AI told a London data center to throttle during a soccer kettle spike

It started with something very British: during a tense, scoreless first half of an England vs. Germany match, millions of Brits made tea. In the real world, that means electric kettles click on, and with them comes a sudden surge in electricity demand. In this story, the kettle spike triggers a different kind of stress test. An AI program sent instructions to a data center in London to slow down some power-hungry chips, helping keep supply aligned with demand and staving off potential blackouts or damage to electrical hardware.

That is the real point. Not “data centers are flexible in theory,” but “a data center can react fast enough to protect the grid while keeping its most important work running.” Emerald AI’s Conductor did exactly that in a simulation engineers ran in December 2025. They re-created the energy demand facing the UK’s grid during a match from the 2020 Euro tournament and then tested how Conductor would have responded had it been online at the time. Conductor is the signature product of Emerald AI, a Washington, DC-based firm, and it is part of a broader push by companies trying to prove that data centers can fit within today’s electric-grid constraints instead of treating the grid like an unlimited backstop.

Emerald’s next step is not another white-paper demo. This year, the company plans to deploy Conductor in a new facility in Virginia’s “Data Center Alley,” connected to the live grid. The operating promise is straightforward: when overall demand spikes, Conductor will turn down the power used by the data center, while still ensuring that servers carry out their timeliest and most important jobs. The project’s partners include Nvidia and Digital Realty, and they are billing it as one of the world’s first “power-flexible AI factories.” For executives, this is a direct shot at the most frustrating part of scaling AI infrastructure: getting enough electricity, fast enough, without forcing the grid operator into panic mode.

Speed is one stake, but it is not the only one. Many tech leaders argue that approval, construction, and connection of new power plants takes far longer than building data centers themselves. The PJM grid operator in Virginia, for instance, needs eight years to bring new generation online, according to RMI, an energy research and advocacy group. Flexibility is pitched as the bridge between AI demand and the “immediate limitations” of the energy grid. Josh Parker, head of sustainability at Nvidia, frames it bluntly: “We need to solve the energy equation,” and “AI factory flexibility is the bridge between the incredible demand for AI and the immediate limitations of our energy grid.”

Then there is the other reality executives have to manage, especially when projects move from planning boards to neighborhoods. Once a facility plugs in, neighbors often complain about electricity draw and its downstream effects on prices. They also raise concerns about noise, pollution, and job impacts. In 2025 alone, organizers stalled more than $150 billion worth of projects, according to Data Center Watch. Policymakers, reading the room, are starting to impose limits on development: more than a dozen states are considering bans, and local moratoriums are in effect in Minneapolis and DeKalb County, Georgia. At the federal level, the GRID Act, a bipartisan bill in the US Senate, proposes to sever new data centers from public grids entirely. Some operators are already moving in that direction by pursuing their own power generation, trying to avoid the grid bottleneck by bypassing it.

But the source also makes a contrarian case: maybe you do not need more turbines everywhere, because the grid is not always at full capacity. The existing system operates near full capacity during only a small number of high-demand hours each year. Some grid experts argue that if data centers can limit power draw during those stretches, they can reduce the need to wait for major infrastructure upgrades or build off-grid generation. Studies cited in the piece suggest there could be “plenty of power available” for data centers that can flex. A widely discussed 2025 Duke University report, for example, found the US grid could offer an additional 76 gigawatts, about 5% of total capacity, enough to accommodate projected data-center growth through 2030 if facilities reduce usage just 0.25% of the time. That equates to about 22 hours a year.

A separate report funded by Google, from Princeton University and two grid-modernization companies, looked at locations for new data centers in the PJM region and 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 a more inflexible one. The same flexibility argument also comes with potential PR upside. By decreasing draw during grid stress, data centers might avoid diverting power from where it is most needed, improving stability. By using existing capacity more effectively, they might reduce the need for new fossil-fuel power plants and spread fixed costs across more electricity users, potentially pushing prices down. And across the whole system, flexibility matters because the “electrons problem” is not just data centers. Taken together with electric vehicles, air-conditioning, and other sectors, the piece notes analysts predict a 25% increase in US electricity demand by 2030 compared with 2023 levels. Flexibility could give grid operators more control over demand spikes and help manage renewables’ intermittency from wind and solar.

Of course, flexibility is not magic. The source warns it adds complexity. For data centers, compromising on energy needs can be a hard sell. Flexibility also requires utilities and grid operators, which tend to be operationally conservative, to change long-held practices. Skeptics argue flexibility distracts from the real need to build grid infrastructure faster and could pose risks to electricity supply. The hyperscalers most associated with massive new builds are leaning toward rigidity: Microsoft and Oracle have proposed enormous new centers, many relying on off-grid, natural-gas-burning power plants. When xAI wanted to speed up the Colossus site outside Memphis, Tennessee, it used gas turbines on flatbed trucks; the facility is now in operation and is facing blowback from regulators and residents over spikes in emissions and other pollution. There may not be enough gas turbines worldwide to meet the demand from data-center operators.

So the question for executives is no longer abstract. It is operational, regulatory, and capital allocation all at once: can you buy time on grid upgrades by proving that your facility can flex in real conditions? Emerald’s December 2025 simulation and its planned Virginia live deployment with Nvidia and Digital Realty are designed to answer that. If it works, power-flexible AI factories could become a practical alternative to waiting eight years for new generation or racing into off-grid bets that invite political blowback. If it fails, regulators and neighbors will keep pushing for constraints, and the whole industry could keep paying the “energy equation” tax in delays, costs, and reputational risk.