AI News · Hot news ·
Musk outlines further Nvidia chip expansion at Colossus
Elon Musk outlined plans to add more Nvidia AI chips to the Colossus data center complex in Memphis. Reports on September 25 said an additional 220,000 GB300 chips were expected to become operational the following week, with further additions planned.
Key points
- Musk said Colossus already houses around 780,000 Nvidia AI chips across Hopper and Blackwell generations.
- Another 220,000 GB300 chips were expected to become operational the following week, with 220,000 more planned for October.
- A conditional December addition could bring the complex to as many as 1.44 million chips before year end.
- Customer access, pricing and product performance effects were not reported.
What happened: Elon Musk has outlined another expansion of Nvidia AI chips at Colossus, the Memphis data center complex serving SpaceX’s xAI business. Barron’s reported on September 25 that Musk expects an additional 220,000 Nvidia GB300 Blackwell chips to become operational the following week, with further additions planned for October and potentially December. The announcement sets out a substantial expansion schedule, but the upcoming installations remain expectations rather than confirmed operational capacity.
The details: Musk said Colossus already houses around 780,000 Nvidia AI chips across the company’s Hopper and Blackwell generations. After the first planned addition of 220,000 GB300 chips, another 220,000 are expected in October. A third batch of the same size is less certain: “If we get lucky, yet another 220k GB300 by late December,” Musk posted on X. Barron’s calculated that the complex could therefore house as many as 1.44 million chips before year end, spread across the Colossus 1 and Colossus 2 facilities. That upper figure depends on all three planned additions arriving on the outlined timetable.
Background: Colossus was already intended to incorporate at least one million graphics processing units, according to a statement from Memphis’s chamber of commerce cited by Barron’s. Musk had also previously said SpaceX would exclusively use Nvidia AI chips. The latest plans therefore extend an existing commitment to Nvidia hardware rather than introduce a new supplier relationship. For Nvidia, Barron’s highlighted the importance of data center construction keeping pace with chip sales, helping avoid hardware accumulating in inventories instead of being put to use.
Who it affects: The announcement is relevant to business teams assessing the infrastructure behind an AI provider’s roadmap. The distinction is between chips housed at a facility, chips expected to become operational and capacity actually confirmed as running. Musk’s figures describe both an existing installation and future targets; they do not establish that every planned chip is already available for use. The report did not specify how the additions would affect customer access, service pricing or product performance.
What to watch: The next checkpoint is whether the first additional 220,000 GB300 chips become operational on schedule, followed by the October expansion. December’s batch warrants particular caution because Musk explicitly made it conditional. Barron’s contrasted the plans with broader questions about how quickly technology companies can build data centers. For readers evaluating the roadmap, confirmation of each operational milestone will be more informative than treating the maximum year end chip count as a completed expansion.
Our take
The planned expansion highlights the scale of infrastructure commitments behind frontier AI. Teams should distinguish announced chip additions from operational capacity when assessing a provider's roadmap.