Every leap in artificial intelligence is, underneath the silicon, a bet on electrons. The models grab the headlines and the GPUs absorb the capital, but the binding constraint on the next decade of computing is neither chips nor talent.. It is power; firm, abundant, affordable, and ideally clean.
The companies racing to build frontier AI have quietly become some of the largest prospective buyers of electricity on the planet, and the scramble to secure megawatts has become the defining infrastructure story of the era.. The scale is hard to overstate. US data centers drew roughly 25GW in 2024; BloombergNEF projects that figure could reach 106GW by 2035, more than quadrupling the sector’s appetite in a decade..
McKinsey pegs US data-center consumption at 606TWh by 2030, up from about 147 in 2023, or close to 12 percent of the nation’s electricity. The Department of Energy and Lawrence Berkeley National Laboratory estimate data centers account for roughly 50GW of the new peak capacity the grid must add by 2030, and could consume up to 12 percent of US electricity by 2028..
Other analysts project US data-center electricity demand could grow by as much as 130 percent by 2030. Globally, the International Energy Agency expects data-center electricity use to more than double to roughly 945TWh by 2030, with the United States driving the single largest increase..
Why power, not silicon, is the bottleneck. The economics of an AI cluster are brutal in their simplicity: a GPU that is not computing is a depreciating asset burning money. Large training and inference clusters must run as close to continuously as possible to justify their staggering capital cost, which means the electricity behind them has to be available every hour of every day.
Intermittency or curtailment directly erodes the return on billions of dollars of compute. A single 100MW IT campus running flat out consumes on the order of 876GWh a year, and can produce tens to hundreds of millions of AI tokens per second.. That appetite has collided with a grid that cannot move fast enough.
Interconnection queues stretch for years, transmission is congested, and new high-voltage lines take most of a decade to permit and build. The result is a pivot that would have seemed exotic two years ago: hyperscalers are increasingly building generation directly alongside their facilities, also known as “behind the meter,” to sidestep the grid and guarantee supply on their own timeline..
The power menu: Eight ways to feed a data center. A developer’s choices differ enormously along the two axes that matter most: reliability, can the source deliver around the clock, and sustainability, what that power costs the climate.. Today’s US mix is led by natural gas near 43 percent, then nuclear around 19 percent, coal near 16 percent, wind around 10 percent, solar near seven percent, hydropower around five percent, and geothermal at a sliver under one-half of one percent..
Natural gas is the incumbent and the default; dispatchable, fast to build, politically favored, but carbon-intensive, with a wave of new gas plants now being announced specifically to serve data centers behind the meter. Oil and diesel survive only as expensive, emissions-heavy backup.
Coal remains a legacy baseload source but is both the dirtiest option and a retiring fleet. Nuclear is the gold standard for firm, carbon-free power, running above 90 percent of the time, but new plants take five to ten years or more, hence the rush toward reactor restarts and small modular designs..
Hydropower is clean and partly firm but geographically capped and increasingly drought-exposed. Wind and solar are the cheapest new energy and nearly carbon-free, but capacity factors of 30–40 percent and 20–30 percent betray their core problem for always-on compute: the wind drops and the sun sets..
Geothermal is the sleeper, under one percent of supply today, yet the one source that combines a capacity factor above 90 percent, near-zero emissions, a small water and land footprint, and independence from the transmission grid. The table below sets the contenders side by side.. – EIA (mix); supplied 2026 analysis (capacity factors, suitability); NREL and CRS (geothermal)..
The procurement scramble. The hunt for electrons has turned hyperscalers into some of the most aggressive energy buyers in the market. With grid interconnection a multi-year wait, they are signing long-dated power-purchase agreements, funding new generation outright, and reviving assets the market had written off..
Gas-turbine order books are backlogged into the late 2020s, idled nuclear units are being eyed for restart, and small modular reactors, long a slide-deck promise, are suddenly attracting real capital, much of it aimed at supplying data centers behind the meter. The common thread is firm power on a corporate timeline that a board can underwrite..
The strategic logic is identical across every option: control. Owning or contracting generation directly insulates a build-out from grid congestion, volatile wholesale prices, and the political friction of being blamed for a neighbor’s rising electric bill. It also reframes electricity from an operating cost into a strategic asset, one whose reliability and carbon profile can be engineered rather than merely purchased..
Reliability is the new currency. For decades, utilities optimized for the lowest-cost kilowatt-hour. AI has rewritten the scoreboard.
The metric that now governs site-selection meetings is the capacity factor — the share of the year a plant actually produces at full output, and the cleanest single proxy for whether a source can keep a GPU fleet saturated.. Geothermal and nuclear lead above 90 percent; gas and coal sit at 50–70 percent; hydro runs 35–50 percent; wind manages 30–40 percent; and solar, even in sunny states, lands at 20–30 percent.
This is why the behind-the-meter model has moved from fringe to frontier: co-locating dedicated generation with compute cuts transmission costs, escapes interconnection bottlenecks, locks in stable long-term pricing, and, with the right resource, secures 24/7 carbon-free power that can command a premium valuation.. The carbon math.
Reliability alone no longer wins the bid. Roughly 56 percent of the electricity powering US data centers still comes from fossil fuels, an uncomfortable fact for hyperscalers that have committed to 24/7 carbon-free operations, Google among them, with a stated 2030 target.. On a lifecycle basis, the divide is stark: natural gas, oil and coal carry materially high emissions, while nuclear, wind, solar, hydro and geothermal all sit near the bottom of the carbon curve.
In a market where buyers increasingly assign a premium to verifiably clean compute, the carbon intensity of a campus is no longer a reporting footnote, it is a valuation input.. The geothermal dark horse. Next-generation geothermal is what happens when the AI power crunch meets the American shale playbook.
Enhanced geothermal systems borrow the directional-drilling and reservoir engineering built up over two decades of oil and gas, applying it to hot dry rock that exists almost everywhere, not just the rare hydrothermal pockets that limited the old industry. That geographic flexibility is the unlock: a firm, carbon-free resource that can, in principle, be drilled near the data centers that need it..
The cost curve is bending fast. EGS capital cost averaged roughly $28,000 per kilowatt in 2021; the DOE’s Enhanced Geothermal Shot targets about $3,700 per kW, an unsubsidized levelized cost near $45 per MWh by 2035, roughly a 90 percent reduction on a trajectory comparable to solar’s collapse in cost the prior decade.
NREL projects total US geothermal capacity could climb from a few GW today to about 38GW by 2035 and 90GW by 2050, at which scale it could supply on the order of 12 percent of US electricity.. The capital is already moving. Google’s partnership with Fervo Energy delivered enhanced-geothermal power to a data center on the Nevada grid; Fervo then closed a $462 million round in December 2025, backed by Google, Breakthrough Energy Ventures and CalSTRS, to build a 500MW project in Utah.
Meta has signed paired 150MW agreements with Sage Geosystems and XGS Energy, and Google has added a 150MW deal with Ormat.. Developers are explicit that the longer-term prize is off-grid, behind-the-meter power delivered straight to compute.. 20 Nov 2025.
That momentum culminated in a landmark public debut. In May 2026, Fervo listed on the Nasdaq under the ticker FRVO, pricing an upsized offering at $27 a share and raising roughly $1.89 billion, the largest clean-energy IPO on record.. Shares opened about 30 percent above the offer price, and the stock has since traded in the mid-$30s, giving Fervo a market capitalization of approximately $10.5 billion.
What makes that number arresting is the denominator beneath it: the company booked only about $138,000 in revenue in 2025; ancillary fees, not commercial power sales, against a net loss of roughly $58 million.. In our view, investors, in other words, are not paying for what Fervo earns today but for what its pipeline implies: some $7.2 billion in long-term contracted revenue, a 500MW project under construction, and a framework agreement with Google.
It is the clearest market verdict yet that firm, carbon-free electrons for AI are worth underwriting, even years ahead of the cash flows they will produce.. The 200-campus scenario. So how far could geothermal go?
Consider a deliberately concrete thought experiment. Forecasters cluster around roughly 80GW of net new US data-center load by the mid-2030s, above DOE’s 50GW data-center figure, below BloombergNEF’s headline totals. Suppose geothermal aimed to meet a significant portion of that, about one-quarter, or roughly 20GW.
Built as standardized 100MW behind-the-meter campuses, that target resolves to a strikingly memorable number: 200 facilities.. The arithmetic is favorable. Because geothermal runs above a 90 percent capacity factor, each 100MW campus generates on the order of 788GWh a year.
Two hundred of them would deliver roughly 158TWh annually, comparable to a quarter to a third of total projected US data-center consumption in the early 2030s. That is not a rounding error; it is potentially a structural share of the entire load.. The caveat is that 20GW of dedicated geothermal would absorb more than half of NREL’s projected 2035 national buildout, so 200 campuses sit at the ambitious end of the near-term envelope.
It would demand a step-change in drilling capacity, permitting throughput and capital. But the two halves of the equation that usually fail to meet, a willing buyer and a viable resource, are, for once, both present and growing. The hyperscalers have the demand and the balance sheets; the rock has the heat..
For investors, the implication is the cleanest part of the story. As AI power demand expands, campuses that pair 24/7 carbon-free generation directly with compute may become one of the most valuable categories of infrastructure of the decade, commanding premium economics precisely because they solve reliability and sustainability at once.
We believe the AI race will be won in data centers. Increasingly, the question will become: what’s powering them?. More in Critical Power.
10 Aug 2026. More in Sustainability. 11 Jun 2026
