Why American Data Centres Are Unplugging From the Grid

The race to build AI infrastructure is exposing the limits of the US power system and pushing data centre developers towards off‑grid generation as a survival strategy.

Global investment in data centres is forecast to reach US$6.7tn by 2030, with about US$2.7tn of that in the US alone, and much of the new load is heading for already stressed grids such as Texas.

AI is changing the physics of digital infrastructure faster than the grid can adapt.

Traditional server racks once drew 5kW to 10kW, but AI‑specialised racks have surged to 50kW to 100kW each in the past two years, with power demand expected to keep climbing as new chips arrive.

Jon Clark, Associate Director at consultancy Gleeds, argues that the balance of power has shifted away from the long‑time cloud giants towards AI specialists rolling out ever more energy‑intensive hardware.

"The chips are driving the power, the cooling, the infrastructure, the sites, where the availability is, where the customers need to be, where the customers are then buying that technology and where they need to be placed," he explains.

In Texas, the Electric Reliability Council of Texas (ERCOT) has warned that data centre demand alone could reach around 78GW by 2031 on an unadjusted basis, equivalent to roughly 36% of state-wide load.

Queues, bottlenecks and the appeal of going off grid

Against that backdrop, interconnection queues have become a gating factor on AI growth rather than a mere procedural hurdle.

Developers now routinely face multi‑year waits to connect major new campuses as large‑load requests pile up in ERCOT’s queue.

Off‑grid or “island‑moded” power has therefore moved from edge case to serious option.

Instead of waiting for a grid connection, data centres co‑locate with dedicated generation, either on the same site or nearby behind private lines, to create a self‑contained power island that can operate independently from the utility network.

Maura Yates, CEO of Mothership Energy, says developers have been drawn to off‑grid designs as they collide with congested queues.

"We're seeing this increased demand for it because the grid-tied interconnect queue is congested and this alternative may provide a way to bypass it," she says.

"We're going to see this demand for off-grid projects because the juice is now worth the squeeze."

Gas becomes the uncomfortable backbone

In the near term, that move off grid is mostly being powered by natural gas.

Developers need dispatchable capacity that can hit “five nines” reliability and run 24/7, which intermittent renewables cannot yet deliver at the scale AI workloads demand without massive overbuild and storage.

Peter Perri III, Managing Partner at Jupiter Island Capital, says the primary bottleneck has flipped.

"What we've seen over the last couple of years is that movement from chips being the primary scaling constraint to power generation being the primary scaling constraint," he says.

Because wind and solar typically operate at relatively low capacity factors, developers would need vast tracts of land to match the output of a single 1GW gas plant that might fit on about 40 acres, often far from the fibre networks AI tenants need.

Natural gas plant supply chains are now feeling the strain as well, with turbine availability, transformers and skilled technicians all emerging as pinch points.

Whitney Switzer, CEO at Sorellis, says developers are fighting over a finite pool of proven equipment and expertise.

"Even when you're thinking about gas generation, there are only a couple of companies that produce the equipment that is validated and tried and true to deploy," she explains.

"When you're in these big projects, especially in remote areas, everybody on site is dependent on certain processes and timelines in these contracts."

Regulators chase the AI gold rush

Policymakers are scrambling to catch up with this AI‑driven build‑out.

In Texas, Senate Bill 6, which took effect in 2025, tightened oversight of very large loads above 75MW, forcing them to disclose on‑site backup generation, accept curtailment during emergencies and contribute more directly to the cost of network upgrades.

At federal level, US President Donald Trump’s administration has coupled pro‑AI rhetoric with efforts to streamline permitting and shield households from the costs of new digital infrastructure.

One proposed statute, the Decentralized Access to Technology Alternatives (DATA) Act of 2026, would exempt fully off‑grid power systems serving data centres from Federal Energy Regulatory Commission rules and North American Electric Reliability Corporation standards, as long as they remain isolated from the bulk grid.

That would effectively carve out a new class of private mini‑utilities, able to sidestep interconnection studies that can add years to project timelines.

Reliability, communities and the partner premium

Even away from Washington and Austin, data centre plans are now colliding with local politics.

By early 2025, research suggested around US$64bn of US data centre projects had been delayed or blocked by bipartisan opposition centred on fears over power prices, water use and neighbourhood impacts.

Sam Lai, Head of Development at Sorellis, says that AI has pushed data centres into the public spotlight.

"There's no getting around the fact that data centres are not super-popular right now," he says.

He argues that early engagement and clear economic benefits, combined with behind‑the‑meter generation that does not compete with local consumers for grid capacity, can soften that resistance.

New models such as Rune’s modular facilities co-located with wind and solar farms show how AI workloads can also mop up surplus renewable output instead of adding to grid stress, hinting at a more symbiotic future.

As more data centres unplug from the grid, execution risk moves squarely onto project partners.

A single facility can cost more than US$1bn, and sprawling AI campuses may host several such buildings, magnifying the cost of any failure.

Brandon Lobb, a Partner in Troutman Pepper Locke’s Energy Transactional Practice Group, argues that AI has changed the hierarchy of concerns.

"AI has shifted the centre of gravity in the energy market," he suggests. "Power availability – not just price – is now the defining variable in digital infrastructure strategy."

With no utility fallback, creditworthiness, technical competence and supply chain resilience are now central filters for both energy providers and data centre tenants, turning off‑grid power from an experiment into a structural feature of the US energy landscape.