Sovereign AI and the Data-Centre Water Reckoning
The water crisis is not the price of artificial intelligence. It is the price of one architecture, and there is another.
In a county where the reservoir sits lower each summer, a notice goes up on the noticeboard at the village hall: hosepipe ban, effective Monday. Gardens will brown. The car will stay dusty. A few miles away, behind a fence and a row of unbranded transformers, a data centre hums through the same drought, drawing fresh water it will warm and evaporate to keep its processors from cooking. Nobody on the parish council voted for that trade. Few of them know it is happening. And yet, increasingly, this is the bargain modern artificial intelligence asks a community to accept without ever putting it to a vote.
There is a story doing the rounds that crystallises the unease. A viral social post claimed a famous technology billionaire had said, in effect, that cooling water for machines should come before human comfort if it brings superintelligence sooner. The quote is unverified and very probably misattributed, so we will not put words in anyone's mouth. But the reason it travelled so far is worth sitting with. It travelled because it named a fear people already had: that somewhere a decision has been taken that the water belongs to the machines now, and the rest of us will be asked to adjust.
The crisis is real. It is also a choice.
Start with what is not in dispute. The resource demands of large-scale computing are real, large, and growing fast, and the best sources we have say so plainly. The International Energy Agency estimates that data centres consumed around 415 terawatt hours of electricity in 2024, roughly 1.5 per cent of the world's total, and projects that figure to more than double to around 945 terawatt hours by 2030, a quantity the agency describes as slightly more than Japan's entire electricity consumption today. Water tracks the same curve. A United Nations University analysis, reported in mid 2026, warned that water linked to artificial intelligence could come to equal the needs of around 1.3 billion people by 2030. The World Economic Forum has noted that a single one megawatt data centre can use up to an estimated 25.5 million litres of water a year for cooling alone, comparable to the daily use of a small town.
Those numbers deserve to be taken seriously, and we take them seriously. But notice what they are actually measuring. They are not measuring artificial intelligence as such. They are measuring one way of building it: enormous, centralised, water cooled megacentres that concentrate a continent's worth of computation onto a single campus, then pump fresh water and grid power through it at industrial scale. The crisis is real. It is also, at root, an architectural choice. And a choice is something a different architecture can decline to make.
The false choice in the headline
The framing that has taken hold, water for humans versus water for machines, is presented as though it were a fact of physics. It is not. It is a property of a particular design. You only have to choose between the reservoir and the model if your intelligence is required to live inside a megacentre that drinks from that reservoir. Move the intelligence somewhere else, distribute it, shrink each unit to the right size for the job, and the dilemma simply does not arise in the same form.
This matters because false choices are politically convenient. If the public can be persuaded that draining the water table is the unavoidable price of progress, then the draining becomes nobody's fault and everybody's duty. The more honest position is that the megacentre is one answer to the question of where intelligence should live, and it happens to be the thirstiest one. There are other answers. Mickai is built around one of them.
A different place for intelligence to live
Mickai is a Sovereign Intelligence Operating System, a SIOS, not an app and not a service you rent from someone else's campus. It runs fifty specialised artificial intelligence brains on hardware the operator owns, distributed and fully capable of running offline. Instead of one monolithic megamodel waiting in a distant hall to answer every question on earth, Mickai runs right sized, specialised models close to where the work actually happens: on edge devices, on workstations, and on its own server lineup, up to the flagship Prometheus edge server. Mickai is also actively training and specialising its own models now, so the capability lives with the operator rather than being leased back to them by the hour.
The water argument follows almost as a side effect. Intelligence that is distributed across hardware people own, sized to its task rather than inflated to monolithic scale, and sited near where it is used, does not need a region's water table routed through a single cooling plant to function. You are not pouring the output of a thousand small workloads into one evaporative furnace. You keep the water. You also keep the intelligence. The two were never genuinely opposed; the megacentre only made them look that way.
It is worth being precise about what is and is not being claimed here, because the temptation to overclaim is exactly the hype we are trying to avoid. Mickai is not going to single handedly end the world's data centre water problem. No single architecture vendor could. What Mickai offers is narrower and more defensible: the architecture that creates the dilemma is a choice, and there is a credible alternative architecture that does not create it in the first place. The honest pitch is not we will save every river. It is you do not have to drain one to have intelligence you control.
Sovereignty, accountability and the commons converge
What makes this more than an engineering footnote is how neatly it folds into the rest of the Mickai thesis. The case for sovereign artificial intelligence has always rested on three legs: own your intelligence, prove what it did, and do not drain the commons to run it. The water reckoning turns out to strengthen all three at once.
Owning your intelligence means it lives on your hardware, under your control, rather than on a campus whose cooling bill is paid in a stranger's drought. Proving what it did is the work of the Open Audit Record, or OAR: consequential actions are sealed and signed using FIPS 204 ML-DSA-65, the published post-quantum standard from the United States National Institute of Standards and Technology, adopted rather than invented. Those sealed records are then anchored by committing a hash of the record to Bitcoin through Pantheon, Mickai's sovereign Bitcoin anchored Layer 1, whose token is PAN with a fixed supply of five billion. Anchoring is not spending and it is not a Layer 2; it is a small, durable proof that a record existed and has not been altered. And not draining the commons is the architectural point of this whole piece. Sovereign artificial intelligence, it turns out, is also the accountable kind and the sustainable kind. They are the same design seen from three angles.
A worked example
Consider a regional hospital trust that wants artificial intelligence across its operations: triage support in accident and emergency, transcription for clinicians, pattern detection in imaging, and a governance layer that can later show, under audit, exactly what the system recommended and when. The hyperscale answer routes all of that to a distant megacentre. Patient context leaves the building, latency rises, the trust becomes a tenant of someone else's compute, and a share of that campus's cooling water and power is, in effect, billed to the public good in a place the trust will never see.
The sovereign answer keeps the work in the building. Specialised Mickai brains run on hardware the trust owns, sized to each task rather than to a one model fits all behemoth. Sensitive data does not leave the estate. The system keeps working when the connection drops, which in a hospital is not a luxury. Every consequential action is sealed into the Open Audit Record, so when an inspector or a coroner asks what the system did, there is a signed, anchored answer rather than a shrug. And the water that would have been evaporated to cool a distant hall stays in the watershed it belongs to. Same clinical capability. None of the manufactured dilemma.
The quotable line
Micky Irons, the inventor behind Mickai, puts the argument in plain terms.
“The water crisis is not a tax artificial intelligence has to levy on the rest of us. It is the bill for one way of building it. Choose a different architecture, one that runs on hardware you own and stays the size it needs to be, and you do not have to choose between the reservoir and the model. You keep both.”
It is a claim with engineering behind it rather than slogan behind it. The portfolio that protects the approach runs to 101 filed United Kingdom patent applications, around 2,234 claims in total, owned by Mickai LTD. They are filed applications, and that is exactly how we describe them. They are evidence that the architecture is real and defensible, not a headline to wave about. The headline is the water that stays in the ground.
What the next few years actually ask of us
The honest reading of the IEA and United Nations figures is not despair. It is a fork in the road. If the world keeps answering the demand for artificial intelligence by building ever larger water cooled megacentres, then the projections, electricity more than doubling towards 945 terawatt hours by 2030, water rising towards the needs of well over a billion people, will simply come true, and the hosepipe bans will keep arriving while the campuses keep humming. That trajectory is not destiny. It is the sum of architectural decisions, taken one planning permission at a time.
The alternative is to treat where intelligence lives as a design question with more than one answer. Push it outward instead of concentrating it. Put it on hardware the user owns. Right size it. Make it prove what it did. The demand for capable artificial intelligence is not going to fall, and it should not have to. What can change is whether meeting that demand requires emptying a reservoir to do it. Mickai exists to show that it does not.
Back at the village hall, the notice about the hosepipe ban is still pinned to the board. None of this reaches that parish in time to refill the reservoir this summer. But the deeper point is the one worth leaving with. The community was told, implicitly, that the water and the future had to be traded against each other, that intelligence at scale simply costs this much of the commons and that is that. It was never true. It was the shape of one architecture mistaken for the nature of the thing. Build it another way, on hardware people own, accountable by design, sized to the work, and the village keeps its water and its future both. That is not a smaller ambition. It is a more honest one.
About Mickai
Mickai is a Sovereign Intelligence Operating System, a SIOS, that runs fifty specialised artificial intelligence brains on hardware the operator owns, distributed and fully offline capable, from edge devices and workstations to its own server lineup led by the flagship Prometheus edge server. Consequential actions are sealed and signed in the Open Audit Record using the FIPS 204 ML-DSA-65 post-quantum standard, with a hash anchored to Bitcoin through Pantheon, Mickai's sovereign Bitcoin anchored Layer 1 (token PAN, fixed supply of five billion). Mickai is actively training and specialising its own models now. The approach is protected by 101 filed United Kingdom patent applications, around 2,234 claims, owned by Mickai LTD, with Micky Irons named as inventor. Mickai is the opposite of the hyperscale centralised cloud: intelligence you own, can prove, and can run without draining the commons.