AI infrastructure growth is running into limits on energy, water, and land, according to researchers at Forrester, which predicts governments will increasingly make datacenter operators pay for the resources they consume.
Governments, utilities, datacenter operators, and enterprises will have to rethink how that growth is funded, managed, and regulated, the firm said in its 2027 sustainability predictions.
The scale of the infrastructure buildout needed to support AI across businesses, governments, and consumer services is beginning to emerge. Earlier this month, PwC put cumulative global datacenter capital expenditure through 2050 at between $22 trillion and $50 trillion, with a central estimate of $31.6 trillion – roughly equivalent to a year of US economic output.
Whether grids and communities can support those ambitions is another matter. "Unlimited AI ambition faces a reckoning with limited resources," Forrester said.
Forrester predicts that governments and utilities will require datacenter operators to finance grid upgrades, pay for reserved capacity, and provide guarantees against speculative demand. It also expects community impact reviews to become a gate that new projects must pass before construction.
"With unprecedented growth demands, utilities cannot assume that every proposed datacenter will arrive on schedule or consume its promised load. Building generation and transmission for speculative demand could leave households and businesses paying for stranded infrastructure.
"Already, the White House has called on hyperscalers to procure their own power, fund upgrades, and pay for reserved capacity even when unused. In 2027, at least two countries will turn this ratepayer-protection principle into tariffs or equivalent obligations. Cloud and colocation costs will become more location-specific, contractually committed, and exposed to power-market risk."
Sunil said datacenter operators should expect long-term minimum payment commitments, financial guarantees, energy tariffs, and requirements to reduce their loads when grids are under strain.
"Cloud economics will consequently become more dependent on location, electricity contracts, grid maturity, and providers' ability to generate or reduce power. Enterprises rapidly scaling AI capacity will increasingly inherit these costs," he said.
The US, however, appears to be moving in the opposite direction on some forms of environmental oversight.
In August, the US Environmental Protection Agency (EPA) proposed eliminating federal minimum requirements for public participation in minor-source air permits. The change could make it easier for some datacenter power projects to proceed without public notice, even as Forrester predicts that community scrutiny will increasingly determine which facilities get built. ®