Why The New AI Labs Aren’t Competing on Models—They’re Competing on Electricity Instead

A section of land appears oddly incomplete as you drive past the edge of Abilene, Texas. Dusty, level, and silent save for the distant roar of wind turbines whirring somewhere behind a line of makeshift barricades. The air feels different, humming faintly with mechanical effort, but the buildings themselves aren’t particularly noteworthy—low, rectangular structures surrounded by parked pickup trucks. In the conventional sense, these are not factories. Artificial intelligence is being nourished by them.

For years, models were the main topic of discussion in the public discourse surrounding AI, along with which company created the most intelligent system or unveiled the most striking demonstration. However, it is clear from being close to these facilities that intelligence is only one aspect of the situation. Silently, electricity—raw and unrelenting—has emerged as the true limitation. Companies with the most megawatts, rather than the best ideas, may emerge victorious in the AI race.

This is now something that executives say aloud. The leadership of Microsoft has issued a warning that the energy supply may decide the future of the industry; this statement seems more like a worry than a strategy. Within a few years, data center electricity consumption is predicted to double, and utilities in some areas are already overloaded. It appears from the rate of construction that the infrastructure is finding it difficult to keep up with the aspirations.

The heat strikes first inside these data centers. Fans roaring somewhere between an airplane cabin and a warehouse vacuum, rows of servers blinking nonstop. Knowing that even a minor malfunction could disrupt systems that millions rely on, engineers cautiously navigate between racks, inspecting cables and keeping an eye on temperatures. The sustainability of this scale remains uncertain, particularly as demand continues to increase.

Subtle changes have occurred in the competition between AI labs. Businesses are negotiating power contracts and scouting areas close to energy sources rather than just employing researchers or improving algorithms. Requests for enormous new electricity loads have increased beyond what the grid can supply in Texas alone. The grid feels outdated because it was designed for a different era.

Some businesses have made the decision to not wait.

In an unprecedented move, Elon Musk’s artificial intelligence company, xAI, installed its own gas turbines, effectively constructing a private power plant in addition to its computer clusters. The temporary but efficient turbines, which produced hundreds of megawatts, arrived on trucks. As this develops, it’s difficult not to get the impression that the sector is emulating its own electrical infrastructure in real time, eschewing once-permanent systems.

The urgency is explained by economics. There are tremendous incentives to secure energy as soon as possible because a single gigawatt of AI computing capacity can produce billions of dollars in revenue annually. Businesses lose more than just time when power connections are delayed. They miss out on their chance. Access to electricity is now seen by investors as being just as valuable as talent.

A geopolitical undertone is also present. Though their energy systems are very different, China, the US, and Europe are all increasing their AI capabilities. China has scale advantages due to its massive electricity production. Europe has to make more difficult trade-offs because of regulations and higher costs. Whether physical infrastructure or political decisions will have a greater influence on the outcome is still up in the air.

Strangely, the technology itself has taken a backseat.

These days, the capabilities of models released by various labs are strikingly similar, leapfrogging one another every few months. Of course, improvements are still important, but the distinctions frequently seem minor rather than significant. The amount of power needed to train and run them doesn’t feel incremental. An entire small town’s worth of electricity can be used to train a frontier model.

This has altered the way that investors assess businesses. Previously an engineering detail, energy efficiency is now a topic of discussion during investor presentations and earnings calls. Businesses now discuss power costs and cooling systems with the same fervor they used to reserve for product features. Who succeeds may depend on efficiency innovations rather than algorithmic ones.

Something about it is a little unnerving.

Marketed as something nearly intangible, artificial intelligence is heavily reliant on very tangible realities. turbines made of steel. lines for transmission. Heat is released into the atmosphere by cooling towers. One gets the impression that AI isn’t even floating in the cloud as you see these facilities spread out over far-flung landscapes. It is securely fastened to the earth.

Local communities are starting to take notice. New substations with transformers that buzz continuously rise alongside peaceful highways in some towns. Although the jobs and investment are welcomed by local officials, the long-term effects are uncertain. After all, electricity has a limit.

Still, the race goes on.

In an effort to maintain growth, businesses are looking into nuclear energy, solar farms, and experimental cooling techniques. There are issues with cost, dependability, and environmental effects associated with each solution. Which strategy will hold up over time, or if demand will just surpass supply once more, is still up in the air.

It’s becoming clear that practical limitations rather than theoretical innovations may be more important for AI’s future. If the world’s smartest model is unable to maintain power, it is worth very little.

And the results of that contest are already being determined in peaceful locations like Abilene, where turbines whirl ceaselessly beneath the Texas sky.