Investors are ignoring the biggest source of data-center water use

How to meet the power demands of AI data centers is, in many ways, the question at the intersection of sustainability and economic growth right now.

But whether the plants generating that power will have enough water is another question, and it’s an underaccounted risk.

According to an analysis by Ceres, a nonprofit, and projections from Bluefield Research, a market research firm specializing in water issues, most of a data center’s water footprint is tied not to cooling but to electricity usage — and there is a meaningful gap in disclosure on indirect water use.

What the lenders financing data centers — the private credit firms now handling 70 to 90 percent of individual project-level borrowing for new construction — do with water consumption data is a separate question that is rarely asked.

The water behind the power

Most coal, natural gas or nuclear power plants heat water into steam to drive a turbine, then use more water to cool the steam, condensing it back into water. As a result, a data center buying electricity from such plants consumes water indirectly, even if its own cooling system consumes little or no water.

Janus Henderson, an asset management firm, cites Lawrence Berkeley National Laboratory research showing that indirect consumption is up to 75 percent of a facility’s total water footprint. Meta, one of the few operators that has published such an estimate, reported an indirect figure for 2024 that was more than 20 times the water its data centers consumed directly.

That water doesn’t appear directly in an electricity bill, and lenders assess power the way they assess other purchased services: cost, reliability and availability. Among the issues these deals hinge on — tenant credit, lease terms, power delivery, queue position — water at the generating plant isn’t usually a consideration. A risk nobody prices at the financing stage will manifest somewhere later — as delay, curtailment or a cost passed down the contract.

The case for not looking

From a financial perspective, letting the water utility assume risk has a reasonable defense: The exposure sits with a regulated company carrying its own water-planning duties, which credit rating agencies already scrutinize.

There is also no settled way to measure or report a data center’s indirect water footprint. Amazon tracks indirect water consumption but has not published figures, citing an absence of industry standards, and Google and Microsoft would not confirm whether they track this. A risk with no standard measure is hard to write into the loan documents between a lender and a data center developer, and a risk that never enters those documents never gets priced into the cost of the debt.

The vulnerability lies in the time horizon. While a contract locks in terms for decades, river flows, drought conditions and groundwater levels can change dramatically from one year to the next, let alone over the long term.

Some states are taking notice. In Texas, where data centers represent nearly 90 percent of new grid power requests, Governor Greg Abbott ordered a freeze on interconnection approvals until developers can verify their water use reporting and ensure grid sustainability. But Texas is still the exception to the rule.

The water that lenders do count

The divide between the water liabilities that lenders evaluate and those they leave to utilities becomes much clearer at the facility level.

Water diligence on a data center deal is more thorough than some critics assume. Lenders commonly require closed-loop or dry cooling, which draw little or no municipal water. They write water efficiency targets into loan covenants and track them. They check water rights and permits before construction, because a contested groundwater permit means delay, which financiers and developers loathe.

The data center projects cited as water failures are the ones this process exists to prevent, such as Google’s halting of a campus in Chile after a court ruled against its groundwater use.

But the less-water-intensive remedies carry their own water costs. A dry-cooled data center uses more electricity than a conventional facility, and generating that electricity consumes yet more water at the power plant. The water leaves the utility plant and reappears upstream, in a place the lender’s diligence may never check.

Where that leaves your company

The mismatch between how carefully carbon is counted and how little water is scrutinized has a structural cause. Carbon accounting has two decades of protocol, scope definitions and assurance behind it, which is why a sustainability team can produce a defensible Scope 3 figure for its cloud use. 

The water consumed generating that cloud provider’s electricity does not yet have an equivalent apparatus, which leaves it uncounted on the corporate side and unpriced on the lending side. That may not hold indefinitely. An initiative led by sustainability standards organization SCS Global Services, World Resources Institute, World Wildlife Fund and the CEO Water Mandate is developing a scopes-based water accounting framework — the closest thing yet to what carbon has had for two decades. 

With neither the lender nor the accounting framework asking, the question falls to whoever is buying the compute — not only the companies leasing colocation capacity, like cloud providers or their model developer customers, but also the enterprises signing cloud or AI service agreements.

The lack of formal standards notwithstanding, sustainability teams negotiating power or hosting agreements can ask whether water availability at the generating plant was assessed. The answer tells them whether the capacity they’re contracting for rests on a resource anyone has checked. A dry-cooled campus in a drought year still depends on electricity from a generating plant that needs a river, and whether that river is available in year 12 of a 15-year lease should not be left to chance.