The race to build the world's largest AI supercomputers has hit an unexpected wall: electricity. While getting access to cutting-edge graphics processing units (GPUs) was once the main challenge for AI companies, power infrastructure has quietly become the real constraint. Elon Musk's xAI is solving this by constructing its own 1.2-gigawatt power plant, while other tech giants are facing legal consequences for trying to cut corners with unpermitted generators.

Why Is Power Suddenly the Biggest Bottleneck in AI?

Running a million GPUs simultaneously requires an enormous amount of electricity. To put this in perspective, xAI's Colossus 2 data center, which will soon house 1.1 million GB300 GPUs (a type of Nvidia's most advanced processor), needs dedicated power infrastructure that most regions simply cannot provide through the grid alone. This has forced AI companies to become energy producers themselves, a role they never expected to play.

Elon Musk announced on September 25, 2026, that 220,000 Nvidia GB300 GPUs will be operational at Colossus 2 within the next week, with another 220,000 coming online in November and potentially 220,000 more by late December. When combined with existing hardware, this brings xAI's total GPU count to approximately 1.44 million units. However, none of this computing power means anything without the electricity to run it.

How Are Companies Handling the Power Problem?

The solutions vary, but they all point to a larger infrastructure crisis in the AI industry:

• Building dedicated power plants: xAI is constructing a 1.2-gigawatt power facility to supply Colossus 2, replacing unpermitted gas turbines that sparked community lawsuits in Memphis and Southaven.
• Using temporary generators: DataOne, a data center company linked to a $33 billion Microsoft deal, deployed 62 unpermitted portable generators at its New Jersey facility before being fined $1.07 million.
• Transitioning to cleaner alternatives: DataOne stated it is moving away from noisy, polluting portable turbines toward more modern fuel cells as it seeks proper air permits.

The contrast between these approaches reveals how differently companies are prioritizing speed versus compliance. xAI acknowledged its unpermitted generators created problems but committed to removing them over the course of a year as its permanent power plant comes online. DataOne, meanwhile, faced immediate regulatory action and community backlash.

What Are Communities Saying About These Data Centers?

Residents in affected areas are frustrated that fines feel like minor inconveniences to billion-dollar companies. In Vineland, New Jersey, longtime resident Steve Brown told reporters that the $1.07 million fine is "a 'don't ask for permission, just ask for forgiveness later' kind of deal." Another community member, Tiffany Leone-Vespa, added that the penalty amounts to "pocket change" for companies with billions in capital.

In Vineland, New Jersey, longtime resident Steve Brown

"It's still kind of a drop in the bucket for a company that has several billion dollars of capital," said Steve Brown, a longtime Vineland resident.

Steve Brown, Vineland resident

The concerns extend beyond fines. Residents complained about unpermitted construction of a 1.5-million-gallon liquefied natural gas (LNG) tank, excessive noise pollution affecting homes within a half-mile radius, and a lack of public consultation before the data center project began. Many residents only learned about the facility after Microsoft announced its partnership with Nebius, the company that contracted DataOne for computing resources.

How Is the AI Industry Planning to Scale Further?

The current power crisis is just the beginning. Elon Musk has stated that xAI plans to grow its data center capacity sevenfold by 2027 and is aiming for 50 million H100-equivalent GPUs by 2030. These are not modest targets. To achieve them, the company will need to solve the power problem at scale, not just for one facility but across multiple locations.

Broadcom, a major chip manufacturer, reported in 2024 that three hyperscale customers are gunning for the one-million-GPU milestone by 2027, though the company did not identify them. This suggests that xAI is not alone in facing these infrastructure challenges. The entire AI industry is racing to build computing capacity faster than the power grid can support it.

Interestingly, xAI's approach to power generation is becoming a competitive advantage. While rivals scramble to secure grid capacity or negotiate with utilities, xAI is building its own infrastructure, ensuring that power will never be the limiting factor in its expansion plans. This strategy, however, comes with regulatory and community relations costs that are only beginning to surface.

The broader lesson is clear: in the race to build the world's most powerful AI systems, electricity has become as critical as the chips themselves. Companies that solve the power problem first will likely dominate the next phase of AI development.