When the final whistle blew on a nerve-wracking, scoreless draw between England and Germany, millions of Britons did what they do in moments of collective stress: they boiled water. The resulting surge of electric kettles clicking on created a genuine crisis for the National Grid. Yet, the utility company was prepared. As the demand spiked, an artificial intelligence system remotely instructed a London data centre to throttle back its most energy-hungry chips. This intervention prevented potential blackouts and hardware damage, marking a radical shift for facilities that have historically consumed power without regard for wider network needs.
Crucially, this was a simulation. In December 2025, engineers recreated the strain on the UK grid during the 2020 Euro tournament to test a new breed of adaptable infrastructure. They wanted to see how their software, Conductor, would have performed had it been active at the time. Conductor is the flagship offering of Emerald AI, a Washington DC-based firm leading a wave of companies exploring whether data centres can operate within the strict confines of the existing electricity network.
This year, Emerald plans to deploy Conductor in a new facility in Virginia’s Data Center Alley, connecting directly to the live grid. When overall demand surges, the system will reduce the centre’s power consumption while ensuring servers continue their most critical tasks. Emerald’s partners, including Nvidia and Digital Realty, describe this as one of the world’s first “power-flexible AI factories.”
Why flexibility matters for makers and artists
Demonstrating that data centres can participate in this give-and-take addresses a major bottleneck for creators and businesses: the time it takes to build power plants. Constructing and approving new generation capacity often takes far longer than erecting a data centre. PJM, the US grid operator in Virginia, requires eight years to bring new generation online, according to RMI. Josh Parker, head of sustainability at Nvidia, notes that “AI factory flexibility is the bridge between the incredible demand for AI and the immediate limitations of our energy grid.”
Speed is vital, but it is not the only concern. Once facilities connect, they often face backlash from neighbours who argue they contribute to rising prices, pollution, and job losses. Organisers stalled over $150 billion in projects in 2025, according to Data Center Watch, prompting policymakers to impose stricter limits. More than a dozen states are considering bans, with moratoriums already in place in locations like Minneapolis and DeKalb County in Georgia. At the federal level, the bipartisan GRID Act proposes severing new data centres from public grids entirely.
However, the solution may lie closer to home. The existing transmission network operates near full capacity only during a small number of high-demand hours. Grid experts argue that if data centres limit their draw during these specific stretches, they can avoid waiting for massive infrastructure upgrades or building their own off-grid generation.
Studies support this approach. A 2025 report from Duke University researchers found the US grid could offer an additional 76 gigawatts-roughly 5% of total capacity-to facilities willing to reduce usage just 0.25% of the time. That equates to about 22 hours a year. Similarly, a report funded by Google, conducted by Princeton University and grid-modernisation companies, found that a 500-megawatt facility capable of flexing for less than 1% of the year could reach full operation three to five years faster than an inflexible one.
Flexible connections also offer a path to resolving public relations issues. By lowering their draw during grid stress, data centres can avoid diverting power from essential services, boosting stability. Using existing capacity might also reduce the need for new fossil-fuel plants and spread fixed costs across more users, potentially pushing prices down.
This “power pinch” is driving research into grid flexibility, a crucial strategy as data centres, alongside electric vehicles and air-conditioning, are predicted to drive a 25% increase in US electricity demand by 2030 compared with 2023 levels.
Ultimately, flexibility gives grid operators more control, allowing them to lead a harmonious ensemble rather than acting as hostages to rigid requirements. Johanna Mathieu, a grid expert at the University of Michigan, states that “Demand flexibility is incredibly useful for power grids,” noting it helps reduce costs and improve reliability.
Despite the benefits, the concept introduces complexity. For data centres, compromising on energy needs is a difficult sell. It requires utilities, which tend to be operationally conservative, to alter long-held practices. Skeptics also argue that flexibility distracts from the urgent need to build more grid infrastructure faster and could pose supply risks.
Nevertheless, technologists and utilities are hoping to prove that flexibility works in the real world, not just in simulations.
The current reality of data centre construction
Most major data centre growth currently defaults to inflexibility. Hyperscalers like Microsoft and Oracle have proposed enormous new centres relying on off-grid, natural-gas-burning plants. When xAI accelerated the buildout of the Colossus site in Tennessee, it deployed gas turbines on flatbed trucks. The facility, now operational, is facing criticism from regulators and residents over emissions and pollution. Furthermore, there simply aren’t enough gas turbines globally to meet the demand from data centre operators.
A primary obstacle is that grids are designed to supply enough power to meet total demand at the highest point, even if that occurs for only a few hours a year. This conservative approach ensures reliability but leaves significant headroom unused. Amit Narayan, CEO of GridCare, explains: “If you were an airline running at 30% utilization, you would not buy more planes. If you are running a grid at 30% utilization, there’s no scientific reason you can’t go to 60.”
Grid operators have long used “demand response,” calling on large facilities to shut down parts of their operations when demand threatens supply. While this avoids firing up fossil-fuel “peaker plants,” it is slow, imprecise, and hard to scale. The 2000s saw the rise of virtual power plants (VPPs), offering a smarter, faster, and more granular alternative where customers from factories to homeowners allow utilities to adjust their draw.
Following the generative AI boom sparked by ChatGPT in 2022, companies began viewing flexibility as a way to set up data centres more easily and affordably. Bringing AI money into existing grids could reduce the need for expensive upgrades and help spread fixed costs, lowering rates for other users. A study from Duke University published earlier this year found that flexibility could reduce rates by 0.5% to 2.8%.
How data centres can maintain operations
The challenge remains figuring out how power-hungry data centres can operate when their flexible connections are throttled. Specialists envision three potential solutions. The simplest involves installing on-site backup power storage or generation to tap when the grid is maxed out, though this comes at the facility’s own expense.
Alternatively, a facility could fill the gap by drawing on a VPP. The utility would reduce electricity to users who signed up for the VPP, and the data centre would pay them for their flexibility. This method avoids major infrastructure but requires a robust utility programme and coordination during grid stress. While VPPs exist to some extent in nearly 40 states, the rules governing them vary widely.
Key takeaways
- Flexible data centres can utilise existing grid capacity during low-demand periods, potentially accelerating deployment by three to five years.
- Adapting power usage can reduce the need for new fossil-fuel plants and help lower electricity rates for other consumers.
- Despite regulatory hurdles and public resistance, flexibility represents a critical bridge between AI demand and current grid limitations.
- Implementing flexibility requires new operational models for utilities and a shift away from the traditional “build more” approach to energy infrastructure.
