, with engineering samples in H2 2026 and mass production targeted Q2 2027. Microsoft joins Meta, OpenAI and Oracle as customers; Meta plans 1 gigawatt of Helios racks by year-end against a lo...
It's a vertical-integration play that mirrors Google's TPU and Meta's Iris silicon, aimed at cutting Nvidia dependence while positioning for public markets (BuildFastWithAI). If it's real, it says the model makers now think the chip is part of the moat, not a commodity you ren...
OpenAI posted first benchmark results for Jalapeño, its Broadcom co-developed inference ASIC, claiming 1.5x to 1.9x more throughput per kilowatt and 1.7x to 3.6x lower end-to-end latency than Nvidia GB200 and GB300 rack systems, measured on the SemiAnalysis InferenceX suite. T...
Four chips detailed: MTIA 300 (in production), 400 (lab testing), 450 and 500 (GenAI inference, 2027). From 300 to 500: HBM bandwidth up 4.5x, compute FLOPS up 25x. New chips every six months. The most aggressive custom silicon roadmap from any hyperscaler, reducing NVIDIA/AMD...
Reuters, via Tech Startups, reports capital released against deployment milestones with Anthropic deploying up to two gigawatts of Instinct MI450 starting 2027. Same structure as Nvidia/OpenAI: compute vendor capital flowing to the lab that commits to buy the silicon. A two-gi...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.