Key reasons to read this article
- The next AI bottleneck has little to do with AI itself, and it could reshape whether the industry can grow.
- One country’s abundant solar energy could prove almost useless for AI if another vital resource is missing.
- A data center can bring billions in investment, but leave someone else to pay for the power and water it consumes.
- Development banks are beginning to look at the infrastructure behind AI. What they do next could matter far beyond the tech sector.
The AI boom is hitting a new set of limitations. For the past few years, the binding constraint on AI’s growth has involved chips and computing power. That constraint has not disappeared, but it is being overshadowed by others: electricity and water supplies, grid capacity, pollution, permitting timelines, and public tolerance.
We are already seeing the effects. In June 2026, New York enacted a one-year restriction on new environmental licenses for hyperscale data centers above 20 MW. The main reason for this was the energy and water consumption involved. A hyperscale AI data center can use up to 5 million gallons of water a day, comparable to the daily consumption of a town of 10,000 to 50,000 people, and its electricity demand can rival that of heavy industry. For communities and governments, the issue is therefore no longer whether a data center brings investment and jobs, but whether local power, water and other infrastructure can absorb its demands.
These pressures are creating tensions within communities. Residents have raised concerns about rising electricity costs, increased noise levels, worsening air quality, and high water usage, while governments are assessing whether the existing infrastructure can support further expansion. The Ypsilanti Community Utilities Authority in Michigan, for example, has imposed a one-year embargo on water supplies and sewerage services to data centers, which is affecting projects involving the Los Alamos National Laboratory and the University of Michigan.
The issue is now becoming greater than all the individual projects. The AI race is entering a phase governed less by export controls on chips and more by the physical capacity of power grids, the availability of water and the politics of local infrastructure.
Why do data centers have such high energy and water requirements?
The obvious answer is heat. Heavy computing generates significant amounts of heat, and cooling systems are essential to keep servers operational. But power and water are not interchangeable constraints. A site may have access to an abundant electricity supply but still be unsuitable for the building of a large facility if it lacks sufficient water or an appropriate cooling system, or vice versa.
According to an infrastructure specialist from the Digital Impact Alliance, who preferred to remain anonymous, Chad has substantial solar potential but an insufficient water supply to cool a large facility, making it what he referred to as a “net negative” site despite scoring well on energy alone.
So, the question is no longer where cheap electricity or renewable energy can be accessed. It is where power, water, connectivity, land, and regulation can work together without creating infrastructure costs that others will be forced to absorb.
AI’s geography could be changing
There is no ideal location for a data center. Developers must take into account a combination of several variables at the same time.
🔹 Water and climate matter. A recent UK report warned that plans to expand data-center capacity could come under pressure due to future water shortages. This is far from a unique situation. Consequently, cooler climates, abundant water supplies and advanced water-reuse technologies could become increasingly valuable to AI hubs.
🔹 Reliable, low-carbon electricity matters just as much. AI centers run 24/7, making a stable electricity supply essential. A region may have huge solar energy potential, but weak or fossil-powered grids can undermine that advantage. The World Bank has similarly highlighted that a stable electricity supply is as vital as data networks.
🔹 Regulatory predictability is another competitive factor. Investors can work to strict rules when these are clear and consistent. What creates risk is uncertainty over permits, costs, or changing regulations. For investors, predictability can matter as much as the availability of resources.
🔹 Talent and digital infrastructure can also shape the geography of AI. India illustrates this particularly well; it does not lead on power reliability or grid headroom, but the scale of its available software and technical talent has led to it becoming a hub for the operational and engineering work AI infrastructure needs.
🔹 Finally, political stability and data sovereignty can override other resource metrics. The Digital Impact Alliance specialist highlighted concerns that nations like Nigeria require domestic data storage capability to prevent external access issues associated with U.S. cloud providers. Brazil exemplifies how various factors, including renewable energy and policy, can influence data-center investments. Ultimately, a deficiency in any variable, particularly sovereignty, can jeopardize the overall competitiveness of a site.
Who pays for the AI infrastructure?
Historically, development finance institutions have funded infrastructure such as roads, ports, and broadband networks for their broader economic value, not because the assets themselves were the most profitable investment opportunities available.
AI infrastructure is starting to see a similar trajectory. Transmission networks, renewable generation, water recycling, digital public infrastructure, and workforce development are the systems that make AI possible and these could all attract investment without being AI companies themselves. Development banks are beginning to explore whether some of this enabling infrastructure could be financed in much the same way broadband infrastructure once was.
In May 2026, the World Bank Group published a coordinated framework assessing infrastructure, policy, risk, and financing conditions across 15 priority countries, identifying where lending by the International Bank for Reconstruction and Development, financing from the International Finance Corporation (IFC), and Multinational Investment Guarantee Agency guarantees could support cloud and data-center markets.
In the same month, the IFC published a handbook separating AI infrastructure from AI companies to enable funding to be channeled towards power, water, and digital foundations rather than the companies that will use them. This approach echoes the logic of broadband-era lending: support the enabling infrastructure and allow private capital to compete on top of it.
Why would development banks support this? According to financial advisor Seun Akintola, the same drivers that justified broadband financing apply to AI infrastructure. Power, connectivity and other enabling systems can help countries to participate in a fast-growing part of the global economy, while generating economic benefits that extend beyond individual infrastructure projects. That makes them difficult to finance through private capital alone, particularly when the returns are long-term and uncertain.
But this shift should not be exaggerated. Asked whether development banks are actually moving toward financing AI infrastructure, the Digital Impact Alliance specialist said he was not aware of any evidence confirming such a shift. That statement matters. Yes, institutions are exploring the opportunity but whether that exploration becomes a major financing trend remains an open question.
Can countries attract AI investment without adversely impacting communities?
Governments want the jobs and prestige that AI investments offer. In turn, investors demand credible environmental and governance standards. Communities, meanwhile, want affordable electricity and a secure water supply. But the underlying issue is simple: if AI creates enormous demand for electricity and water supplies and new infrastructure, who should pay for the capacity it requires, and who should benefit from it?
Oregon offers one possible policy response. This U.S. state requires big data centers to pay for grid upgrades through higher tariffs, rather than offloading any costs onto households.
Microsoft President Brad Smith has emphasized that infrastructure success relies on community benefits, indicating that ignoring the needs of a community is no longer acceptable when hoping to attract AI investment.
Where does AI infrastructure go next?
The broadband parallel offers the clearest lens for this transition. Development finance historically funded digital networks not merely for direct profitability, but because connectivity became a prerequisite for wider economic participation. AI could follow a similar path.
The IFC handbook codifies this logic by separating AI-enabling infrastructure from the AI companies that use it. Yet this remains an emerging financing model rather than an established approach.
Ultimately, the next leg of the AI race may therefore be decided less by algorithmic supremacy and more by energy and water security, grid capacity, and political will. For governments and development financiers, that could make AI infrastructure the next major test of whether public and private capital can build the physical foundations of a new digital economy without shifting its environmental and financial costs onto the communities that host it.