The AI race is becoming a race for electricity

We have learned to price chips, and we have learned to price compute. The next shortage is the power that runs them both, and it is not where the chips are.

"The limiting factor for AI deployment is fundamentally electrical power." When Elon Musk said that, it sounded like a provocation. Today it reads like a Wall Street analyst report. An analyst quoted by Al Jazeera put it even more bluntly. "The US has the chips and is short on power, while China has the power and is short on chips." Go figure.

That mismatch is starting to decide where AI gets built, what a kilowatt hour costs and who holds pricing power. So the AI trade is no longer just a semiconductor story. More and more, it is an energy investment.

Demand is doubling

The International Energy Agency estimated that data centre consumption was about 485 terawatt hours in 2025 and expects it todouble, to 950, by 2030. That is almost 3 per cent of all the electricity the world consumes right now. In 2024 the US alone accounted for about 45 per cent of the demand, China for 25 and Europe for 15.

What worries me is not the size of that number. It is the speed of the growth. You can build and connect a data center in two to five years. Expanding a grid can take more than ten, as the UN Economic Commission for Europe warned in September. Demand is simply arriving faster than the wires that have to carry it to the consumers. Crazy!

Four bottlenecks, all priced

The global power squeeze

Lets start with the machines. The quickest way to add firm power is natural gas, and gas turbines have become hard to get. GE Vernova's backlog of orders and reservations hit 116 gigawatts in July of 2026, and the company expects 125 by year end. It is already taking orders for 2031. Siemens Energy and Mitsubishi Heavy Industries are booked four to five years out. A data centre approved today could wait longer for its turbine than for its chips, and that assumes public anger over power bills does not slow things down even further.

That anger has a source. PJM, the largest US grid and the one serving the data center corridor of northern Virginia, cleared capacity for 2028 to 2029 at $325 per megawatt day. That is more than eleven times the price of two years ago, and it would have gone higher without a negotiated cap. PJM's own market monitor points to data center as the main motive. Around Chicago, some households already pay about 50 per cent more for power than they did two years ago. That trend is widespread nationally.

Europe has hit the wall in its own way. Ireland has restricted new data center connections around Dublin, and the Netherlands now limits where they can go. Back in the US, PJM alone has more than 100 gigawatts of new generation waiting in its connection queue.

Then there is the fuel. The Hormuz crisis is still disrupting energy flows, and that hurts most in the AI hubs that import their fuel, such as Japan, South Korea, Taiwan and much of Europe. For them it is a second risk on top of the first. Having your own energy at home is a competitive advantage again.

Who has the power?

China has built the biggest energy cushion. In 2025 alone it added more than 430 gigawatts of wind and solar, and Bloomberg NEF expects it to add over six times as much generation capacity as the US over the next five years. What China lacks is the chips. By Beijing's own estimates, its data center run at just 20 to 30 per cent ability.

America is the mirror image. It has the chips, the capital and plenty of gas, but the grid, permitting and the turbine queue keep slowing things down. So the hyperscale’s (Amazon Web Services, Microsoft Azure, Google Cloud Platform just to name a few) are working around the system, building their own generation and investing to restart or extend nuclear plants.

Then there are the energy rich middle powers, the Gulf states, the Nordics and Canada. They are selling exactly what the others lack, electricity available all day, every day, whatever the weather is good or bad, often at a lower price. But having power is not the same as being able to deliver itt. Iranian threats against data centers have pushed the UAE to break up its largest AI campus into dispersed, hardened sites. The cheapest megawatt in the world is not worth much if you cannot protect it.

What could go wrong

Years in risk management taught me one rule that rarely fails. Every shortage eventually invites an over investment. Hundreds of gigawatts of new generation are being ordered on the assumption that AI demand keeps compounding. If models get more efficient faster than expected, or AI revenues disappoint, some of that capacity will end up stranded. China's underused data centers are a preview.

The other risk is political. When families see their bills go up to serve data centers, regulators react. In the US, proposals to make data centers bring their own power, or pay more to connect, are gaining ground. The cost is moving from households to developers, and that will change the math on where it makes sense to build.

What to watch

If this is your cup of tea, there are a few things worth following. Power prices in PJM and the other grids packed with data centers come first, because that is where AI demand hits household bills before it ever shows up in inflation. Then the order books for turbines and transformers, which say more about how long new power really takes than any announcement does. Hormuz matters too. As long as the standoff drags on, every AI hub that imports its fuel pays more for its electricity. And keep an eye on China’s access to advanced chips. If Beijing gets them, the world’s biggest power surplus finally meets the world’s most important technology, and the race looks very different.

For three years the market kept asking who had the best chips. The better question now is who can actually plug them to the grid. The next round of AI will go to whoever has firm power, can deliver it quickly and can keep it safe. Electricity used to be the most boring commodity in the economy. It may now be the most strategic. And being an electrician is a great skill to have.