Mineral, Computing and Water

Water and climate series · Part 3 of 3

1. The mountain is the refuge · 2. Water is the condition · 3. Mineral, compute and water

Two new demands are arriving on Bolivian water at the same time. One is called critical minerals. The other is called artificial intelligence. Both are presented as a historic opportunity, and both are paid for with the same resource that was already in trouble in the previous parts.

Part 1 · Part 2

First, how many minerals

The figure changes depending on who says it, and that variation is itself a finding.

The map we built at sala.red, cross-referencing the US strategic list against the national geological registry, places 47 critical minerals in Bolivian territory. Only Beni and Pando have no record.

Other counts give different numbers because they start from different lists. None of them rests on a published national cadastre, because none exists.

Negotiations are running on an inventory the state does not publish.

On 27 April 2026, Mining Minister Marco Antonio Calderón de la Barca signed a memorandum of understanding on critical minerals in Washington with deputy secretary Caleb Orr. Bolivia joined the bloc of fifty countries the United States is courting to counter China. The zones of interest are Santa Cruz, Cochabamba and Potosí, and in the Chiquitania Indigenous peoples have denounced the absence of prior consultation.

The full text of the memorandum is not public. What is known comes from a State Department release, a post by the US embassy on X, and press coverage. There is no published Bolivian official version, no treatment in the Legislative Assembly, and no citizen access to the content.

It is signed over an inventory that is not published, and reported through a document that is not either.

Lithium is paid for in water

Here the numbers do exist, thanks to work by researcher Gonzalo Mondaca of CEDIB on the contracts signed with the Chinese consortium and with Uranium One Group.

The promise of direct lithium extraction was to use less fresh water. The contracts say otherwise. The Russian contract sets a demand of 1.8 million cubic metres a year. The two Chinese contracts add up to 5.1 million cubic metres a year.

Mondaca calculated that, if executed, those contracts would demand up to five times more fresh water than the state company YLB uses, and fifteen times more than the rain that falls over the Uyuni salt flat and its surroundings.

The reference figure he uses comes from a 2007 water balance by hydrologist Jorge Molina Carpio. And there the real problem appears: nobody knows how the groundwater of the micro-catchments surrounding the salt flat connects to the salt flat itself. Industrial-scale extraction is being signed over a hydrogeological system nobody has characterised.

Brine volumes are larger still. At full capacity YLB would use twenty million cubic metres a year to produce fifteen thousand tonnes of lithium carbonate. The Chinese and Russian operators would use around nine million each for similar or larger output.

And now compute

The second demand does not arrive with pickaxes and trucks. It arrives with fibre optics and the word hub.

The global numbers are no longer abstract. Direct water consumption by data centres reached around 560 billion litres in 2025. A 2026 United Nations report projects that if artificial intelligence rises to forty percent of data centre electricity consumption by 2030, associated water consumption would reach 9.3 trillion litres, enough to cover the annual basic domestic water needs of more than 1.3 billion people in Sub-Saharan Africa.

Google reported 10.9 billion gallons consumed in 2025, up 34 percent year on year and more than double its 2021 level.

Where AI's water is actually billed

This is the point almost nobody explains, and the one that matters most to Bolivia.

In June 2026 Nvidia presented a closed-loop cooling design that cuts on-site water consumption to near zero. It solves the four percent. The other ninety-six percent remains, split between the plant generating the electricity and the mine producing the mineral.

Two different things share the same name here, and they need separating.

A national data centre stores and processes Bolivian information: public records, health, education, backups for local companies. It is useful public infrastructure and it is what was announced for La Paz and Cochabamba. It consumes water and electricity, at a scale of tens of megawatts.

A hyperscale data centre trains and runs artificial intelligence models for the world. It consumes hundreds of megawatts and its location is decided by submarine cables, latency and energy matrix. That one is not coming to Bolivia, and Ferrere's own analysis explains why.

Bolivia can have the data centre that serves Bolivia. What it will not have is the one that drives the AI economy. From that chain it gets the mine and the power generation, which is where ninety-six percent of the water is billed.

Both consume water from the same country. The difference is who they work for.

The regional precedent is already signed. Paraguay and Taiwan are building a binational AI data centre with two hundred million dollars of investment and ten megawatts of capacity, backed by the Trump administration. The split is explicit: Taiwan supplies the Nvidia chips, Paraguay supplies low-cost electricity and basic infrastructure.

Bolivia already has the announcement. Under Supreme Decree 5509, on low-orbit satellites, the president announced that the government intends to build data centres in the cities of La Paz and Cochabamba.

The law firm Ferrere analysed that prospect in May 2026, writing for foreign investors. It estimates that a one megawatt data centre can use up to 25.5 million litres of water per year. And it concludes that it is unrealistic to think departments with limited water resources, such as Cochabamba, could host large-scale data centres for international technology companies. It adds that even proposing projects of this kind could generate social conflict in a region historically marked by episodes such as the Water War.

The government announces data centres in Cochabamba. The lawyers of foreign investment warn that the water is not there and that it would blow up socially.

The incentive machinery, meanwhile, comes and goes. Supreme Decree 5503 created the Extraordinary Regime for the Promotion and Protection of Strategic Investments, with reinforced legal and tax stability for up to fifteen years, approval of investment contracts by supreme decree, and a fast-track procedure with a thirty working day maximum applying positive administrative silence.

That decree was repealed under pressure from social groups. Ferrere's analysis suggests conditions like these could return if the Legislative Assembly passes equivalent rules. Which means the discussion is not closed. It is waiting.

And one fact makes the whole scenario fragile. Around sixty percent of Bolivia's energy matrix depends on natural gas, and some projections point to supply problems within three or four years and reserve depletion in roughly a decade. For scale: the data centre Google is building in Uruguay will require up to 420 GWh a year, equivalent to five percent of that country's energy consumption.

Raw material is not sovereignty

Bolivia has the lithium. It does not have the cell factory. It has the tin and the zinc. It does not have the smelter that separates gallium. It will have the energy feeding somebody else's compute. It will not have the compute.

Potosí exported silver and bought cloth. Today we export brine and buy batteries. The geology changed. The position in the chain did not.

With raw material you supply. With infrastructure you decide. Bolivia has spent five hundred years on the first side.

Technological sovereignty means measurable things. Refining here. Compute installed in the country, with people trained to use it. Bolivian data under Bolivian law. Models trained on our languages, with licences decided by the communities that speak them. None of that appears as a side effect of an extraction contract. It gets negotiated and written into the paper.

The royalty percentage goes up and the position stays exactly where it was. What changes it is keeping a share of the capacity.

And there water is the card we are not playing. It is scarce, it is local, it cannot be imported. Whoever needs it has to come here for it. That is bargaining power, and right now it is given away for free.

The volumes, side by side

Pick a use and compare it with what Cochabamba delivers to people.

The comparison that weighs most: the two Chinese lithium contracts, in fresh water alone, equal the annual domestic consumption of more than ninety thousand people. A hundred megawatt data centre weighs half of that, before counting the water used by the plant powering it.

The three traps

One. Water is billed where nobody is watching

The salt flat and the southern altiplano have little population and no permanent press coverage. Consumption figures come from independent researchers, not from mandatory public reporting. The same happens with data centres worldwide: water use is disclosed as a company-wide aggregate, almost never per site, which masks the impact on the specific watershed.

Two. Efficiency gets announced and volume keeps growing

DLE promised less fresh water and the contracts ask for more. Closed-loop cooling solves four percent of the problem and is communicated as though it solved all of it. When efficiency per unit improves and the number of units multiplies, total consumption rises anyway.

Three. Water is not fungible

A litre replenished in a Scottish catchment does not offset a litre evaporated in Arizona. Nor does a litre returned in Santa Cruz offset a litre pulled from the Uyuni aquifer. Corporate water replenishment programmes rest precisely on treating water as interchangeable, and it is not.

What to do now

1. Publish the signed agreements. The April 2026 memorandum has no public text in Bolivia. Full publication of resource memoranda and contracts, with treatment in the Assembly, before they take effect.immediate
2. Publish the full mineral inventory. Three different figures circulating means the negotiation is running without a cadastre. A public registry, with location and estimated volume, before signing more memoranda.before 2027
3. Updated water balance for the southern altiplano. The reference figure is from 2007 and nobody knows how the micro-catchments connect to the salt flat. Without it, any extraction contract is signed blind.start this year
4. Mandatory per-site water footprint. Cubic metres per year, exact source and affected catchment, published quarterly. Applies equally to lithium plants and to any data centre that lands.before 2028
5. Real prior consultation in the Chiquitania and the altiplano. Indigenous peoples have already denounced its absence. Conflict arrives when the agreement is negotiated afterwards.immediate
6. No positive administrative silence for water-consuming projects. Fast track for commercial permits, fine. For a facility that commits a catchment for twenty years, no.when legislating
7. Price water as an industrial input. If water is free or nearly free, efficiency never arrives. Price is the only signal that changes how a plant is designed.before 2028
8. Negotiate capacity, not just rent. If Bolivia supplies mineral, energy and water, the counterpart is measured in installed capacity: refining in the country, compute accessible to universities and communities, technical transfer with written deadlines. Paraguay negotiated chips for energy. That is the floor.in every agreement
9. Tie every concession to the water works from Part 2. Networks, leaks, well cadastre, amunas and qochas. If the mineral leaves, a fraction returns as water infrastructure, written into the contract.in every contract

What is really at stake

Part 1 showed that the mountain protects us from heat. Part 2, that water is already short through infrastructure before climate. This part adds the third element: two global industries arriving to bill water in the same territory, in the same decade, while peak water keeps declining.

The argument in favour is real. Bolivia needs foreign currency, and lithium and critical minerals are the strongest card it holds. Nobody serious proposes leaving them underground.

The question is what comes back in exchange and who pays the water bill. A contract that hands over mineral and water in return for royalties is the same deal Potosí got in the sixteenth century, with better pumping technology. We remain the resource.

Artificial intelligence presents itself as something that happens in the cloud. It happens in a salt flat, in a power plant and in an aquifer nobody measured. Bolivia is about to host both ends of that chain and none of the capacities in the middle. The decision about the terms is taken now, and paid in water for the next fifty years.