Datacenters are boosting US clean energy, but also propping up fossil power

AI datacenters are becoming one of the clearest examples of a power-market contradiction in the United States: the same demand boom that is helping drive wind, solar and battery investment is also pushing utilities and developers toward fossil-fuel generation, on-site gas and other stopgap power sources.

The dynamic sits at the center of fresh reporting published Friday by The Guardian, which described datacenter demand as a new engine for clean-energy growth even as it adds to climate pressure. The story is reinforced by a June 18 Federal Energy Regulatory Commission order aimed at speeding grid connections for large AI loads, and by new research estimating the scale of electricity use and emissions from U.S. hyperscale datacenters.

A boom with two directions

The datacenter buildout is creating demand for more electricity at a moment when developers and utilities are already struggling to add new generation and transmission quickly enough. That has made AI infrastructure a powerful source of new business for clean-power projects, but also a source of renewed demand for gas and coal in some markets.

The Guardian reported that datacenter load growth is helping support investment in wind, solar, batteries and hybrid power systems. In some cases, technology companies are using combinations of off-grid or behind-the-meter resources rather than waiting for conventional grid supply.

That approach can speed projects, but it also changes what kind of power gets built. When operators want energy immediately, the cheapest and cleanest long-term option is not always the fastest available one.

Federal regulators move faster

The policy backdrop changed on June 18, when AP reported that FERC unanimously voted to speed interconnection rules for large AI datacenters. According to AP, regional grid operators were instructed to update their procedures so big new loads can connect more quickly.

AP also reported that states will retain control over retail rates and that datacenter customers are expected to cover the cost of the grid upgrades needed to serve them. That is a crucial point: the speed of connection is being paired with the expectation that new AI demand should not simply be shifted onto ordinary power customers.

The order underscores how quickly regulators are being forced to respond to a new kind of load. Datacenters are no longer just another commercial customer class. In many regions, they are now large enough to affect planning, pricing and the timing of new grid investments.

Clean power gets a boost

On the clean-energy side, the datacenter boom is supporting real projects and financing decisions. The Guardian reported examples of major technology-linked developments that combine wind, solar, batteries and gas in search of reliable power.

One cited example is DTE Energy’s decision to build a 330 megawatt battery system rather than a new gas plant to support a 1.4 gigawatt Oracle datacenter, according to consultant Douglas Jester. That is a notable signal: in at least some cases, datacenter demand is helping tip investment toward storage instead of another fossil plant.

Bloom Energy is another beneficiary of the trend. Its first-quarter results, covered by The Wall Street Journal and MarketWatch in April, were tied to rising digital-power demand from datacenters. Oracle also shifted its New Mexico Project Jupiter power plan away from a gas plant and toward Bloom fuel cells, according to later reporting on the project.

Those examples matter because they show how the AI boom can create new markets for technologies that sit between the grid and the datacenter. Fuel cells, batteries and hybrid systems are increasingly part of the response to the need for faster power delivery.

Fossil backup still has a role

But the clean-energy gains do not erase the fossil side of the story. The Guardian reported that some utilities are keeping coal and gas plants online, or considering new fossil generation, because datacenter demand is arriving faster than the rest of the power system can adapt.

That is the basic structural problem. Datacenter operators want power quickly. Grid planners have to reconcile that demand with interconnection queues, transmission constraints, local permitting and the time it takes to build new resources. When those timelines do not match, the quickest available option is often the dirtiest.

The result is that even as AI load growth helps spur clean-energy spending, it can also prolong the life of existing fossil plants or create pressure for new gas capacity. The climate effect depends on which response wins out in each market.

How big is the climate burden?

The concern is not hypothetical. A June 3 preprint on arXiv estimated that 403 U.S. hyperscale datacenters used about 68 to 99 terawatt-hours of electricity and were associated with about 37 to 54 million metric tons of carbon dioxide.

The paper estimated that roughly 54% of the attributed generation came from fossil fuels and put datacenter electricity-weighted carbon intensity at about 545 grams of CO2 per kilowatt-hour, above the U.S. grid average cited in the study. That does not prove every datacenter is equally carbon-intensive, but it does quantify the scale of the emissions issue behind the AI buildout.

Taken together, the research and reporting show why the debate is moving beyond simple claims that AI is either clean-tech friendly or climate hostile. Both are happening at once, depending on the project and the power source.

The companies and utilities shaping the market

Several actors are now central to the scramble. The Guardian’s reporting points to datacenter operators including Google, Oracle, Microsoft, OpenAI and xAI, along with utilities and grid planners such as DTE Energy and regional grid operators.

Clean-power suppliers are also becoming important. Bloom Energy is benefiting from digital-power demand, while Nextpower and similar firms are positioned to sell equipment into a market where large loads need fast, flexible supply.

Energy economist Lucas Davis and consultant Douglas Jester were among the outside voices referenced in the reporting. Their involvement reflects the main unresolved question in this market: how much of the apparent clean-energy boom is truly additional, and how much is simply being pulled forward or reshuffled from projects that were already planned.

What happens next

The near-term outlook depends on what utilities and developers file next. The most important signals will be whether new datacenter loads are paired with batteries, solar arrays, fuel cells or gas plants, and how regulators decide to allocate costs and speed approvals.

The policy conversation is likely to stay active in Washington and in the states. AP’s reporting on the FERC order shows that federal regulators are already adjusting the rules for large AI loads, while state authorities still control retail rates and many of the local siting and cost questions.

More utility filings will also show whether datacenter demand is increasing clean-energy construction that would not otherwise have happened, or mostly changing the timing and location of projects already in the pipeline. That distinction matters for climate policy, because only new buildout represents a real net gain.

For now, the evidence points in both directions at once. Datacenters are helping drive new investment in wind, solar and batteries, but they are also deepening the pressure on a grid that still relies heavily on fossil fuels.

The AI infrastructure boom is therefore not just a technology story. It is now a power-system story, a climate story and a cost-allocation story at the same time.