Musk: Colossus 2 to Nearly Double Nvidia GB300 GPUs by Year-End

# What Musk announced

Elon Musk posted that SpaceXAI plans to increase the number of Nvidia AI accelerators at its Colossus 2 data center in Memphis, Tennessee. He gave specific counts for the chips already in place and for the upcoming additions.

Colossus 2 is reported to be running 110,000 of Nvidia's GB200 chips and 440,000 of Nvidia's GB300 chips. Musk said 220,000 more GB300 chips will come online in the next week and another 220,000 in November. He added that, if timing and logistics allow, a further 220,000 GB300 chips might be added by late December.

# Why the numbers matter

Those additions would materially increase the total GB300 inventory at Colossus 2. Musk described Colossus as the "world's largest AI supercomputer," a title he has used since the original Colossus 1 cluster, which was built in 2024 to train xAI's Grok model. The expansion of Colossus 2 shows an aggressive push to scale large-scale training and inference capacity.

# What Musk did not say

Musk did not specify whether Colossus 2 currently has the spare electrical, cooling, networking, or rack-space capacity to operate the new chips without additional infrastructure. That raises practical questions about how the added chips will be powered and integrated into existing clusters.

# How Colossus is being used

Colossus started as xAI's training facility but has been expanded to support other workloads. Earlier this year xAI began renting compute at the Colossus complex to third parties, including Anthropic and Google's Alphabet. The data center is positioned to serve both internal AI model work and third-party compute demand.

# Timing and scale

Musk framed the November batch as planned and the December batch as conditional: "If we get lucky, yet another 220k GB300 by late December." Taken together with the immediate week-ahead 220k addition, that sequence would roughly double the GB300 count on site compared with the 440,000 baseline Musk provided.

# Immediate implications

Short term, the announced chip deliveries and activations are operational milestones that could increase Colossus 2's throughput for model training and inference. For customers already renting capacity, the added GPUs would expand available compute. For Nvidia, large-scale purchases like this continue to represent a major source of demand for its advanced accelerators.

# Open practical questions

The announcement leaves several operational items unanswered: whether the center needs more power generation, whether additional cooling or network upgrades are planned, and how the new chips will be staged and integrated. Musk's post did not address those specifics.

# Bottom line

Musk announced concrete GPU counts and a near-term schedule that would substantially increase Colossus 2's Nvidia GB300 inventory by the end of the year. The post confirms aggressive scaling but stops short of describing the infrastructure or timelines required to fully deploy and operate the additional chips.