SpaceX Activates Colossus II at Record Speed to Train Grok 5

On Thursday, May 21, 2026 (UTC), SpaceX confirmed the completion and full activation of its second AI computing megacluster, Colossus II. According to data from the official S-1 prospectus submitted to the Securities and Exchange Commission (SEC), the new infrastructure was fully assembled and powered in a record time of just 64 days, surpassing the 91-day schedule of the company's first cluster.

The project added a batch of 110,000 GB300 processors supplied by Nvidia to the computing campus in Memphis, Tennessee, providing an additional 220 megawatts of dedicated compute power. With this hardware expansion, Colossus II immediately began extreme-scale operations to train Grok 5, the next major language model from the integrated AI startup, xAI.

Dynamic Reinforcement Learning and Decision-Making

The speed of construction underscores SpaceX's plans for infrastructure verticalization to accelerate the global race toward general artificial intelligence. Alongside the financial announcement, the company's founder, Elon Musk, publicly stated that Grok 5's internal architecture will be structured to learn almost immediately from real-time interactions, approaching the cognitive processing speed observed in highly intelligent humans.

According to industry experts, the focus on continuous reinforcement learning aims to address the historical issue of catastrophic forgetting common to traditional static models. Training on the supercomputer, which combines hybrid use of GB200 cards and the new GB300 silicon, will enable the AI to adaptively absorb and process real-time data streams from the X social media platform and automotive telemetry data.

Hardware Expansion Roadmap and AGI

SpaceX also outlined the next steps for the physical expansion of the supercomputer. The next investment phase plans to add another 220,000 Blackwell graphics accelerators and increase power supply capacity with an additional 400 megawatts of dedicated transmission infrastructure in partnership with local utilities.

With this roadmap, the xAI and SpaceX alliance plans to achieve an integrated cluster with over 1 gigawatt of installed capacity in the coming months. The empirical behavior tests of Grok 5 will be evaluated by certifying and computing audit bodies following the final consolidation of the Memphis supercluster learning blocks.