The AI race is heating up, and Elon Musk’s firm xAI is not holding back. The company has set out to scale its compute infrastructure to the equivalent of 50 million Nvidia H100 GPUs by 2030 to enable its high-performance models like Grok.
Musk validated the ambition on X (formerly Twitter), adding that the company isn’t accumulating chips but aims to build compute that’s “100x more efficient.” An effort like that would give xAI a gigantic edge at training and putting into practice large AI systems.
While other AI companies are running tens of thousands of GPUs, xAI is going for tens of millions, not in raw chip counts, but in H100-equivalent compute, which Musk describes as the “AI compute standard.” xAI now possesses approximately 230,000 GPUs, consisting of 30,000 Nvidia GB200s, and plans to increase to 500,000 GPUs in the short term through its Memphis hub of supercomputing known as Colossus.
To finance this infrastructure, xAI is raising up to $12 billion in debt capital, according to a report by The Wall Street Journal and Reuters. The funds will be utilized to expand data centers and energy operations with an estimated $13 billion in capital spending in 2025 alone.
It’s not just about hardware, however; it’s also about power. xAI is reportedly buying an international power plant and hauling it into the U.S., enabling the company to independently support compute clusters requiring up to 2 gigawatts of electricity, a load comparable to powering nearly 2 million homes. At Colossus, the company already has dozens of gas turbines and Tesla Megapacks operating, but is being constrained by the grid with only 150MW cleared by the local utility. To get around that, it’s also building an $80 million wastewater treatment plant to cool its data centers.
There is a second mega data center already up and running in Atlanta, with nearly 12,500 Nvidia H100 graphics processing units, and built using municipal bond money, which is a sign of how rapidly xAI is expanding nationwide.
This bold move comes just days after OpenAI CEO Sam Altman claimed his company would have over 1 million GPUs listed online by the end of 2025, and might potentially need up to 100 million to meet global AI demand.
Altman’s comments point to how compute is not only a valuable but scarce commodity in the fight to control AI. Musk’s plan, however, signals that he’s not just playing catch-up; he’s trying to leap ahead.
In the wider tech environment, there are also other big players like Meta making bold acquisitions, such as buying Play AI to build voice technologies, showing just how competitive the frontier of AI has become.
But scale is different for xAI. 50 million GPUs’ worth of computing isn’t performance; it’s becoming the world’s most capable AI infrastructure leader.
