Memphis, Tenn. — xAI Plans 1.2 Million Nvidia GPUs for Colossus

Elon Musk's xAI is working to bring roughly 1.2 million of Nvidia's newest GPUs online at its Colossus data center, an effort aimed at expanding training capacity to compete with OpenAI, according to a report by Wccftech.

What Is xAI Trying to Build at Colossus?

Colossus is the mega data center xAI has used to train its Grok family of models. The buildout described in the report would push the facility's GPU count to about 1.2 million units, a scale Wccftech frames as an effort by Musk to out-build rival compute clusters used for frontier AI training.

How Many GPUs Are Involved, and Which Chips?

The figure cited is roughly 1.2 million graphics processing units, described in the report as Nvidia's latest generation of AI accelerators. The report does not specify a chip family or model number, and neither Nvidia nor xAI has published a technical specification sheet tied to this expansion that HTT News could locate.

Why Is xAI Racing to Match OpenAI's Compute?

The expansion is described as a direct response to the compute buildouts underway at OpenAI, which has its own large-scale data center commitments for training successive model generations. More GPUs translate into faster training runs and larger models, which is why compute capacity has become a headline metric in the competition among frontier AI labs, according to the report.

Have Nvidia or xAI Confirmed the Numbers?

The 1.2 million figure originates with Wccftech's reporting rather than a joint statement, securities filing, or press release from Nvidia or xAI. HTT News was not able to independently verify the GPU count, a delivery timeline, or a total cost through a primary company document. Readers should treat the number as a reported estimate pending confirmation from either company.

What's Confirmed, What Isn't, and What Comes Next

• Confirmed by the report: xAI is pursuing a large-scale expansion of Colossus's GPU footprint, per Wccftech.
• Confirmed by the report: The stated purpose is to compete with OpenAI on training compute.
• Not yet confirmed: An exact GPU model, a delivery or completion date, or a total unit cost.
• Not yet confirmed: Public statements from Nvidia or xAI verifying the 1.2 million figure.

Until xAI or Nvidia issues its own accounting, the scale of the buildout, and how quickly it can be installed and powered, remains an open question tied to Nvidia's production capacity and the availability of electricity and cooling infrastructure at the scale frontier AI training now demands.