Chips & Compute

Data Center Boom Faces a Bigger Risk Than Water or Power: Finance

Tech firms plan $750B in U.S. AI infrastructure by end-2026, yet economists say data centers must get 2.7x more productive to justify the spend — a bigger risk than water or power.

Data center boom faces challenges beyond water and power
Data center boom faces challenges beyond water and powerfrabre / pixabay
By Rebecca Stone5 min read

Why it matters

  • U.S. tech firms plan more than $750 billion in new infrastructure and equipment by the end of 2026, possibly exceeding $1 trillion annually by 2027 (Wachter research, June 2026).
  • Data centers must become about 2.7 times more productive to support an assumed 15% investment return, per economists Jonathan and Jessica Wachter.
  • The U.S. had more than 4,000 data centers totaling roughly 50 GW of capacity by end-2025; they consumed about 66 billion liters of water in 2023.
  • 69% of respondents in a recent NBC News poll said they would oppose an AI data center in their neighborhood.
  • Nvidia is one of the chief lenders financing AI data center expansion.

Major technology companies plan to spend more than $750 billion on new infrastructure and equipment in the United States by the end of 2026 — and, according to economists, that money only pays off if AI data centers become roughly 2.7 times more productive than they are today.

That figure comes from research published in June 2026 by economists Jonathan and Jessica Wachter, whose estimates frame a warning that experts are now raising above the familiar concerns about electricity and water: the AI data center construction boom carries significant financial risk, and communities could be left holding billions of dollars in partially completed infrastructure if demand for high-performance AI computing fails to meet expectations.

The scale of the buildout is already historic. According to a report published by BBC Science Focus, the United States had more than 4,000 data centers by the end of 2025, representing approximately 50 gigawatts of capacity. U.S. data centers consumed approximately 66 billion liters of water in 2023, a figure that has anchored public debate about the industry's environmental footprint.

But water and power, the experts told the BBC, are not the biggest problem.

What is the financial risk?

The core issue is that AI data centers are considerably more expensive than standard data centers, and the cost arrives up front. Companies must commit enormous sums — for buildings, power infrastructure, cooling systems and computing hardware — long before any AI service generates a single dollar of revenue.

Those investments assume AI services will generate sufficient revenue to justify the construction of massive computing facilities. There are no guarantees that revenue will show up.

The Wachters' research quantifies the gap. To support an assumed 15% investment return, data centers would need to become approximately 2.7 times more productive. Failure to achieve those productivity improvements could expose companies to significant financial difficulties — expensive facilities that do not generate enough revenue to cover their construction costs.

The trajectory of spending steepens from there. The same research estimates annual spending could exceed $1 trillion by 2027, up from the more than $750 billion planned for the United States by the end of 2026.

Who is lending the money?

How the infrastructure is being financed compounds the risk. Technology companies are increasingly relying on borrowed money rather than funding all of their expansion through existing profits.

One of the chief lenders in this financing activity is Nvidia — the same company that sells the graphics processors at the heart of AI computing. Nvidia's role as a financier ties hardware supply, credit and compute demand into a single chain. If AI adoption returns fall short of expectations, companies could be left with expensive facilities that do not generate enough revenue while still carrying substantial debt obligations.

Why are communities pushing back?

The financial questions arrive as local opposition to data center construction hardens. A recent NBC News poll cited by the publication found that 69% of respondents would oppose construction of an AI data center in their neighborhoods.

That sentiment showed up at the ballot box. In recent elections, pro-data center candidates lost badly, while candidates who publicly opposed data center construction did well.

The opposition persists even though data centers can deliver real economic benefits to host communities. Those benefits include increased property tax revenues, permanent employment and infrastructure improvements. Developers sometimes finance upgrades to electrical substations, roads and emergency services to accommodate their facilities.

The scale of those benefits is the point of contention. Data centers are designed to be self-operating and need minimal hands-on maintenance. The number of permanent jobs created by individual data centers is generally modest, often numbering only in the hundreds.

Dr. Ben Green, an assistant professor of information and public policy at the University of Michigan, told the BBC that communities hosting these facilities do not always receive benefits relative to the scale of the investments.

The bargain can also be lopsided in practice. Some developers have promised communities certain benefits in return for permission to build. A developer might contribute several million dollars to local fire departments while leaving other concerns — persistent equipment noise, for example, or environmental impacts — unresolved.

Could data centers become stranded assets?

The combination of three forces — uncertain AI demand, improving computing efficiency, and enormous capital requirements — raises the possibility that some planned data centers could become stranded or semi-abandoned assets.

Improving computing efficiency cuts in an unexpected direction. If future hardware and software deliver more computation per dollar and per watt, facilities designed around today's most demanding AI workloads might eventually run less demanding or less critical workloads, undermining the business case that justified their construction.

The industry's central challenge, the experts suggest, may therefore be less about whether companies can build enough data centers and more about whether the computing capacity they are constructing will remain economically viable over the lifetime of those facilities. A building financed over decades must keep earning through years of rapid change in chip performance, model efficiency and AI demand.

The stakes extend beyond the technology sector's balance sheets. Communities that approved construction in exchange for tax revenue and jobs could inherit partially completed infrastructure worth billions of dollars if projects stall. Lenders exposed to the sector, Nvidia among them, would carry the credit risk. And local governments that budgeted around projected property tax gains would face shortfalls.

What happens next?

The numbers that decide the outcome are already on the calendar. Spending above $750 billion by the end of 2026, a possible $1 trillion annual pace by 2027, and the Wachters' 2.7-times productivity requirement give the industry concrete thresholds to meet — or miss. Whether AI services generate the revenue those figures assume will determine whether the 4,000-plus U.S. data centers built so far mark the foundation of a durable industry or the high-water mark of an overextended one.

Source: networkworld.com

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. Voters Approved 40 of 41 Ballot Measures Restricting AI Data Centers
  2. AI Spending May Hit $3T–$4T While Compute Stays Sold Out
  3. Chinese data center builders chase the U.S. AI infrastructure boom
  4. Google's Emissions Jumped 50% in 2024 to Feed Generative AI
  5. Startup Drogue raises $20M to power floating data centers with ocean heat

« Previous articleNext article »