An aerial view of construction of a data centre in Lithia Springs, Georgia, on Sept 3, 2025. There are reasons beyond politics that help explain why smog-spewing fossil fuels have become the go-to power source for futuristic data centers. The pairing is almost unavoidable – at least for now. — Dustin Chambers/The New York Times
It’s been a big week for artificial intelligence data centres. That means it’s also been a big week for coal and natural gas.
This past week Nvidia announced a US$100bil (RM421.19bil) investment to support OpenAI’s enormous build-out of data centres that use its chips. The next day, OpenAI said it had signed deals with SoftBank and Oracle to build five new data centres as part of the Stargate Project, a US$500bil (RM2.10 trillion) plan for AI infrastructure. (The three companies unveiled it at the White House back in January.)
The announcements are the latest in a global push to speed the construction of AI data centres. OpenAI, Amazon, Google, Meta and Microsoft are together spending more than US$325bil (RM1.36 trillion) on them by the end of the year. To stay on the bleeding edge, the companies want the latest processors, cooling systems, facilities – all running 24/7 on mind-bending quanta of electricity.
In the US, more than half of that power is coming from fossil fuels.
President Donald Trump, who called green energy a “scam” at the UN General Assembly on Tuesday, has enthusiastically endorsed natural gas, coal and oil. He has also subsidised them.
But there are reasons beyond politics that help explain why smog-spewing fossil fuels have become the go-to power source for futuristic data centres. The pairing is almost unavoidable – at least for now.
Renewables
Sprawling solar farms, windmills and hydroelectric dams are the best energy options for the planet, and usually the cheapest. Their economic upside has made them, collectively, the fastest-growing power source for data centres worldwide.
But renewables often can’t shoulder the load alone, despite being a major part of the AI power plan. That’s because servers hum and whir around the clock – not just when the sun is up or the wind is blowing. They demand a constant, stable flow of electricity. If power falters, even for a few seconds, companies lose thousands of dollars, sometimes more.
There’s a fix: Companies can pair solar and wind farms with massive batteries that store power and then release it in a steady stream. But storing energy that way is relatively pricey and may still fall short of providing the nonstop energy that data centres need.
“Batteries are a great way to shift daytime electricity to evening electricity but not a great way to shift July electricity to January electricity,” said Matthew Bunn, a professor at Harvard who studies energy policy. So even the greenest facilities rely on fossil fuels or the local grid for backup, he said.
Another challenge: The biggest data centre campuses will consume multiple gigawatts of power. (As part of last week’s deal, OpenAI agreed to use Nvidia chips in at least 10 gigawatts’ worth of data centres.) To continuously produce just a single gigawatt, a renewable-energy plant would need around 12.5 million solar panels – enough to cover nearly 5,000 football fields. Wind turbines would need even more room. Many data centres near cities and towns don’t have that kind of space.
Nuclear
That’s where nuclear plants come in. They have smaller footprints, generate steady power and, like renewables, emit no carbon.
But they’re expensive. That’s why the nuclear industry has been in a decades-long rut. It boomed back in the 1970s, when the global energy crisis quadrupled oil prices. But Americans’ enthusiasm for nuclear energy soured after a series of headline-grabbing accidents, like the 1979 partial meltdown at Three Mile Island.
Around the same time, our electricity needs started to decline, which tends to happen in mature economies. Oil prices came down, so we stopped building nuclear reactors.
The industry has been groping for a good sales pitch ever since. With AI, it finally has one: Energy demand is soaring, and nuclear companies can help fill the gap.
A slight snag: They’ll need another seven or eight years to do it, best-case scenario, said Jacopo Buongiorno, a nuclear science professor at Massachusetts Institute of Technology. That’s how long it takes to build nuclear plants.
So, it’s a gamble: Tech companies investing in nuclear power (Microsoft, Google and Amazon, among others) are betting billions that AI demand will continue to rise a decade from now, when those nuclear facilities open for business. But it’s not clear their bets will pan out.
Fossil fuels
The US has vast natural gas reserves in underground reservoirs and offshore deposits, so it’s cheap and available. And the infrastructure to harness it can be ready fast: “A year or two, and you have a gas plant,” Buongiorno said.
If data centres continue expanding at their going rate, their energy needs will far surpass the current supply by 2030. So tech companies that need to bridge that widening gap are reaching for fossil fuels. Natural gas is already the top power source for US data centres, according to the International Energy Agency, and it’s on track to dominate through at least 2030.
The only other energy source that can be deployed in one to two years – aligning with the construction timeline for most data centres – is solar, which has its own drawbacks.
Trump’s policies are only making natural gas more attractive. The administration was already subsidising fossil fuels, and now it’s eliminating regulations and green-energy tax credits to bolster them further.
Trump says the new policies will help American companies develop AI tools unencumbered by pesky rules and oversight. Climate advocates say he’s stacking the deck for the fossil lobby.
His plan may work. It may also accelerate climate change by pumping heat-trapping gases into an atmosphere already at its highest-recorded temperature ever. For now, though, tech companies are seeing an opportunity to invest. – ©2025 The New York Times Company
This article originally appeared in The New York Times.
