The oldest water war on record produced the oldest arbitration, the oldest boundary marker, and the oldest treaty. None of it produced peace. It produced something else, and that is the thing the newest disputants are doing without.
There is a limestone slab in the Louvre, carved around 2460 BCE, called the Stele of the Vultures. It shows soldiers of the city-state of Lagash marching in phalanx over the bodies of their neighbors from Umma, while vultures carry off the losers' heads. It commemorates the earliest well-documented war between states in human history, and the war was about water: a border watercourse and an irrigated strip of land called Gu-Edin, which Umma, sitting upstream in the canal system, kept taking from Lagash, sitting downstream.
Here is the detail that should interest you more than the vultures. About a hundred and fifty years before that stele was carved, this exact dispute had already been settled. A king named Mesilim, ruler of Kish, a third city with no stake in the fight, was brought in as a neutral arbiter around 2600 BCE. He heard the claims, fixed the boundary between Umma and Lagash, and marked it with a stele planted in the ground. A 2018 paper calls the result "Mesopotamia 2550 B.C.: The Earliest Boundary Water Treaty." So the oldest recorded water war also produced the oldest recorded third-party arbitration, the oldest boundary marker, and the oldest treaty. Every institutional move a modern negotiator would recognize, the neutral adjudicator, the published boundary, the written settlement, was invented for this specific quarrel, essentially immediately.
And it did not work. Umma breached the settlement, repeatedly, across generations. The conflict grinds through the whole Early Dynastic III period, roughly 2600 to 2350 BCE. The arbitration failed, the boundary was violated, the stele of the treaty was answered a century and a half later by the stele of the vultures. If the claim of this essay were "ancient water institutions solved the problem and we should copy them," the essay would be dead in its second paragraph.
The claim is different, and it is the thing the ancient story actually proves. Those institutions did not produce peace. They produced legibility. Because there was a named arbiter, a marked boundary, and a recorded claim, we can still read who did what to whom, forty-five centuries later, from the losing and winning parties' own documents. The dispute was written down into the world. Hold that thought while I tell you about Oregon.
The city that sued the newspaper
In 2021, the city of The Dalles, Oregon, went to court to keep a number secret. The number was how much municipal water Google's data centers in town consume, and the suit was filed against The Oregonian, the newspaper whose reporters had requested the records. The city was carrying the argument on Google's behalf, and the legal theory is worth stating carefully, because it is the purest artifact this whole subject has produced: the company argued that its water consumption was a trade secret, since competitors might infer from it how Google cools its servers.
Thirteen months later, in December 2022, the city dropped the case. Note the verb. It did not lose; nothing was ruled. It abandoned the suit, released ten years of data covering 2012 through 2021, agreed to disclose annually from then on, and paid $53,000 toward the legal costs of the Reporters Committee for the Freedom of the Press. The number, once public, was 355.1 million gallons in 2021, which local reporting translated as enough to cover the city's seven square miles three inches deep.
Set the two stories side by side and the comparison is not flattering to us. In 2550 BCE, a water dispute produced a named arbiter applying a stated boundary, recorded durably enough to outlast the civilizations involved. In 2022 CE, a water dispute produced no ruling, no precedent, a withdrawn lawsuit, and a records release that exists only because journalists were willing to be sued and a nonprofit paid lawyers. The next town that wants to know what its own water system has promised a data center starts from zero, because nothing was adjudicated and the operative theory on the other side is still that the answer is proprietary. Four and a half thousand years on, the argument has become less documented, not more. That is the sentence this essay exists to put in front of you.
The toolkit nobody is using
The strange part is that the toolkit for this exact problem is not lost knowledge. It is working law, refined over centuries, and in the American West it is brutally specific.
Western states allocate water by prior appropriation: first in time, first in right. Every water right has a date, and in a shortage the state cuts off the most junior rights first, working backward up the seniority ladder until supply and demand meet. Farms with 1890 rights water their fields while subdivisions with 1990 rights go dry, and everyone knows this in advance, because the dates are published and the rule is mechanical. Eastern states run the older riparian doctrine instead, where those adjacent to a watercourse may make reasonable use of it, and courts referee what reasonable means. You can argue about which system is wiser. What you cannot argue is that either is secret: the whole design of both is that claims are known, ranked or reviewable, and enforced by a public process.
Now place the AI buildout inside that frame. A hyperscale data center is reported to draw one to five million gallons a day, comparable to a small city. And in any prior-appropriation state, a facility built in the 2020s is, by construction, the most junior claimant on the river. The essay's third old virtue, seniority that cannot be bought retroactively, is not a moral flourish here. It is the operative statute. When the shortage comes, and in the Colorado basin it is not hypothetical, the newest straws come out of the water first. Anyone financing a twenty-year facility on a junior right is carrying curtailment risk on their balance sheet whether they have priced it or not, and the water-rights date is public even when the consumption figure is a fought-over secret.
The disputes, meanwhile, are already running. Arizona communities are contesting data-center growth as supply cuts loom over the state. In California's Imperial County, a developer went to court in June 2026 over a county freeze on data-center development. These are separate fights in separate states under separate postures, and I name them separately on purpose; the pattern they share is only this: each is being negotiated ad hoc, largely in private, deal by deal, as if no one had ever allocated scarce water before.
What a thousand years of Thursdays teaches
The commission that prompted this essay proposed three virtues from the old institutions: adjudication in public, allocation by published rule, and seniority that cannot be bought retroactively. I want to complicate the list with the most charming counterexample available, because it sharpens the point rather than blunting it.
Every Thursday at noon, at the Door of the Apostles of the Valencia cathedral, the Tribunal de las Aguas convenes in the open air, as it has for roughly a thousand years. It hears disputes among the irrigators of the Valencian plain. The judges are farmers elected by the canal communities. The proceedings are entirely oral. There are no lawyers, no filings, and, this is the part to sit with, no written records at all. It is commonly described as the oldest functioning court in Europe, and it keeps no archive of a single case it has ever heard.
So the model institution for public adjudication flatly fails the published-record criterion, and it has thrived for a millennium anyway. Why? Because publicity and documentation are two different mechanisms for the same underlying good, and they can substitute for each other. Valencia's transparency is not that anyone writes it down; it is that everyone watches. The parties are neighbors, the judges are peers who farm the same canals, and the sanction is your standing in a community that will reconvene, in public, next Thursday, forever. Mesopotamia had the other mechanism: no weekly assembly of neighbors, but a stone in the ground and a scribe. Legibility by record, or legibility by audience. Every durable water institution in this story has at least one.
And that is the precise indictment of the current moment. A data-center water negotiation, as actually practiced, has neither. It is not conducted in public, and its record, where one exists, is wrapped in nondisclosure agreements and, in the case actually litigated, a trade-secret theory, so that the documentation surfaces only when a newspaper spends thirteen months being sued. Not the record without the audience, not the audience without the record. Neither. Whatever you think of oral tribunals or bronze-age arbitration, they clear a bar the present arrangements do not.
The one where the institution worked
Lest this read as nostalgia, the strongest modern case is recent, and it is the thesis operating correctly. In February 2024, a Chilean environmental court partially reversed the permit for a $200 million Google data center in Cerrillos, in Santiago, requiring the application to be reassessed for climate effects amid real pressure on the capital's aquifer. The project as first approved would have drawn about 7.6 million liters of potable water per day, around two million gallons. Google paused the project and announced it would change the cooling system, with a new process starting from scratch.
Look at the anatomy: a public body, applying a published standard, in an open process, revised a specific permit, and the engineering changed in response. Nobody needed a leak or a lawsuit against journalists. The company was not destroyed; it redesigned, which is what companies are good at when the constraint is legible. That is Mesilim's move running on modern rails, and it happened not in the jurisdictions that write the most about water law but in Santiago.
I should say plainly what this essay is and is not claiming, because the policy genre here is crowded. That data centers strain water systems, that prior appropriation matters, that disclosure is deficient: all of this is established commentary, and there is a live rival remedy worth taking seriously, the argument, made at length in Reason among others, that water markets price scarcity better than administrative allocation. Fine. But notice that a market presupposes the very layer this essay is about: you cannot trade rights that are undefined, unranked, and undisclosed, and a price negotiated under NDA between one buyer and one city is not a market, it is a bilateral secret. Markets, courts, tribunals, and arbiters are competing mechanisms on top of a shared foundation, which is that the claims be legible. The contribution here is only the long view: nobody in the current debate seems to have noticed that the foundation is four and a half millennia old, that it was invented for the first water war on record, and that the newest disputants are the first in the whole lineage to try operating without it.
What to do with this
If you build, site, or approve compute, four portable conclusions.
First, publish your water numbers before someone sues for them. The Dalles fight bought thirteen months of secrecy, cost real money, generated a decade of disclosure anyway, and left the trade-secret theory tainted for the next negotiation. Annual disclosure is now the settlement's own norm. Voluntary disclosure is cheaper than litigated disclosure in every currency that matters, including the one this industry runs on, which is permission to build the next one.
Second, know your seniority date, because the river does. In a prior-appropriation state your facility is the most junior right on the system, first curtailed in shortage, and no amount of capex changes the date. Engineer for it: Chile's lesson is that cooling design is a variable, not a given, and a redesign forced by a court in public cost Google a delay, not the project.
Third, prefer the public process even when the private one is offered. A permit won in an open proceeding, like Cerrillos, is a settled fact that survives scrutiny; a sweetheart arrangement in an NDA is a liability with a countdown attached, waiting for its own Oregonian. The old institutions are not slower than the new secrecy. The Dalles took thirteen months to produce nothing binding.
And fourth, apply the stele test. Mesilim's boundary stone was legible to strangers four and a half thousand years later. The Valencia tribunal is legible to anyone standing in the plaza on a Thursday. Ask of any water arrangement your organization signs: could an outsider, later, reconstruct who agreed to what, under which rule? If the answer is no, you have not settled a dispute. You have scheduled one.
Sources
The Umma-Lagash conflict, Mesilim's arbitration, and the Stele of the Vultures (c. 2460 BCE, Louvre) per standard treatments of Early Dynastic Mesopotamia, with the treaty framing from "Mesopotamia 2550 B.C.: The Earliest Boundary Water Treaty" (2018), which identifies Mesilim of Kish as external arbiter and the Gu-edena valley as the disputed ground. Mesilim's date is given variously in the literature; this piece uses "around 2600 BCE" and the cited paper's own title places the treaty at 2550 B.C., so the gap to the stele is approximate, on the order of a century and a half.
The Dalles: suit filed 2021 against The Oregonian on trade-secret grounds and dropped in December 2022 after thirteen months, with ten years of data (2012-2021) released, annual disclosure agreed going forward, $53,000 paid toward the costs of the Reporters Committee for Freedom of the Press, and 355.1 million gallons consumed in 2021. Per Courthouse News, the Reporters Committee, KGW, and Data Center Dynamics. The "seven square miles three inches deep" comparison is local reporting's, repeated here as theirs.
Cerrillos, Santiago: February 2024 partial reversal of the permit for a $200 million Google data center, with the cooling towers' requirement of 7.6 million liters (about two million gallons) of potable water per day surfaced in the proceedings, and a subsequent design change toward air cooling. Per Data Center Dynamics and February 2024 wire reporting.
Prior appropriation and riparian doctrine per standard water-law references; the hyperscale draw of one to five million gallons a day is a reported range, not a measurement of any specific facility.
The Tribunal de las Aguas per tribunaldelasaguas.org and general references: Thursdays at noon at the Door of the Apostles, judges elected by the canal communities, oral procedure, no written case records, roughly a thousand years.
Imperial County, California: a developer went to court in June 2026 over the county's freeze on data-center development, per June 2026 reporting.
The water-markets counterargument per Reason (March 2026), "The data center water panic has a better answer: Water markets."
The stele test, for systems: could an outsider, later, reconstruct who agreed to what, under which rule?
That question is the whole of provenance, and it is the one an agent system fails by default. Chain of Consciousness is the machinery for passing it: a tamper-evident record of what a system decided and on what basis, written at the moment of the decision rather than reconstructed afterward under pressure, and readable by someone who was not in the room and does not have to trust you. A boundary stone, for things that do not last four thousand years on their own.
pip install chain-of-consciousness · npm install chain-of-consciousness
Or the whole stack, provenance and ratings and verification together: pip install agent-trust-stack / npm install agent-trust-stack.
Top comments (0)