Microsoft, Google, and Meta are racing to lock down massive natural gas power plants for their AI data centers, and the scale is staggering. TechCrunch AI reports that in just the past week, these three companies announced projects totaling over 13 gigawatts of gas-fired capacity across Texas and Louisiana. Meta’s Hyperion site alone, at 7.46 GW, could power the entire state of South Dakota.
What’s driving this? Pure, uncut FOMO. The AI boom demands enormous amounts of electricity, and renewables plus grid connections can’t scale fast enough. Natural gas, especially in the southern U.S. where deposits are massive, offers a faster path to powering the next generation of AI infrastructure.
But speed comes with serious risk.
The turbine bottleneck is already here
The rush has created a hardware shortage that’s hard to overstate. According to Wood Mackenzie data cited by TechCrunch AI, gas turbine prices are projected to rise 195% by year’s end compared to 2019. New orders can’t even be placed until 2028, and delivery takes six years from there. That means companies ordering today won’t see turbines until 2034.
Think about that timeline. These companies are betting that AI will still need exponential power growth eight years from now. That’s a massive wager on a technology curve that could flatten, shift to more efficient architectures, or face regulatory headwinds.
The “bring your own power” illusion
Tech companies are increasingly moving gas plants “behind the meter,” connecting them directly to data centers and bypassing the electrical grid. The pitch sounds clean: we’re not straining public infrastructure.
The reality is different. As TechCrunch AI points out, they’re just shifting demand from the electrical grid to the natural gas grid. Natural gas generates roughly 40% of U.S. electricity, according to the Energy Information Administration. When Big Tech vacuums up gas supplies, it drives up costs for everyone, including households, hospitals, and industries that genuinely can’t switch to renewables.
Petrochemical plants, for instance, need natural gas as feedstock, not just fuel. They can’t swap in solar panels. Data centers, ironically, can.
Three scenarios that could turn ugly
- Price volatility. None of these companies have disclosed contract terms. If gas prices spike due to geopolitical disruption, production slowdowns, or extreme weather, the financial exposure could be enormous.
- Political backlash. The Texas freeze of 2021 showed what happens when gas runs short. Imagine that scenario with AI data centers competing against home heating. The political fallout would be swift and brutal.
- Stranded assets. If AI efficiency improves faster than expected, or if nuclear and renewables catch up, billions in gas infrastructure becomes dead weight on balance sheets.
What practitioners and businesses should watch
The smart play here isn’t to follow Big Tech into the gas rush. It’s to recognize what this signals about the AI infrastructure market:
- Power is the real bottleneck for AI scaling, not chips, not models, not data. Companies with energy advantages will have competitive moats.
- Diversification matters. Organizations building AI infrastructure should hedge across energy sources rather than going all-in on gas.
- Efficiency gains are undervalued. Model optimization, better chip architectures, and smarter workload scheduling could reduce power needs dramatically, making today’s gas bets look like overkill.
What stands out here is the pattern recognition failure. The tech industry has a long history of overbuilding during hype cycles, from fiber optic cables in the dot-com era to crypto mining rigs. The companies involved have the balance sheets to absorb losses if things go sideways. Smaller players following their lead might not.
The AI power race is real, but locking into a finite fossil fuel at peak demand pricing isn’t a strategy. It’s a gamble. And the house, in this case, might be the one that needs heating.
Full details are available in the original TechCrunch AI report.