Meta's self-developed chip Iris will go into production to reduce reliance on Nvidia

CoinWorld News: Meta plans to begin production of its next-generation self-developed AI chip, Iris, starting in September, aiming to reduce reliance on Nvidia. The chip is designed by Meta itself, with Broadcom participating in development and TSMC responsible for production. Iris is part of Meta’s self-developed accelerator MTIA series and is mainly used for AI tasks behind Facebook and Instagram. It will be used alongside Nvidia and AMD GPUs, but will not completely replace externally sourced chips. Iris testing took only 6 weeks, and no major issues were found. Meta also plans to release a new chip every six months by the end of 2027. This year, it plans to deploy 7GW of computing power, doubling to 14GW in 2027, with AI infrastructure spending of up to $145 billion. Self-developed chips can lower costs and reduce dependence on external GPUs, but in the short term they still cannot fully meet chip demand.
META4.68%
NVDA-0.70%
AVGO3.20%
TSM0.15%
AMD5.67%
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 6
  • 2
  • Share
Comment
Add a comment
Add a comment
ByteSizedAlpha
· 7h ago
One iteration every six months — this update cycle is even more intense than the iPhone's.
View OriginalReply0
APeacockSpreadingItsTailLooks
· 10h ago
The MTIA series name is quite cool, but can its actual performance compete with the H100?
View OriginalReply0
KiteAndBlock
· 10h ago
Meta finally goes after Nvidia—its $145 billion infrastructure budget is ruthless.
View OriginalReply0
GateUser-78aae297
· 10h ago
From 7GW to 14GW, the computing power arms race is entering a white-hot phase.
View OriginalReply0
GateUser-d2b4d9c6
· 10h ago
From buying GPUs to building its own, Meta's transformation is just like Apple's M1 move back then.
View OriginalReply0
PerpNightRunner
· 10h ago
The cost reduction from self-developed chips is real, but completely decoupling from NVIDIA? Not happening within five years.
View OriginalReply0
  • Pinned