Meta simultaneously builds 5 GW-level AI clusters, computing power may surpass OpenAI and Anthropic by end of year

CoinWorld news: Meta is simultaneously building five large AI clusters, each with a power capacity exceeding 1 GW. According to well-known semiconductor analysis firm Semianalysis, Meta's total computing power may surpass that of OpenAI and Anthropic by the end of this year. Among them, the Hyperion facility in Louisiana is building approximately 1 GW of infrastructure, with future expansion potential to 5 GW. The Prometheus facility in Ohio, which is already partially operational, has a planned scale exceeding 3 GW, consisting of six campuses and 27 data centers. Meta connects these facilities into a cluster via high-speed networks, and plans to link campuses that are more than 2,000 kilometers apart in the future. Due to high latency over long distances, Meta places pre-training in a single region and distributes reinforcement learning tasks across different campuses. Additionally, Meta has assembled a team of about 3,000 people dedicated to creating reinforcement learning tasks and training environments.
META4.68%
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
GateUser-656cc6e4
· 5h ago
3000 people are specifically working on RL tasks and environments, and this team size is larger than the entire workforce of many AI companies.
View OriginalReply0
GateUser-c3de680b
· 5h ago
Meta is really piling on computing power—5 GW-level clusters. Will it surpass OpenAI + Anthropic by year-end?
View OriginalReply0
GateUser-991fc58a
· 5h ago
Prometheus 3GW is divided into 6 campuses with 27 data centers. This infrastructure density looks like a money burner.
View OriginalReply0
NotYourExitLP
· 5h ago
The expansion space from 1GW to 5GW is large enough, and the name Hyperion is also quite chuunibyou, haha.
View OriginalReply0
AncientKeysUnlockNewChains
· 5h ago
From Louisiana to Ohio over 2,000 kilometers, how many milliseconds of fiber optic latency? Curious.
View OriginalReply0
HexiHoodie
· 5h ago
The long-distance latency issue is solved using pretrained localization + RL distribution; the engineering approach is quite pragmatic.
View OriginalReply0
  • Pinned