Elon Musk Said xAI's Colossus 2 Supercomputer Could More Than Double Its Nvidia Chip Count by the End of the Year

Elon Musk said this week that Colossus 2, the AI supercomputer cluster his xAI company operates near Memphis, Tennessee, could more than double its current Nvidia chip count by the end of the year, laying out the most detailed timetable yet for a build-out he has previously said is meant to close the compute gap with rival AI labs. The report on Musk's announcement puts the cluster's current configuration at 110,000 Nvidia GB200 chips and 440,000 GB300 chips.

A staged rollout through December

Musk said a further 220,000 GB300 chips are due to come online within the next week, with another 220,000 expected in November and a final 220,000 possible in late December "if we get lucky," a schedule that would take the Memphis cluster to roughly 1.21 million GB200 and GB300 processors combined by year end, including about 1.1 million GB300s. Invezz's coverage of the plan notes that xAI had previously set a target of one million GPUs at the Memphis facility during 2026, meaning Musk's latest figures would take the site well past that original goal if the full schedule holds.

The pattern behind the numbers

The scale of Colossus 2 reflects a broader dynamic across the AI industry this year, in which every major lab has treated raw chip count as a competitive metric worth disclosing in granular, almost promotional detail, mirroring the kind of infrastructure announcements OpenAI and Anthropic have made around their own data centre commitments. Musk's own hedge, that the final tranche of chips depends on getting "lucky," is a rare moment of candour about how much of this year's AI infrastructure race still runs on supply chains, power availability and construction timelines that even the companies spending tens of billions of dollars on them cannot fully control.

What a doubled cluster is meant to buy

xAI has framed Colossus 2's expansion as central to keeping its Grok models competitive against systems trained on the far larger compute budgets of OpenAI and Google, an argument that treats chip count almost as a proxy for model quality. Whether the cluster reaches Musk's stated 1.21 million chip target on schedule will depend on the same energy and supply constraints that have already slowed comparable projects elsewhere in the industry this year, including Oracle's own flagship data centre in New Mexico.