Crypto’s Energy Mix in 2026: Betting on Nuclear, Running on Gas

On Friday, 16 October 2026, Bitcoin’s mining difficulty climbed almost 6% to a fresh record near 140.66 trillion, the largest upward adjustment since June, according to difficulty trackers. The price of a single coin, meanwhile, sat around $82,000, lower on the week. More machines, drawing more power, competing for a thinner margin: that is the mundane engine behind every headline about crypto and energy.

For most of the past decade, the fight over that power was a fuel fight. Renewables versus fossil. Green versus dirty. Clean coin versus carbon bomb. In late 2026 that framing is quietly out of date. The variable that now decides crypto’s carbon number is not which fuel the machines burn but which kind of electricity they need: firm power that runs 24 hours a day, or intermittent power that comes and goes with the weather. And the reason the question flipped is that the largest new buyers of firm power on the planet are the same companies that used to be Bitcoin miners.

Why the mix question changed shape

Proof-of-work mining has one unusual property that shaped its entire energy history: it can switch off in an instant and lose nothing but revenue. A mine is the ideal interruptible customer. That is why miners spent years chasing the cheapest, least reliable electrons on earth: stranded gas at remote wellheads, hydro spilling over a dam in the wet season, wind a grid cannot absorb at three in the morning. When power is almost free because nobody else wants it, intermittency is a feature, not a bug.

The AI boom broke that logic. Training and serving large models is not interruptible; a hall full of accelerators has to run, and a dropped job can cost more than the power saved. So when Bitcoin miners began converting their sites and, more importantly, their grid connections into AI data centers, the kind of electricity they needed changed overnight, from cheap-and-flexible to firm-and-constant. The mix debate inherited a new axis. It is no longer only about carbon per kilowatt-hour; it is about whether the next megawatt a miner-turned-operator signs for is a nuclear reactor, a gas turbine, or a promise of one.

This piece is about the input side of that question: where the electrons come from, and how the scramble for firm power is reshaping them. The output side, what happens to the heat those machines throw off, we examined separately in our look at mining’s waste heat. Both ledgers matter, and neither is captured by the single percentage that regulators and activists fight over.

What firm power means, and why miners suddenly want it

Grid engineers sort generation into two buckets. Firm, or dispatchable, power is available on demand whatever the weather: nuclear, natural gas, geothermal, coal, and hydro with a reservoir behind it. Variable, or intermittent, power depends on conditions you cannot schedule, mainly wind and solar. A healthy grid needs both, plus the transmission and storage to stitch them together. The cheapest electricity in any market is usually variable power at the moment nobody wants it; the most valuable is firm power at the moment everybody does.

Bitcoin’s original edge was that it monetized the first category. A mine could park next to a flare stack, a curtailed wind farm, or an off-peak dam and buy power that would otherwise be wasted, then throttle down the instant the grid needed it back. That is still a real and useful role, and it is why Texas regulators treat large mines as a demand-response resource rather than a nuisance. But a site that signs a twenty-year lease with an AI tenant cannot behave that way. The tenant needs uptime measured in nines, which means the operator needs power that does not blink.

That is the pivot in one sentence: the crypto-energy business is migrating from the cheapest intermittent power to the most contested firm power, and it is doing so at the exact moment hyperscalers, utilities, and reshored factories are all bidding for the same firm electrons. The mix will be decided by who wins those auctions, not by any slogan about going green.

The 2026 mix, in numbers

Start with the most credible bottom-up measurement. The Cambridge Centre for Alternative Finance (CCAF) surveyed 49 mining firms across 23 countries, covering roughly half the network, for its April 2025 study. It put Bitcoin’s draw at about 138 TWh a year, roughly 0.5% of global electricity, and 39.8 million tonnes of CO2 equivalent. Sustainable sources, renewables plus nuclear, supplied 52.4%, up from 37.6% in 2022.

The headline share matters less than what moved underneath it. Coal, which powered 36.6% of mining in 2022, collapsed to 8.9%. Natural gas surged from 25.0% to 38.2%, making it the single largest fuel. Renewables reached 42.6% (hydropower 23.4%, wind 15.4%, solar 3.2%) and nuclear supplied 9.8%. The industry did not simply go green; it switched from coal to gas and added firm low-carbon capacity around the edges.

A preliminary Cambridge update presented late in 2025 pushed the numbers further: about 190 TWh (a 38% jump in eighteen months as hashrate climbed), 59.4% low-carbon, and, for the first time, hydropower overtaking gas as the largest single source, with emissions near 48 Mt, according to reporting on the data. The trend line gets cleaner each year even as the absolute energy grows.

Nuclear is already in the mix, and the pipeline is enormous

The 9.8% nuclear slice is easy to overlook, but it is the fastest-moving line in the table. Nuclear is the purest firm low-carbon power there is: no weather dependency, no fuel-price spikes, a capacity factor above 90%. For an operator who has to keep an AI tenant online, that profile is close to ideal, and 2026 produced a run of deals that would have sounded fanciful two years ago.

The flagship is Riot Platforms. In May 2026 the miner signed a memorandum of understanding with Terrestrial Energy to co-locate the developer’s Integral Molten Salt Reactor (IMSR) plants with Riot data centers, targeting up to 4 gigawatts of capacity from multiple 390 MW units at sites in Texas and Kentucky, with a natural-gas bridge to supply power until the reactors exist, per Riot’s announcement. Terrestrial’s chief executive, Simon Irish, used a 2026 investor panel to pitch reactors as the answer to AI’s appetite, describing a “secular expansion in nuclear energy supply” driven by an “extraordinary confluence of factors,” in remarks his company published.

Riot is not alone. Constellation Energy is restarting the undamaged Unit 1 at Three Mile Island, rebranded the Crane Clean Energy Center, an 835 MW reactor under a twenty-year power-purchase agreement with Microsoft; the roughly $1.6 billion project cleared a key federal approval in mid-2026 and targets a 2027 restart, Utility Dive reported. Meta, meanwhile, agreed in January 2026 to back a 1.2 GW Oklo Aurora small-reactor campus in Pike County, Ohio, part of up to 6.6 GW of nuclear it lined up across three developers, according to Meta. None of these are crypto deals in the old sense, but they set the price and availability of the firm power every miner-turned-operator must now compete for.

The reason small modular reactors (SMRs) feature so heavily is that they fit the buyer. A traditional gigawatt-scale plant is a civic megaproject; an SMR is meant to be factory-built, trucked to site, and ganged together to match a data center’s footprint, which is exactly how Oklo’s 75 MW Aurora units and Terrestrial’s 390 MW IMSR plants are pitched. Whether they can be built on time and on budget is the open question, since no commercial SMR is yet operating in the United States. But the design intent, firm low-carbon power sized to a single campus, is precisely what a crypto-to-AI operator wants behind its own fence.

Almost none of it arrives before 2027

Read the dates in that table again and the catch is obvious. The Crane restart is aimed at 2027. The Oklo campus might deliver first power around 2030 and reach its full 1.2 GW only by 2034. Terrestrial’s IMSR units are a 2030s proposition, which is why the Riot agreement openly builds in a gas bridge to run the data centers until reactors exist. Even the AI leases that nominally promise power in late 2027 and 2028 depend on substations and turbines that are not built yet.

Nuclear is a decade business wearing a two-year headline. A new reactor, large or small, takes years to license, build, and fuel, and the United States has restarted exactly one shuttered unit so far. That gap between the announcement cycle and the construction cycle is the single most important fact about crypto’s firm-power turn. The deals are real and the ambition is genuine, but the clean firm electrons they promise are mostly a 2030s event. The mix between now and then will be set by whatever can be energized this year.

So gas fills the gap

What can be energized this year is natural gas. It is already the mix’s largest single fuel at 38.2%, and the firm-power squeeze is pushing that share up, not down. Gas turbines can be ordered, sited, and connected far faster than a reactor, they run whenever you want them to, and in gas-rich basins they are cheap. For an operator who promised an AI tenant uptime starting in 2027, a behind-the-meter gas plant is often the only firm option that can actually be delivered on schedule.

Some of that gas is genuinely better than the alternative. Mines and data centers placed at oil wellheads can burn methane that would otherwise be flared or vented, and because methane is roughly eighty times as potent as CO2 over twenty years, capturing it can beat the grid on net emissions. The green-leaning analyst Daniel Batten of CH4 Capital argues that methane mitigation offsets a meaningful slice of Bitcoin’s footprint and that the sustainable share is understated when that credit is ignored, a case he lays out in his published research. But flared-gas capture is a niche within a niche. The bulk of new firm gas is ordinary pipeline gas burned to keep servers running, and it will shape the 2026 to 2029 mix far more than any reactor will.

Gas also lets operators step off the grid entirely. A behind-the-meter plant at a wellhead or a dedicated gas campus avoids interconnection queues that now stretch for years in PJM and ERCOT, which is part of why miners pioneered the model and AI developers are copying it. The climate math depends entirely on the gas: captured flare gas that would otherwise vent is a net win, while a new combined-cycle plant built to serve compute is simply new fossil demand. The disclosure forms, as we will see, do not distinguish between the two.

Geography decides firmness

Firm power is intensely local. You cannot ship a reactor’s output across the country, and gas is cheapest where it comes out of the ground. That is why the map of crypto-and-AI compute is redrawing itself around a handful of US regions: the Permian and Appalachian gas basins, the PJM grid across the mid-Atlantic where the Crane and Oklo projects sit, and ERCOT in Texas, where land, gas, and a permissive grid operator meet. Each place offers a different firm-power recipe, and the fuel mix a given operator ends up with is mostly an accident of where it could get an interconnection.

The Cambridge survey carries a geographic caveat worth repeating: its reported US share, around three-quarters of surveyed capacity, reflects who answered the survey, not true global hashrate, which independent trackers put closer to the high-30s to low-40s percent for the United States. The direction is what matters. The center of gravity has moved to North America, and North America’s answer to the firm-power question is gas now, nuclear later. Where a web of Chinese hydropower once defined Bitcoin’s mix, American gas and the promise of American reactors increasingly do.

The AI collision: same megawatts, new buyers

The forcing function behind all of this is demand that dwarfs mining. The International Energy Agency estimates that data centers used about 485 TWh of electricity in 2025 and will reach roughly 945 TWh by 2030, close to 3% of global electricity, with AI the fastest-growing driver, in its Energy and AI analysis. Set that against Bitcoin’s roughly 190 TWh and the balance of power is plain: miners no longer set the price of the electrons they used to dominate.

That is why so many of them stopped competing and started leasing. In 2026 a cohort of listed miners signed multidecade contracts to host AI workloads instead of hashing: Riot’s roughly $9.1 billion, 191 MW deal at its Rockdale, Texas campus, reported to be with Anthropic and running through 2048; TeraWulf’s $19 billion, 401 MW Anthropic lease in Kentucky, as CoinDesk detailed; and a broader wave of AI and high-performance-computing commitments that crossed $70 billion across the sector. The land and power that used to make blocks now host a different kind of compute.

This also explains why hashrate and energy have pulled apart in a way they never did before. Difficulty keeps setting records because the machines still running are more efficient and more numerous, yet the fleet’s growth is slowing as the best-powered sites defect to AI. The economics of that shift, and why the market now prices a miner’s megawatts above its coins, is a story we told in our look at hashrate as an asset class.

Ethereum chose the other exit

None of this applies to the second-largest crypto network, and the contrast is the cleanest argument in the whole debate. When Ethereum switched from proof of work to proof of stake at the Merge in September 2022, it cut its electricity use by more than 99.98% overnight. The network now draws about 0.0026 TWh a year, roughly 2,601 megawatt-hours, the output of a small office park rather than a country, by the Ethereum Foundation’s accounting. Validators run on ordinary servers that sip tens of watts; there is no firm-power problem because there is barely a power problem.

That difference reframes the firm-power scramble as specifically a proof-of-work issue. Securing Bitcoin means converting electricity into hashes, so its security budget and its energy budget are the same line. Securing Ethereum means locking up capital, so its energy footprint is incidental. Whether staking’s capital lock-up carries risks of its own is a separate question, one we unpack in our comparison of liquid-staking providers, but on the narrow metric of joules per unit of security, the two networks are not in the same universe.

Cleaner later, dirtier first: why firmness cuts both ways

Put the pieces together and the firm-power turn points in two directions at once. In the long run it is bullish for the mix. If even a fraction of the announced nuclear capacity gets built, crypto-and-AI compute could become an anchor tenant for the first wave of new Western reactors in a generation, the kind of guaranteed, creditworthy, around-the-clock demand that makes a reactor financeable in the first place. Nuclear’s 9.8% share could climb into the twenties, and the carbon number would fall with it.

In the short run it points the other way. Because the reactors are not ready and the AI tenants will not wait, the firm power being procured right now is overwhelmingly gas, some of it new gas that would not have been built otherwise. The same demand signal that makes a 2032 reactor financeable also makes a 2027 gas turbine profitable. Which effect dominates depends entirely on how long the bridge lasts and whether the reactors at the far end actually arrive. Miners also still sit on the other side of the ledger as flexible load that can absorb surplus renewables, so the net grid effect is genuinely contested, not obviously good or bad.

The number nobody can agree on

Every figure above comes with an asterisk, because there is no meter on the Bitcoin network. Cambridge builds its estimate bottom-up, from hardware efficiency, surveyed fuel mixes, and location data. The rival Digiconomist index, run by the economist Alex de Vries, works top-down from miner revenue and assumptions about how much of it goes to electricity; it currently puts consumption at around 204 TWh, higher than Cambridge, along with roughly 114 Mt of CO2, 22.8 kilotonnes of electronic waste, and 3,222 gigalitres of water a year. The gap between the two methods is the real measurement war, and it is wide enough to swing any policy argument.

Cambridge frames its own survey as a corrective to exactly this guesswork. The study “directly addresses a persistent data gap by relying on direct practitioner insights rather than abstractions,” said Alexander Neumueller, the CCAF’s research lead for digital assets, energy and climate impact. That is a polite way of saying the loudest numbers in the debate are often the least grounded.

De Vries also presses the footprint that electricity figures miss: the water consumed by power plants and cooling, and the e-waste from application-specific chips that go obsolete in well under two years. Those externalities are real and belong in the ledger. So does the caveat on the other side: the much-quoted “per transaction” metrics, Digiconomist’s model implies something like 791 kWh and more than 12,000 litres of water per Bitcoin transaction, are misleading, because energy tracks hashrate and blocks, not the number of payments, and layer-two batching breaks the link entirely. The honest summary is that nobody can state Bitcoin’s energy mix to the decimal, which is exactly why the firm-versus-intermittent framing is more useful than another contested percentage.

What the mix must now disclose: SEC versus MiCA

Regulators have split on whether any of this must be reported, and the split runs straight down the Atlantic. In the United States, the SEC’s 2024 climate-disclosure rule, which would have forced large issuers to report emissions, was stayed in early 2025 and then formally proposed for rescission in 2026; the agency has cast the original rule as regulatory overreach, and federal climate disclosure is effectively dormant, per the SEC’s own filings. For a US-listed miner, the green number is a marketing choice, not a legal obligation.

Europe went the other way. Under MiCA, crypto-asset white papers must carry sustainability indicators, with total annual energy consumption always mandatory and renewable share, energy intensity, and greenhouse-gas emissions required above a consumption threshold, under Commission Delegated Regulation (EU) 2025/422, whose white-paper rules apply from late 2025, as published in the EU Official Journal. Notably, none of these frameworks asks the question this article is built around. They want a fuel mix and a carbon figure; they do not ask whether the power is firm or intermittent, and they neither reward a flared-gas site nor penalize a new gas turbine. The disclosure regime is still fighting the last war. For where US policy heads next, see our regulatory countdown.

Texas and the flexible-load paradox

Nowhere is the tension sharper than in Texas, the center of US mining. On the grid run by ERCOT, large mines have become the biggest block of controllable demand, and a 2025 state law, Senate Bill 6, now lets the operator curtail large non-critical loads during emergencies and sets up mandatory and voluntary demand-response programs for them, as Utility Dive reported. Lee Bratcher, who leads the Texas Blockchain Council, has long made the industry’s best environmental case: mines are price-sensitive load that runs when power is abundant and cheap overnight and powers down on the hottest afternoons when Texans need it most. That flexibility is mining’s strongest climate argument.

Here is the paradox the firm-power turn creates. A pure Bitcoin mine can be the perfect interruptible load, the grid’s shock absorber. An AI data center is the opposite, a must-run load that makes the grid tighter. As Texas operators convert sites to AI hosting, they trade away the very flexibility that justified them environmentally and ask the grid for firm capacity instead. The same company, on the same land, flips from the grid’s best friend to one more customer bidding up the price of power. Multiply that across every converted site and you have the mix’s defining tension in 2026.

What to watch into 2027

A few concrete markers will tell you which way the mix is breaking long before the next annual study lands.

• The Crane restart date. If Constellation energizes Three Mile Island Unit 1 on schedule in 2027, it proves firm low-carbon power can actually be delivered to this demand, not just announced.
• How long the gas bridges run. Watch whether deals like Riot and Terrestrial treat gas as a short bridge or a permanent backstop; the duration is the carbon story.
• SMR licensing. Any Nuclear Regulatory Commission milestone for Oklo’s Aurora or Terrestrial’s IMSR pulls the 2030s pipeline closer to real.
• The next Cambridge update. A confirmed move of hydropower past gas, and of nuclear above 10%, would show the firm-power turn reaching the actual mix rather than the press releases.
• Disclosure convergence. Whether the SEC leaves climate reporting dormant while MiCA’s indicators bed in will decide how much of this is even visible to investors.

The through-line is simple. Crypto’s energy question used to be about the dirtiest electron in the mix. In 2026 it is about the most reliable one, and whether an industry that spent a decade mastering the cheap, flexible tail of the grid can re-tool around its expensive, firm core without torching its own best environmental argument. The percentage will keep being argued over. The firmness is where the real money, and the real carbon, now sit.

Frequently Asked Questions

What is Bitcoin’s energy mix in 2026?

The most cited bottom-up study, from the Cambridge Centre for Alternative Finance, found that about 52.4% of Bitcoin mining ran on sustainable sources (42.6% renewables plus 9.8% nuclear), with natural gas the largest single fuel at 38.2% and coal down to 8.9%. A late-2025 preliminary update put the low-carbon share near 59.4% as hydropower overtook gas, though top-down models such as Digiconomist estimate a higher total consumption and a different balance.

What is firm power and why do crypto miners need it now?

Firm, or dispatchable, power is electricity available on demand regardless of weather, such as nuclear, natural gas, and geothermal, as opposed to intermittent wind and solar. Miners historically chased cheap intermittent power because they could switch off at any time. As they convert sites into AI data centers that must run around the clock, they now need firm power that never stops, which is scarcer and more expensive.

Are Bitcoin miners really building nuclear reactors?

Not directly, but several have signed deals that point that way. Riot Platforms agreed with Terrestrial Energy in 2026 to co-locate molten-salt reactors with its data centers, targeting up to 4 GW, while hyperscalers such as Microsoft (via the Three Mile Island restart) and Meta (via Oklo) have locked up gigawatts of nuclear. Almost all of this capacity arrives in 2027 at the earliest, and most of it not until the 2030s.

How does Ethereum’s energy use compare to Bitcoin’s?

They are not close. Ethereum switched to proof of stake at the 2022 Merge and cut its electricity use by more than 99.98%, to roughly 0.0026 TWh a year. Bitcoin’s proof of work still consumes somewhere between about 138 and 205 TWh annually depending on the model, because its security is produced by computation rather than by staked capital.

Is Bitcoin mining getting cleaner or dirtier?

Both, depending on the time frame. Each annual study shows a higher sustainable share as coal gives way to gas, hydro, and nuclear. But the near-term scramble for firm power to serve AI tenants is being met mostly by new natural gas, because reactors cannot be built fast enough, so the next few years may add fossil capacity even as the long-run trend points lower.

By the HOGE Wire mining and energy desk.