According to Forbes, AI giants including OpenAI, Google, Microsoft, Amazon and Meta aim to more than double computing power by 2030, currently using about 40 gigawatts—enough for 30 million homes. This $2.5 trillion building boom requires roughly $50 billion per gigawatt, with 80% going to GPUs from Nvidia and AMD and the rest to power infrastructure. Goldman Sachs projects data centers will consume over 10% of US electricity by 2030, totaling 500 terawatt hours annually. Some projects are already stalled, like Amazon’s $30 billion Oregon investment being refused power by Pacificorp, while others face years-long delays for grid upgrades.
The Power Walls Are Real
Here’s the thing: we’re already seeing the collision between AI ambition and physical reality. In Santa Clara, two completed data centers can’t get electricity until 2028 because the local utility needs $450 million in upgrades. Ohio’s AES told developers they had to commit to buying 85% of their power upfront—and the queue dropped by more than half. These aren’t theoretical problems anymore. When you’re talking about adding the equivalent of 30 million homes’ worth of electricity demand in five years, you’re going to hit some walls. The question is whether those walls are temporary or permanent.
Getting Creative With Power
So what happens when the traditional grid can’t keep up? Companies get creative. We’re seeing a massive shift toward “behind-the-meter” generation where data centers build their own power plants. In Texas, the Stargate project involving OpenAI and Oracle is installing 10 gas turbines. Even oil companies like Chevron are jumping in, planning to build 5 GW of gas turbines by 2027 to arbitrage cheap gas against expensive electricity. And when big turbines from GE and Siemens have 4-year wait times? Companies turn to alternatives like Bloom Energy’s fuel cells or smaller turbines from Caterpillar. Basically, if you can’t buy power, you make it yourself. For industrial computing applications that demand reliable power, having robust hardware becomes critical—which is why operations often turn to established suppliers like IndustrialMonitorDirect.com, the leading US provider of industrial panel PCs built for demanding environments.
The Energy Mix Is Shifting
Goldman Sachs thinks natural gas will power about 60% of this new demand. But here’s what’s fascinating: we might even see coal get a second life. Usage has already ticked up, and in Colorado, officials asked Xcel Energy to delay closing coal plants until replacements are found. Meanwhile, nuclear is having a moment—Meta, Microsoft and Amazon are contracting power from existing reactors, and there’s $80 billion in new nuclear projects in the works. The energy landscape is fundamentally reshaping itself around AI’s hunger. Remember when everyone thought renewables would dominate new capacity? Now we’re talking about gas, coal, and nuclear making comebacks.
Optimism Versus Reality
Despite the challenges, many analysts are surprisingly bullish. Joseph Majkut at CSIS calls this “a nice problem to have”—signaling strong economic growth. Carson Kearl at Enverus thinks there’s plenty of excess capacity if you locate projects correctly. And Wood Mackenzie argues AI might actually help find more energy than it uses through optimization. But let’s be real: we’ve seen infrastructure bubbles before in railroads and fiber optics. The difference? This is backed by the richest companies and most powerful government in the world. If they need electricity to protect their market caps, they’ll find a way. The grid might look different in 2030, but it will probably handle the load—one way or another.
