Why is Musk's xAI Importing Power for Giant Data Centre?

Elon Musk’s artificial intelligence company, xAI, continues to scale its infrastructure footprint by acquiring a power plant overseas and preparing to ship it to the US.

The plant will support a major data centre expansion that intends to run one million Nvidia Blackwell GPUs and require 2 GW of energy, raising urgent questions over the environmental cost of such projects.

Powering one million GPUs with global resources

After raising US$10bn through a mixture of debt and equity financing, xAI moves to acquire an overseas power plant in a bid to manage its electricity demands for a US-based data centre.

The company’s financing is divided into a US$5bn debt raise and a US$5bn strategic equity investment arranged by Morgan Stanley.

xAI’s plans mark a fivefold scale-up from its current Colossus supercomputer, which runs on 300 MW and hosts 200,000 Nvidia Hopper GPUs.

The Colossus system, already considered energy-intensive, is now being surpassed by a facility requiring several times more power.

Industry estimates suggest that a facility running one million Nvidia Blackwell GPUs would consume between 1,400 MW and 1,960 MW.

These figures account not just for the GPUs themselves, but for surrounding systems like central processing units (CPUs), storage, networking equipment and cooling systems.

Together, these extra components typically add 30% to 50% more energy usage to the GPUs’ draw, a factor expressed through a metric known as power usage effectiveness (PUE).

The sheer volume of power needed illustrates how large-scale AI development is reshaping the infrastructure requirements of the tech sector.

However, sustainability concerns follow closely behind these expansion efforts, as data centres demand not only huge energy inputs but also considerable water usage for cooling.

Gas turbines raise clean air questions

Alongside the overseas power plant acquisition, xAI is also building out its domestic energy generation capacity.

At its Memphis, Tennessee site, the company has been granted permits to operate 15 natural gas turbines.

These are a form of internal combustion engine that converts natural gas into mechanical energy to generate electricity.

The Memphis data centre launched in July 2024 with a 150 MW capacity.

Its footprint is expanding with the installation of an extra 100,000 Nvidia Hopper GPUs, effectively doubling capacity and intensifying its resource needs.

But this move has sparked controversy.

The Southern Environmental Law Center (SELC) has threatened legal action, citing the Clean Air Act and accusing xAI of running its natural gas generators without the correct environmental permits.

SELC claims that 26 turbines were installed and are operating unlawfully, raising concerns over air pollution and regulatory oversight.

The case highlights a broader tension in AI infrastructure expansion, between the speed at which companies like xAI need to act and the legal frameworks that regulate energy use and emissions.

The lawsuit points to an increasing public scrutiny over how energy-intensive technologies are deployed, particularly when fossil fuels are involved.

Energy, scale and the sustainability dilemma

xAI’s actions are part of a wider trend in the artificial intelligence industry.

As developers aim to train larger models, they increasingly concentrate their computing resources into single locations, requiring large-scale power generation to keep operations running.

This consolidation strategy also includes investing in energy infrastructure and bringing top researchers into the fold.

However, from a sustainability perspective, these moves represent a trade-off.

Natural gas is a fossil fuel that emits CO₂ when burned, contributing to global GHG levels.

Though it is sometimes marketed as a 'transition fuel', its environmental impact remains contentious.

The AI sector’s reliance on such sources places it in conflict with net-zero ambitions.

The power plant purchase is one piece of xAI’s wider strategy to keep pace with competitors in the race for AI supremacy.

Reports from Reuters suggest the company may raise a further US$20bn in equity, which could value it between US$120bn and US$200bn.

Such growth ambitions are tightly bound to the availability of energy and the environmental implications that come with it.