Elon Musk announced that xAI’s first 1GW AI compute cluster, Colossus 2, is now fully operational. Here’s the kicker, the company isn’t stopping there. Plans are already underway to scale the system to 1.5GW by April, bringing the total GPU count to over 900k units.
Based on current estimates using Nvidia’s GB200 systems, a 1GW cluster translates to roughly 300k to 500k AI chips. Delivering around 2,750 exaFLOPS of compute power at BF16 / FP16 training precision. Colossus 2 now ranks among the most powerful AI supercomputers ever built, and it represents just one component of Elon’s broader infrastructure push.
During a recent podcast appearance, Elon revealed that Tesla’s Cortex 2 AI cluster at Giga Texas is expected to reach 500MW by 2026. Would deliver roughly 1,375 exaFLOPS of compute capacity, an enormous amount of processing power, even for training and simulating Tesla’s FSD system.
What’s All This Compute Really For? capacity may exceed what FSD actually requires, which raises questions about Tesla’s true intentions. The company could be preparing for large-scale AI training beyond autonomous vehicles — specifically for Optimus, its humanoid robot project. Training a general-purpose robotic system capable of perception, reasoning, and real-world interaction demands far more data, simulation, and compute than autonomous driving alone.
Between xAI’s Colossus 2 infrastructure and Tesla’s expanding Cortex capabilities, Elon is wagering that massive compute forms the foundation for his long-term AI strategy. Whether that’s smarter vehicles, more capable robots, or entirely new applications remains to be seen. One thing’s certain, though, Elon isn’t thinking small when it comes to Colossus 2.
Related Post
xAI Buys 5 Natural Gas Turbines to Power 600K+ GPU Cluster
xAI Colossus 2 Supercomputer Targets Million-GPU Milestone with $20B Investment
Grok 4 Release: xAI Claims #1 AI Model Crown in Independent Testing