Meta to put AI chip into production in September as it looks to double computing capacity, memo shows By Reuters
Meta Platforms plans to begin producing its in-house AI chip “Iris” in September, aiming to raise total computing capacity to 14 gigawatts in 2027, up from 7 gigawatts this year. An internal memo reviewed by Reuters says testing took six weeks with no major issues. Meta expects up to $145 billion in AI infrastructure spending in 2024 and is working with Broadcom and TSMC.
How this was made
The 30-second read
Why it matters
The article adds new, time-specific execution details and quantified compute deployment plans, which can re-rate expectations for Meta’s AI capex efficiency and supply-chain independence.
Market read
Traders can update positioning around Meta’s AI infrastructure trajectory based on new production timing and explicit gigawatt targets.
What to watch
The memo’s capacity targets rely on sustained supply agreements amid “chipflation” and memory shortages; any partner delays or cost overruns could blunt the expected benefit.
Background
Meta has been developing in-house AI accelerators (MTIA) for years; Reuters says the “Iris” chip is now moving into manufacturing with a reported September start.
Ticker impact
Reuters reports Meta will begin manufacturing its in-house AI chip “Iris” in September and target 14 gigawatts of computing next year.
Near-term volatility likely, with upside bias if investors view Iris as reducing long-run AI infrastructure costs and supply risk.
The memo provides new, concrete milestones (September manufacturing start, 7 GW this year, 14 GW in 2027) and names key partners (Broadcom, TSMC) that affect execution credibility.
Market effects
If Meta’s custom chips progress, it may pressure GPU demand growth rates at Nvidia/AMD at the margin and increase scrutiny of hyperscaler in-house silicon roadmaps.
Execution depends on TSMC manufacturing capacity and Broadcom design support, keeping Taiwan and US semiconductor supply-chain sentiment in focus.
Compute-capacity targets and large AI capex (up to $145B) reinforce global data-center and memory/optics demand themes.
Counterpoint
Quick testing and planned production do not guarantee yield, performance-per-watt, or cost advantages versus leading GPUs, so the cost-reduction thesis may take longer to materialize.
Key entities
- companyMeta Platforms
Plans to start manufacturing its in-house AI chip “Iris” in September and target 14 gigawatts of computing capacity in 2027.
- companyBroadcom
Working with Meta to help design the Iris chip, per the internal memo reviewed by Reuters.
- companyTaiwan Semiconductor Manufacturing Co
Manufacturing partner for Iris, per the internal memo reviewed by Reuters.
