Musk Says Colossus 2 Data Center Could Double Nvidia Chip Count By Year-End

TLDR

• Elon Musk says xAI’s Colossus 2 data center in Memphis currently runs 110,000 Nvidia GB200 chips and 440,000 GB300 chips.
• Musk posted on X that 220,000 more GB300 chips are set to go online next week.
• Another 220,000 GB300 chips are expected to be added in November.
• A further 220,000 chips could arrive in late December if timing works out, Musk said.
• The expansion would roughly double the cluster’s current chip count by the end of the year.

---

Elon Musk gave new details this week on how fast his xAI business plans to grow its Colossus 2 computing cluster. The facility is located in the Memphis, Tennessee area, near Southaven, Mississippi.

Elon Musk says $SPCX will have ~1 million GPUs online next week as another 220K GB300s come online at Colossus 2.

From there another 220K are planned for November and potentially 220K more by late December taking xAI to ~1.44 million GPUs by year end. pic.twitter.com/J70u0j8Z5r

— Shay Boloor (@StockSavvyShay) September 25, 2026

Musk shared the update in a post on X early Friday. He said the center may more than double its current Nvidia chip count by the end of 2026.

According to Musk, Colossus 2 currently uses 110,000 Nvidia GB200 chips. It also runs 440,000 GB300 chips, which are a newer generation of Nvidia hardware.

More Chips Coming Over The Next Few Months

Musk said 220,000 additional GB300 chips are expected to come online next week. That would mark the first wave of the planned expansion.

Another 220,000 GB300 chips are expected to be added in November. Musk said a further 220,000 chips could follow in late December, though he added that this batch depends on timing working out.

If all the planned additions happen, Colossus 2 would end the year with a chip count far higher than what it runs today. Musk has not given a single combined total for the year-end goal.

The SpaceX unit had previously said it aimed to equip the Memphis facility with 1 million graphics processing units. The new timeline is the most detailed schedule Musk has given so far for reaching that goal.

Musk did not say whether the site currently has enough power and readily available capacity to support the new chips right away. Data centers of this size require large amounts of electricity and cooling infrastructure.

What Colossus Is Used For

Musk has described the Colossus complex as the world’s largest AI supercomputer. The first version, Colossus 1, was built in 2024 to train xAI’s Grok chatbot.

Since then, the complex has grown beyond training Grok alone. xAI has also started renting out computing power from the site to other companies.

Earlier this year, xAI began renting capacity to Anthropic and to Alphabet’s Google. This means the Memphis cluster now supports AI projects outside of xAI as well.

The expansion comes as demand for Nvidia’s AI chips remains high across the tech industry. Companies building large language models need massive computing power to train and run them.

Nvidia’s GB200 and GB300 chips are both part of its Blackwell chip lineup. These chips are designed for high-performance AI training and processing tasks.

The pace of Musk’s plans shows how quickly AI infrastructure projects are scaling this year. Colossus 2 has grown from its original chip count to hundreds of thousands of units in a short span.

Musk’s post did not include a breakdown of costs tied to the new chip additions. He also did not mention any new power agreements tied to the expansion.

As of Friday, the most recent detail from Musk is that the next batch of 220,000 GB300 chips is due to go live next week, with the November and December additions to follow after that.