Bitcoin’s Water Footprint in 2026: The Number Nobody Measures

Bitcoin spent the first days of November 2026 on the back foot. The price slipped below $77,000 on 4 November and touched a multi-month low near $75,700 as institutional outflows and nervous trading around the United States midterm count dragged the whole market lower, according to Fortune. Prices move. The questions underneath them do not. One of the most durable, and least settled, is how much of the planet’s fresh water a proof-of-work network actually uses.

The energy debate has grown up. We now have a Cambridge estimate, a rival Digiconomist estimate, a methane-offset school of thought, and, in the European Union, a mandatory electricity-disclosure field that sits on crypto white papers. Water has none of that scaffolding. It is estimated by a handful of researchers, disclosed by almost no one, and delivered to the public through a single arresting image: one Bitcoin transaction, we are told, can use as much water as a backyard swimming pool. That image is not fabricated. It is also one of the most misleading figures in the entire sustainability conversation, and taking it apart is the fastest way to understand what Bitcoin’s water footprint really is.

Three facts do most of the work. First, the large majority of that water is indirect: it is consumed at power stations, not at mining sites. Second, most mining rigs are air-cooled and barely touch water directly. Third, the number is quoted per transaction, a denominator that has almost nothing to do with how much water the network draws. Add the defining twist of 2026, the scramble to convert mining halls into artificial-intelligence data centers, and the picture gets more interesting rather than less, because the AI industry is now having the exact same argument about its own water use. This piece walks through all of it: where the water goes, who measures it, why the numbers diverge by orders of magnitude, and what an investor or a policymaker should actually watch.

The number that launched a thousand headlines

The swimming-pool line comes from Alex de Vries, the economist behind the Digiconomist research site, in a 2023 commentary in the journal Cell Reports Sustainability. His headline estimate: a single Bitcoin transaction carried an average water footprint of about 16,000 liters in 2021 once the electricity and cooling behind it were counted, roughly 6.2 million times the water behind one credit-card swipe. Scaled up, he put the network’s global water use at more than 1,600 gigaliters in 2021, projecting it toward about 2,237 gigaliters in 2023. For context, he estimated United States mining alone at somewhere between 8.6 and 35.1 gigaliters a year, and Kazakhstan, a mining hub already living through a water crisis, at 997.9 gigaliters in 2021.

Those figures made the rounds. CoinDesk summarized the moment as Bitcoin going from boiling the oceans to draining them. De Vries keeps the counter running on Digiconomist, where the live Bitcoin index in early November 2026 put fresh-water use at roughly 12,000 liters per transaction and about 3,222 gigaliters a year, which the site likens to the annual water use of Switzerland. The same model tracks electricity at 204.44 TWh and carbon at 114.03 Mt CO2, well above the figures from Cambridge, a gap we will come back to.

Here is the first tell that something is off with the framing. The per-transaction water number, by de Vries’s own account, rises and falls with the price of Bitcoin, not with how many people are transacting. A denominator that tracks the token’s market cap rather than user demand is doing something other than measuring water per payment. Hold that thought; it reappears later as the single biggest distortion in the whole debate.

Direct versus indirect: where the water actually goes

The crucial distinction, and the one almost always lost in the headline, is between direct and indirect water. Direct water is what a mine uses on-site, mostly to cool the machines. Indirect water is what the power plant upstream consumes to generate the electricity the miner buys. By de Vries’s own accounting, the indirect component dominates: roughly 80 to 90 percent of Bitcoin’s total water footprint is consumed at power stations, not at mining sites. Only the remaining tenth to fifth is direct, on-site cooling, and even that assumes the miner uses water-based cooling typical of conventional data centers.

That reframes the entire argument. Bitcoin, for the most part, does not drink water at the mine. It inherits the water intensity of its electricity. A gas turbine needs water to cool its steam cycle. A coal or nuclear plant evaporates large volumes in its cooling towers. And a hydroelectric dam, counterintuitively, carries one of the heaviest water footprints of all, because a reservoir exposes an enormous surface area to the sun and loses water to evaporation every day. When a miner plugs into a hydro-heavy grid, it looks clean on carbon and heavy on water at the same time. The two ledgers point in opposite directions, which is exactly why you cannot read one from the other.

Why most mining rigs barely sip

If indirect water dominates, the obvious question is how much a mine actually consumes on-site. The honest answer, for the typical operation, is remarkably little, because the typical operation is air-cooled. High-speed fans push air across the hashboards and carry the heat away without touching a drop of water. That is why the direct footprint is a design choice, not a fixed cost of mining.

Marathon Digital, now MARA, one of the largest public miners, makes the point bluntly in its own materials: its single largest facility used only about 38,000 gallons of water in 2024, and that was for basic amenities like bathrooms and a kitchen, not for cooling. The company contrasts that with a hyperscale data center, whose daily draw can exceed what a mining site would not consume in a decade. The figure is a corporate talking point, so treat it as one, but the underlying physics is sound: an air-cooled ASIC hall is not a water-intensive building.

The water-hungry options are the ones built for density and heat. Evaporative or adiabatic cooling deliberately evaporates water to chill incoming air, which is why it shows up as a genuine direct draw. Immersion cooling, by contrast, submerges the boards in a non-conductive dielectric fluid in a sealed loop and consumes essentially no water; paired with dry coolers it can cut water use by 95 to 98 percent against evaporative towers. Direct-to-chip systems circulate a water-glycol mix through closed cold plates, again with limited evaporation. Cooling, in other words, is a capital decision as much as an environmental one, and it sits inside the same margin math that governs where and how miners build, a calculus we have traced in our look at mining margins when the coin costs more to make.

The hydropower paradox

Since most of the footprint is indirect, the generation mix is where the water story is actually decided, and the single largest driver is the one that makes Bitcoin look greenest on carbon: hydropower. The April 2025 Cambridge study put the sustainable share of Bitcoin’s electricity at 52.4 percent, with renewables at 42.6 percent and hydro the largest single renewable at 23.4 percent of the total. A December 2025 preliminary update went further, lifting consumption to about 190 TWh and the low-carbon share to 59.4 percent, with hydropower overtaking natural gas as the number-one source. Good news for the carbon ledger. Awkward news for the water ledger, because reservoir evaporation is one of the most water-intensive forms of power on a per-megawatt-hour basis.

This is where honest accounting matters. Attributing a dam’s evaporation to Bitcoin is a defensible modeling choice, but it is a choice, and a contested one. The reservoir exists for grid power, flood control and irrigation; the water evaporates from its surface whether or not a miner buys any of the electrons. There is also a difference between water withdrawn (borrowed from a river and largely returned) and water consumed (evaporated and gone). Load the footprint with reservoir evaporation and Bitcoin looks like a water glutton. Count only what the mine physically consumes and most of the number disappears. Neither framing is dishonest; they answer different questions. The problem is that the public gets a single swimming-pool figure with the framing stripped out.

The per-transaction trap

Return to that swimming pool. The reason it misleads is that Bitcoin’s water use, like its energy use, is driven by hashrate and the generation mix behind it, not by how many transactions the network settles. Miners run the same machines flat out whether the block they win is packed with payments or nearly empty. Divide a roughly fixed resource draw by a variable, and often small, transaction count and you manufacture a per-transaction figure that can swing wildly without a single watt or liter changing hands. De Vries himself notes the number rises and falls with the Bitcoin price, because price pulls in more hashrate, which pulls in more power, which pulls in more water. It has little to do with the person sending coins.

The Ethereum Foundation makes the same point about energy in plain terms, arguing that a per-transaction metric is misleading because fewer transactions would not lead to lower energy use, which is not how the network behaves. The logic is identical for water. And the denominator keeps getting worse as a measure of real activity, because batching and layer-two settlement pack ever more economic throughput behind each on-chain transaction, the same compression dynamic reshaping fee markets that we covered in our piece on who keeps the sequencer margin as the L2 fee war ends. One on-chain transaction in 2026 can represent thousands of netted payments. Pinning a swimming pool on it is a category error.

A more honest unit is resource use per unit of security, or simply total annual draw measured against the service the network provides. That reframes the question from a consumer-shaming headline to a policy one: is the settlement assurance worth the resources it consumes? Because water and energy both ultimately track hashrate, and hashrate tracks price, the entire footprint is downstream of the market cycle, a linkage we unpacked in our halving cycle math on why a pause is not a pivot. When the price falls, as it did this week, the footprint eventually follows.

A gallon is not a gallon

Even a perfectly measured water figure would hide the thing that matters most: where the water falls. Water stress is local. A liter evaporated from a Canadian hydro basin or a Norwegian fjord in April is close to meaningless; the same liter pulled from an aquifer in drought-stricken West Texas or Central Asia in August can be a genuine problem. Global aggregates smear that distinction into nothing.

De Vries’s own regional breakdown shows why. He put Kazakhstan’s mining-related water use at 997.9 gigaliters in 2021, in a country already rationing water, while the United States range of 8.6 to 35.1 gigaliters spans everything from hydro-rich and water-rich regions to the arid grids of the Southwest. A tonne of carbon does the same damage wherever it is emitted, so a global carbon figure is meaningful. A liter of water does not, so a global water figure is not. This is the same lesson that geography teaches on the carbon side, where a megawatt-hour from Quebec hydro and a megawatt-hour from an off-grid gas flare are not remotely the same product, and it cuts even harder for water.

The practical upshot: a responsible water assessment is site-specific. It asks what the local basin can spare, what season the draw happens in, and whether the water is consumed or returned. A network-wide gigaliter total, however carefully modeled, cannot answer any of those questions, which is precisely why it travels so well as a headline and so poorly as an input to policy.

Proof of stake barely registers

The cleanest way to see how much of Bitcoin’s footprint is a function of proof of work is to look at the network that abandoned it. When Ethereum switched to proof of stake at the Merge in September 2022, its electricity use fell by more than 99.98 percent, to roughly 0.0026 TWh a year, the output of a small town rather than a mid-sized country. A validator is a normal computer drawing tens of watts. There is no hashrate arms race, no hall of fans, and therefore essentially no cooling water to speak of. The direct water footprint of staking is a rounding error, and its indirect footprint shrinks in lockstep with its vanishingly small electricity demand.

That is not a verdict that proof of stake is better money, a separate argument entirely, but it does settle the physics. The resources Bitcoin consumes are the price of its particular security model, the same model that funds the network’s long-term defense through issuance and fees, a tension we examined in our coverage of who pays Bitcoin’s security bill. Proof of stake buys its security differently and pays for it in capital at risk rather than in megawatts and liters. The table below sets the two consensus models beside the workload that is now competing with both of them for power and water.

The AI pivot changes the water story

Here is the 2026 plot twist. For two years the biggest force reshaping Bitcoin’s energy mix has been the stampede to repurpose mining sites for artificial intelligence and high-performance computing, which pay far more per megawatt than hashing does. That shift is well understood on the carbon side. Its water side is not. And it may push the direct footprint in the wrong direction.

The reason is thermal density. An air-cooled ASIC hall sheds heat with fans and little water. A rack of AI accelerators runs far hotter per square meter and increasingly demands liquid cooling, and in hot climates that often means evaporative or hybrid systems that do consume water on-site. So the very conversion that improves a site’s economics, and sometimes its carbon profile by chasing firm clean power, can raise its direct water draw, flipping Bitcoin’s defining advantage (barely sipping at the mine) into something closer to a conventional data center’s thirst. The same surge in AI compute demand is spawning entire markets for GPU capacity, including decentralized ones; we looked at one attempt to build that demand in our piece on Render’s Forth vote and whether a gateway can build AI demand.

None of this is captured in a Bitcoin water model, because once a site runs AI workloads it is no longer a Bitcoin site. The footprint migrates out of the crypto ledger and into the data-center ledger, where, as it happens, the same measurement fight is already underway.

AI’s water debate is Bitcoin’s, rerun

If you want to know how the Bitcoin water argument ends, watch the AI version, because it is running the same script a few years later. The viral AI statistic, that a short exchange with a chatbot drinks a bottle of water, came from the same place conceptually as the swimming pool: a whole-chain estimate that counts the power plant upstream. And it is now being walked back for the same reason.

Shaolei Ren of the University of California, Riverside, whose team produced the most cited estimates of AI’s water use, has since revised his figure for a single GPT-4 prompt down to roughly 15 milliliters, of which only about 5 milliliters is on-site cooling. The rest is the power station behind the socket, plus the water used to manufacture the chips. In other words, on-site cooling is perhaps 15 to 25 percent of the true water story and the generation mix is the majority, almost exactly the split de Vries found for Bitcoin. Two industries, two headline panics, one underlying structure: the water that matters is mostly indirect, and the direct draw depends on the cooling design.

The scale is real, to be clear. Google reported company-wide water consumption climbing 34 percent to 10.9 billion gallons in 2025, driven largely by data centers, though it says it replenished 78 percent of that freshwater through stewardship projects. One widely cited advocacy projection warns that United States AI data centers could consume as much as 720 billion gallons of water a year by 2028. The International Energy Agency’s landmark study on energy and AI frames the whole picture: electricity demand from data centers is set to roughly double by 2030, and water sits alongside power as the binding physical constraint (IEA, Energy and AI). The industry’s answer is the same one open to miners: closed loops. Microsoft chief executive Satya Nadella has said the company’s newest designs use as little water in a year as a restaurant, by filling the cooling loop once and recirculating rather than evaporating continuously, as reported by Tom’s Hardware.

The measurement gap

Everything above rests on estimates, and the estimates are thin. Energy modeling for Bitcoin now has two mature, competing schools: Cambridge’s bottom-up approach, which surveys miners and sums known hardware, and Digiconomist’s top-down economic model, which infers consumption from miner revenue. They disagree by a wide margin, roughly 138 TWh against more than 200 TWh, and that disagreement is itself informative. Water modeling is a generation behind. It mostly derives water from the energy figure using regional assumptions about how electricity is generated and how mines are cooled, which means every uncertainty in the energy number is inherited and then multiplied by fresh uncertainty about cooling and local water intensity.

Alexander Neumueller, who leads the Cambridge Centre for Alternative Finance’s work on digital-asset energy and climate impact, has described the core problem for energy in words that apply with even more force to water. The Cambridge study, he said, directly addresses a persistent data gap by relying on direct practitioner insights rather than abstractions. There is no equivalent practitioner survey for water. Almost no miner meters and publishes its consumption, almost no regulator requires it, and the few public figures, like MARA’s, are volunteered and unaudited.

So the honest position is wide error bars. Is Bitcoin’s water footprint closer to the physical draw at the mine, which for an air-cooled fleet is small, or to the full upstream reservoir-and-cooling-tower total, which is large? The truthful answer in 2026 is that nobody can say to better than a factor of several, and anyone quoting a precise liters-per-transaction figure without that caveat is selling certainty that does not exist.

The disclosure that is not required

If water is the externality nobody measures, part of the reason is that nobody has to publish it. The European Union came closest to forcing the issue. Under the Markets in Crypto-Assets framework, the technical standards in Commission Delegated Regulation (EU) 2025/422 make exactly one sustainability indicator always mandatory on a crypto white paper: the total annual electricity consumption of the consensus mechanism, in kilowatt-hours. Cross a threshold of 500,000 kWh a year and a few supplementary fields kick in, including renewable share, energy intensity per transaction and greenhouse-gas emissions. Water is not on the mandatory list. It sits among the optional indicators, alongside e-waste, that an issuer may disclose but is never compelled to. So even the most ambitious crypto disclosure regime in the world treats water as nice-to-have.

In the United States the picture is starker. The Securities and Exchange Commission adopted a climate-disclosure rule in March 2024, then saw it stayed in litigation, dropped its defense of it, and in 2026 moved to rescind it outright, proposing to unwind the rule rather than enforce it. Whatever one thinks of that, the effect is that federal environmental disclosure for public miners is dormant: no mandatory carbon field, let alone a water one. A United States investor who wants to know a miner’s water exposure has to reconstruct it from the company’s cooling choices and site locations, because no filing will spell it out. The contrast with the energy mix is sharp, where at least one jurisdiction now forces a number onto the page.

The case for the defense

It would be easy to read all of this as an indictment, so it is worth stating the defense plainly, because it is strong. Most mining is air-cooled and consumes almost no water at the mine. Where liquid cooling is used, immersion and direct-to-chip systems are closed loops that recirculate the same fluid rather than evaporating it. Miners can, and increasingly do, site in water-rich regions where a marginal liter costs the local basin little. And mining’s interruptible, location-flexible nature lets it chase stranded hydro and curtailed renewables that a fixed data center cannot.

There is even a water-positive story in waste heat. When a miner’s exhaust heats a Nordic district-heating network, it displaces boilers that would otherwise burn oil, gas or peat, fuels whose own life cycle carries a water cost. Daniel Batten of CH4 Capital, the most prominent voice arguing that the standard footprint is overstated, makes a related case on carbon, contending that mining at methane sources can be net beneficial and that the headline models systematically ignore the industry’s avoided emissions. The water analogue is simpler: if roughly 80 to 90 percent of the footprint is the grid’s, then cleaning up the grid, not shutting down miners, is where the water is actually saved, and a miner that funds new hydro or firms up renewables is part of that story, not apart from it.

The defense has limits. Evaporative cooling is genuinely thirsty, water-stressed regions genuinely host mines, and the AI pivot genuinely raises the direct draw. But a fair reading lands far from the swimming-pool caricature. Bitcoin’s water problem is real, local, and mostly a power-sector problem wearing a crypto costume.

Water as a siting constraint

The most useful way to think about water in 2026 is not as a footprint to shame but as a constraint that shapes where the industry can go next. For a decade, miners chose sites almost entirely on the price of power. That is changing. As mining, AI and conventional data centers all bid for the same megawatts, they are also bidding for the same water rights, cooling permits and community goodwill, and in the fastest-growing regions water is becoming the gating factor rather than power.

West Texas is the clearest example. It offers cheap, often stranded power and a welcoming regulator, which is why so much hashrate and now so much AI compute has landed there. It is also chronically water-stressed, and the arrival of evaporatively cooled AI campuses is turning water from an afterthought into a permitting fight. Expect the next wave of large sites to advertise their cooling design the way they once advertised their power price, because closed-loop or air-cooled builds can get permitted where an evaporative build cannot. Expect, too, more projects in genuinely water-rich geographies: Nordic hydro, Canadian basins, and anywhere a miner can pair with a heat offtaker that turns the thermal output into a local asset rather than a liability.

This is the quiet way the water question actually bites. Not through a global gigaliter total that moves no decisions, but through a specific permit in a specific drought-year county that decides whether a half-billion-dollar build happens at all. Water will not shut down Bitcoin. It will increasingly decide the map.

What to actually watch

Strip away the headlines and a short checklist tells you most of what you need to know about any given mine’s water exposure. First, the cooling method: air cooling or closed-loop immersion means a trivial direct draw, while evaporative or adiabatic cooling means a real one. Second, the generation mix at the site: a hydro-heavy grid is clean on carbon but heavy on indirect water, a gas-heavy grid is the reverse, and an off-grid flare build can be light on both. Third, local water stress: the same gallon means very different things in a fjord and in a desert. Fourth, the disclosure regime: in the EU, check the white paper’s sustainability fields, and in the United States, assume you will have to reconstruct the number yourself.

And one figure to distrust on sight: water per transaction. It is the number most likely to appear in a headline and least likely to tell you anything true, because it divides a price-driven resource draw by a batching-compressed denominator that has nothing to do with it. Total annual draw, measured against a specific basin and a specific cooling design, is the honest frame. Everything else is a swimming pool in search of a scandal.

The deeper point is that Bitcoin’s water story is not really a crypto story at all. It is a power-sector story, an accounting story and a siting story, dressed up as a crypto one because the token makes a better headline than a cooling tower. Measure it honestly and the network’s water bill shrinks to something smaller, more local and more fixable than the swimming pool suggests, and the real work, cleaning up the grid it draws from, turns out to be the same work the rest of the economy already owes.

Frequently Asked Questions

How much water does one Bitcoin transaction use?

Estimates run from roughly 12,000 to 16,000 liters per transaction, but the figure is misleading. Most of that water is indirect, consumed at power plants rather than at the mine, and the per-transaction framing divides a largely price-driven resource draw by a transaction count that keeps shrinking as batching and layer-two settlement pack more activity behind each on-chain transaction.

Does Bitcoin mining consume water directly?

Mostly not. Most mining rigs are air-cooled and use almost no water on-site, and by Alex de Vries’s own accounting roughly 80 to 90 percent of the footprint is indirect. MARA has said its largest facility used only about 38,000 gallons in 2024, for amenities like bathrooms and a kitchen rather than cooling.

Is Bitcoin’s water use worse than AI data centers?

They are hard to compare directly. AI accelerators run hotter and often need liquid or evaporative cooling, so a single AI site’s direct water draw can exceed an air-cooled mine’s, but both are dominated by the water used to generate their electricity. Google reported company-wide water use of about 10.9 billion gallons in 2025, driven largely by data centers.

Does proof of stake use water?

Essentially none. When Ethereum moved to proof of stake at the 2022 Merge, its electricity use fell by more than 99.98 percent to about 0.0026 TWh a year, and because there is no mining hardware to cool, its direct water footprint is a rounding error.

Do crypto miners have to disclose their water use?

In general, no. In the United States the SEC’s climate-disclosure rule is dormant and has been proposed for rescission, so there is no mandatory water field. In the European Union, MiCA’s sustainability rules make electricity disclosure mandatory but treat water as an optional indicator that issuers may report but are not required to.

HOGE Wire mining and energy desk.