The news, 365 days behind — on purpose Delayed live · replaying 2025

One Year Ago.AI

Remember how fast this is.

18JAN2024replayed
one year on
computeMeta · Nvidia

Zuckerberg says Meta is building AGI, will have 350,000 Nvidia H100 GPUs by end of 2024

The Facebook parent company plans to assemble compute power equivalent to 600,000 H100s and open-source the resulting AI systems.

Mark Zuckerberg announced via Instagram Reels that Meta is redirecting Meta-wide efforts toward building artificial general intelligence (AGI), with a plan to amass 350,000 Nvidia H100 GPUs by the end of 2024. Combined with other chips, the total compute power will be equivalent to roughly 600,000 H100s.

Zuckerberg said the company’s Fundamental AI Research (FAIR) and GenAI teams will now work closely together to develop core AGI capabilities. Meta is currently training Llama 3 and intends to release it openly, following its previous open-source strategy. The GPU count—valued at close to $9 billion.

Meta’s chief scientist Yann LeCun later wrote on X that FAIR is now a sister organization of GenAI, the AI product division.

One year later — open only if you can handle spoilers

Meta did not reach 350,000 H100s by end of 2024 due to supply constraints; Llama 3 was released in smaller sizes before a larger model arrived in mid-2025. The open-source pledge remained controversial as safety researchers argued it could accelerate misuse.

Replay thisPost on XRedditHNLinkedIn