Elon Musk has revealed significant updates on xAI’s supercomputing infrastructure development, highlighting the company’s ambitious push toward unprecedented computational scale. Organization’s current system, Colossus 1, served as the training foundation for Grok models—likely powering both Grok 3 and Grok 4 development based on operational timelines. However, xAI continues utilizing cloud providers for inference operations rather than maintaining complete in-house capabilities.
Colossus 1 operates with approximately 230k GPUs, comprising 150k NVIDIA H100 units, 50k H200 chips, and 30k GB200 processors according to previous company disclosures. While this configuration represents substantial computational power, it’s merely the foundation for xAI’s broader infrastructure strategy.
xAI’s upcoming Colossus 2 system aims to claim the distinction of becoming the world’s first million-GPU cluster. Initial deployment will feature 550k GB200 and GB300 units, scheduled to come online within the coming weeks. Once operational, Colossus 2 will establish itself as the first cluster globally to exceed 500k GPUs.
Memphis facility housing Colossus 2 represents a $20 billion investment in GPUs and supporting equipment. Construction began in mid-February at the Tulane site in Shelby County, Tennessee, with the company already deploying the initial computing infrastructure phase.
Tesla Megapacks will provide critical support during peak demand periods, ensuring stable power supply while protecting Shelby County residents from potential blackouts. Elon emphasized the system’s liquid-cooled design, addressing the thermal management challenges inherent in such massive computational deployments.
The facility installation includes an additional 110k NVIDIA GB200 GPUs, powered through diverse energy sources similar to the original Colossus configuration. xAI Memphis confirms they’re building redundancies for Colossus 2 that mirror the success achieved at the first facility.
Looking toward the future, Elon outlined xAI’s five-year objective: deploying 50 million units of H100-equivalent AI compute with superior power efficiency. When Colossus 2 achieves full operation, will it spark the next wave of AI computational breakthroughs, or will competitors quickly close the gap?
Grok 4 Release: xAI Claims #1 AI Model Crown in Independent Testing
xAI Expands GPU Supercluster to 340K Units, Deepens Tesla Partnership
xAI Seeks $9.3B Fundraising Round as Monthly Burn Rate Hits $1B