AI data centers want more power, and Google, Nvidia and Emerald AI are betting they can get it faster by learning when to use less.
Google, Nvidia and Emerald AI have launched the AI Energy Management Alliance (AEMA), an 18-member coalition pushing utilities and regulators to treat flexible AI data centers as controllable grid resources. The idea is straightforward: facilities that can reliably cut or shift electricity demand when the grid is strained could have another path to securing the enormous power connections AI infrastructure requires.
The approach could change the economics of AI data center development. Instead of waiting solely for utilities to build enough generation and transmission capacity to meet peak demand, developers could trade verifiable power flexibility for faster or larger grid connections.
How flexible data centers would work
A data center could respond to grid conditions by shifting or pausing lower-priority computing tasks, drawing from batteries, using on-site generation or reducing consumption during emergencies.
AEMA says its framework will focus on measurable performance rather than specific technology. That includes how quickly a facility can respond, how long it can reduce demand, how predictable its response is and how it performs during grid emergencies. The alliance also wants standardized technical requirements and data-sharing rules, along with faster interconnection processes for facilities that make credible and verifiable commitments.
“The most effective way to accelerate American AI is to make every data center a good citizen of the grid,” said Emerald AI CEO Varun Sivaram.
Google already has agreements covering about 1 gigawatt of electricity demand that it can reduce when needed, according to Axios.
The power bottleneck
The initiative comes as securing electricity has become a major challenge for new U.S. data centers. Sivaram said in a Fortune commentary that a new facility can wait a decade or more for a grid connection, while the U.S. grid is only about 50% utilized on average.
AEMA estimates that making AI facilities moderately flexible during the grid’s most stressed periods could unlock up to 100 gigawatts of existing capacity. The companies are also testing the concept. Emerald AI and Nvidia have completed six demonstrations, while Nvidia, Digital Realty and Emerald AI plan to activate a nearly 100-megawatt power-flexible AI facility in Virginia later this year.
What changes for data center developers and the channel
The proposal could give data center developers another route to securing power without relying entirely on new generation and transmission projects. Instead of designing every facility around uninterrupted access to electricity at its maximum possible demand, operators could use workload scheduling, batteries, and other power-management tools to reduce consumption when the grid is strained.
That could matter as access to electricity becomes a greater constraint on the expansion of AI infrastructure. Nvidia has already moved deeper into the power side of the market, including investing in Cloverleaf Infrastructure as developers compete for land, electricity and grid access.
If utilities begin rewarding facilities that can reliably reduce demand, the ability to manage power use by a data center could become another factor in site selection and infrastructure planning, alongside available megawatts, networking, cooling, and access to AI hardware.
But the model depends on data centers actually delivering the flexibility they promise. Sivaram told Axios that faster or larger connections should require flexibility that is “verifiable and enforceable.” Utilities would need confidence that operators can reduce demand quickly and predictably when called upon, rather than simply promising flexibility to secure a connection.
That requirement could also create opportunities across the channel. Data center operators and their customers may need help determining which AI workloads can be delayed or shifted, how much battery capacity is required, when on-site generation should be used, and how power-management systems should respond to utility signals.
Energy is already beginning to move into the channel conversation. AppDirect, for example, sees energy procurement as an emerging opportunity for MSPs, resellers, and technology advisors who help customers navigate AI infrastructure decisions. Flexible data centers could widen that opportunity by adding workload orchestration, power monitoring and grid-response planning to the mix.
The trade-off is that not every AI workload is equally flexible. Training jobs and other batch computing tasks may offer more room for scheduling changes, while latency-sensitive inference and business-critical applications may be harder to interrupt. Developers will therefore have to demonstrate not only how much electricity they can shed, but which workloads can absorb those reductions without creating problems for customers.
That makes regulation and measurement just as important as the underlying technology. AEMA can develop technical standards and demonstrate that flexible computing works, but utilities and regulators ultimately have to decide how those capabilities translate into interconnection agreements and whether flexibility warrants faster or larger grid connections.
If that model gains traction, power management could become another layer of the AI infrastructure stack. For channel partners, the question may increasingly shift from simply helping customers find enough computing capacity to helping them build AI environments that can prove when, where, and how they can use less electricity.
More Channel Insider news: KKR’s Helix targets AI data centers as the AI boom makes access to electricity a growing infrastructure bottleneck.