Elon Musk Unveils Gun Range and Spam Clusters Adding 420,000 NVIDIA GB300 GPUs
Elon Musk details Gun Range and Spam AI clusters adding 420,000 NVIDIA GB300 GPUs to Colossus 2, splitting frontier model training from production inference.
Elon Musk announced plans on November 1, 2026, for two massive new AI clusters—codenamed 'Gun Range' and 'Spam'—deploying approximately 420,000 NVIDIA GB300 GPUs by the end of 2026. The expansion pushes total capacity at the Colossus 2 site toward 1.2 million Blackwell-generation accelerators, bifurcating dedicated frontier model training from high-throughput production inference.
Gun Range and Spam: 420,000 Next-Gen Blackwell Accelerators
On November 1, 2026, Elon Musk revealed the operational roadmap for the next expansion phase of xAIs computing infrastructure, unveiling two massive cluster facilities codenamed 'Gun Range' and 'Spam.' Together, the two buildouts will bring approximately 420,000 NVIDIA GB300 GPUs online by year-end, representing an estimated $35+ billion hardware deployment. The rapid expansion pushes the total compute footprint at the Colossus 2 campus to roughly 1.2 million Blackwell-class GPUs, establishing one of the largest privately operated AI supercomputing installations in history.
Architectural Bifurcation Between Training and Production Inference
The expansion establishes a deliberate architectural separation between model pre-training and real-world deployment. The Gun Range cluster—constructed on the acquired site of a former municipal police firing range—houses 220,000 GB300 GPUs equipped with 800-gigabit network interface cards (NICs), optimized for tightly coupled, high-bandwidth distributed training of next-generation Grok models by late November 2026. Conversely, the Spam facility—repurposed from an industrial food processing warehouse—will deploy roughly 200,000 GB300 GPUs in December configured specifically for inference, utilizing simplified networking topology with less than half the fiber optics to maximize per-token energy efficiency and serving latency.
Sovereign Compute Moats and Accelerating AI Timelines
Musk credited around-the-clock engineering by SpaceX and xAI site contractors for maintaining the compressed installation schedule. By bifurcating training and inference across dedicated silicon at industrial scale, xAI aims to insulate its production offerings from third-party cloud capacity constraints while accelerating the deployment of autonomous agents and real-time reasoning capabilities across Tesla, X, and enterprise partners before the close of 2026.