Musk Reveals xAI Compute Fleet: ~780K GPUs Across Colossus Clusters, Up to 1.44M by Year-End

On September 25, Elon Musk gave a rare detailed breakdown of xAI's compute fleet on X. Colossus 1 runs 150K H100, 50K H200 and 30K GB200; Colossus 2 already has 110K GB200 and 440K GB300, with 220K more GB300 arriving next week—bringing Colossus 2 alone to 550K GPUs and both clusters to roughly 780K, crossing 1 million next week.

Confirmed

• Musk's own disclosure of Colossus 1's configuration: 150K H100 + 50K H200 + 30K GB200
• Colossus 2's current scale: 110K GB200 + 440K GB300, plus 220K GB300 next week (relayed by @minchoi, @MickeySteamboat and others)
• Per @XFreeze's extrapolation of Musk's ramp schedule: 440K GB300 running now, 660K next week, 880K in November, and 1.1M by end of December if all goes well, with total cluster GPUs approaching 1.44M; @markk computed the two clusters currently hold 780K GPUs, reaching 1M next week; @minchoi also relayed a more aggressive claim of 3.3M H100-equivalent compute by year-end

Unconfirmed

• Some reposts (e.g., @ivanbezdomny's "880K GB300" and @CurieuxExplorer's "660K more in four weeks") diverge from the original 440K figure; Musk's original post is authoritative
• The year-end targets—1.1M GB300, 1.44M total GPUs, and 3.3M H100-equivalent—carry an "if all goes well" caveat and remain uncertain

Why it matters

• If the ramp lands on schedule, xAI would operate one of the world's largest single GPU clusters, reshaping its compute race with OpenAI and Anthropic
• Colossus 2's GB300-heavy mix shows xAI's aggressive adoption of Nvidia's latest GPUs and indirectly confirms Nvidia's supply capacity

2026-09-25 ~ 2026-09-26 · 11 related posts

• [source] Musk details xAI compute: Colossus 2 to hit 880k GB300s by year-end — elonmusk · 2026-09-25

10 near-duplicate retellings: minchoi · XFreeze · elonmusk · ivan_bezdomny · MickeySteamboat · CurieuxExplorer · XFreeze · minchoi · mark_k · rohanpaul_ai