Colossus 2: Musk Announces 220,000 GB 300 Next Week

Elon Musk aims for the top spot in AI for SpaceX within a six-month horizon. He is relying on Colossus 2, already equipped with hundreds of thousands of GB 200 and GB 300 systems, and announces new numerical integrations starting next week, then in November and possibly in December. The volumes and timelines are presented by Elon Musk himself.

Musk Aims for AI Leadership and Highlights Hardware Challenges

Elon Musk claims that SpaceX could become the number one in AI within six months. He describes hardware as a complex field and believes it is incredibly difficult to quickly bring massive computing power online. He asserts that SpaceX has already demonstrated exceptional capabilities in this area and that they will improve. By December, he mentions a potential tranche of 220,000 GB 300 systems, contingent on "luck."

What's Already in Place in Memphis for Colossus

xAI operates on two supercomputer clusters, Colossus 1 and Colossus 2, located in Memphis, Tennessee. According to Elon Musk, Colossus 2 currently integrates 110,000 GB 200 systems. He adds that 440,000 GB 300 systems are already in place.

The Plan for GB 300 Additions and the Hardware Context

Elon Musk plans to integrate 220,000 new GB 300 systems next week, followed by another delivery of 220,000 units in November. He indicates that the capabilities of Colossus 2 are set to increase significantly in the coming months and that xAI will benefit from a substantial influx of GPUs. This trajectory is set against a backdrop where energy powers large data centers and where building these infrastructures requires a very large number of GPUs, ideally from NVIDIA. Elon Musk has a long-standing relationship with Jensen Huang, the founder of NVIDIA, a relationship that may have facilitated exceptional hardware orders.