xAI Plans Further GB300 Deployment at Memphis Colossus 2 Cluster
Редакция ЦОД:Главное
xAI plans to add three further batches of 220,000 NVIDIA GB300 chips to its Colossus 2 AI cluster in Memphis, according to a statement by Elon Musk. The final batch, targeted for late December 2026, was explicitly conditional, leaving the year-end expansion target unconfirmed.
- Colossus 2 was stated to have 110,000 GB200 and 440,000 GB300 chips deployed as of September 25, 2026.
- xAI plans three additional 220,000-chip GB300 deployments, but the late-December tranche is explicitly conditional.
- The reported up-to-1.2-million-chip year-end figure is a target, not confirmed installed or operational capacity.
- No power, grid, CAPEX, facility-size or construction-readiness data were disclosed.
Elon Musk said Colossus 2 in Memphis had deployed 110,000 NVIDIA GB200 chips and 440,000 GB300 chips as of September 25, 2026. xAI plans a further 220,000 GB300 chips to become fully operational the following week, another 220,000 in November, and a third 220,000-chip batch in late December.
The December tranche is conditional. Musk framed it as dependent on favourable progress, so it should be treated as a potential deployment rather than committed operational capacity. A source headline described the possible year-end total at Colossus 2 as up to 1.2 million NVIDIA chips, but that figure remains a target rather than an achieved configuration.
Were all three planned GB300 batches to enter service, Colossus 2 would become an exceptionally large example of concentrated AI-accelerator deployment at a single facility. For the data-center supply chain, the stated schedule could point to continued demand for NVIDIA accelerator deliveries, systems integration and the supporting infrastructure needed to commission AI compute at speed.
The available information does not disclose the facility's electrical capacity, power-supply arrangement, grid infrastructure, CAPEX, floor area or construction-readiness status. Consequently, the announced chip counts cannot be translated into a power requirement, data-center capacity or operational status beyond the equipment deployment statements.
For context, Musk also cited a separate Colossus 1 configuration of 150,000 NVIDIA H100 chips, 50,000 H200 chips and 30,000 GB200 chips.
If delivered and brought into operation on the stated schedule, the expansion would represent an unusually concentrated deployment of AI accelerators at a single site. It could reinforce demand for leading-edge NVIDIA systems, rapid AI data-center fit-out and associated infrastructure, although the source provides no information on power capacity, grid supply, CAPEX or site readiness.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.