Meta is building 10 new natural gas power plants to fuel its upcoming Hyperion AI data center, according to a TechCrunch exclusive. The scale is staggering - enough capacity to power an entire state like South Dakota. The move signals a dramatic shift in how Big Tech is confronting AI's insatiable energy demands, abandoning earlier renewable energy commitments in favor of reliable fossil fuel baseload power.
Meta just made its biggest bet yet that AI's future runs on natural gas, not sunshine and wind. The company's planned Hyperion AI data center will draw power from 10 dedicated natural gas plants, a scale of fossil fuel infrastructure unprecedented for a single tech facility.
The energy math is jaw-dropping. TechCrunch's comparison to South Dakota isn't hyperbole - the state consumed roughly 13,000 gigawatt-hours in 2025, and Meta's new gas fleet appears designed to match that threshold. For context, that's enough to train hundreds of frontier AI models simultaneously or run inference for billions of users across Meta's apps.
This marks a stunning reversal from Meta's 2020 commitment to reach net-zero emissions across its value chain by 2030. Back then, CEO Mark Zuckerberg championed renewable energy partnerships and carbon removal investments. But the company's AI ambitions, particularly around its Llama large language models and metaverse computing demands, have collided hard with grid realities.
Google and Microsoft face identical pressures. Google's emissions jumped 48% since 2019, while Microsoft's rose 30% - both driven largely by data center expansion. Amazon has quietly secured nuclear power agreements for AWS facilities, while Microsoft inked a deal to restart Three Mile Island's reactor.
But Meta's approach differs in scale and speed. Rather than waiting years for nuclear restarts or renewable buildouts, the company is essentially building its own private utility grid. Natural gas plants can be constructed in 18-24 months versus 5-7 years for nuclear or the intermittency challenges of solar and wind.
The decision reveals the brutal calculus now facing Big Tech. Training GPT-5-scale models requires sustained, reliable power measured in hundreds of megawatts. A single H100 GPU cluster can draw 20-40 megawatts continuously. You can't train a frontier model when the wind stops blowing.
Meta hasn't disclosed Hyperion's location, but energy experts suggest it'll need proximity to existing gas pipeline infrastructure and favorable regulatory environments. Texas, Ohio, and Pennsylvania emerge as likely candidates given their gas abundance and data center-friendly policies.
The climate implications cut both ways. Natural gas emits roughly half the CO2 of coal per unit energy, and modern combined-cycle plants hit 60% efficiency. But it's still fossil fuel combustion at massive scale, likely adding millions of tons of annual emissions to Meta's carbon footprint.
Environmental groups are already mobilizing. The move could trigger regulatory scrutiny under air quality permits and potentially state-level climate laws. California, where Meta is headquartered, has aggressive emissions reduction mandates that may conflict with out-of-state gas infrastructure investments.
What's particularly striking is the timing. This surfaces just as OpenAI debates its own infrastructure strategy and Nvidia ships ever-more-powerful chips that demand exponentially more electricity. The entire AI industry is hitting a wall where model capabilities outpace available clean energy.
Meta's gas play could be a bridge strategy - securing reliable power now while longer-term solutions mature. Or it might represent a more permanent acknowledgment that AI's energy appetite won't be satisfied by renewables alone. Either way, it's a watershed moment for an industry that's spent years cultivating a green image.
The real question is whether competitors follow suit. If Amazon, Google, and Microsoft all start building dedicated fossil fuel infrastructure, we're looking at a fundamental reordering of energy markets and climate trajectories. South Dakota-sized power draws, replicated across multiple companies, would represent one of the fastest buildouts of gas generation in modern history.
Meta's Hyperion project forces the tech industry's dirtiest secret into daylight - AI's explosive growth is incompatible with current renewable energy timelines. Whether this becomes the new normal or sparks a race toward alternative solutions like next-gen nuclear will determine whether AI's promise comes at an unacceptable climate cost. Every major tech company now faces the same choice Meta just made: slow down AI development, build fossil fuel infrastructure, or find breakthrough energy solutions that don't yet exist at scale. The next 18 months will reveal which path the industry takes.
