Google, NVIDIA and Emerald AI Launch Alliance to Make AI Data Centers More Grid-Friendly
Google, NVIDIA and Emerald AI have launched the AI Energy Management Alliance (AEMA), a new coalition focused on changing how large AI data centers use electricity.
The alliance, announced Wednesday, September 16, brings together companies and organizations from the artificial intelligence and energy industries to promote data centers that can adjust their power consumption when electricity grids come under pressure. Axios reported that the coalition includes 20 members, including Anthropic, National Grid, AES, Constellation, NRG and RWE.
The idea addresses one of the biggest infrastructure challenges facing the AI industry: rapidly growing electricity demand and the limited speed at which new grid capacity can be brought online.
AI Data Centers Could Become Flexible Power Users
Traditional data centers generally operate as large, relatively predictable electricity loads. AI facilities, however, can potentially change how much power they consume by adjusting computing workloads, using energy storage or coordinating with other power resources.
AEMA wants utilities and grid operators to recognize that flexibility as a resource.
Under the alliance’s proposed framework, data centers could agree to reduce or adjust their electricity consumption during periods of grid stress. In return, facilities that make credible and verifiable flexibility commitments could potentially receive faster or more flexible pathways for connecting to the grid.
The alliance says the approach could make better use of existing electricity infrastructure while reducing the need for some costly upgrades.
That does not mean AI data centers would simply shut down when the grid is strained. Instead, flexibility could involve shifting workloads, temporarily reducing computing demand, discharging batteries or using paired generation, depending on the facility.
Google Already Has 1 GW of Demand Flexibility
Google has already incorporated approximately 1 gigawatt of demand flexibility into utility agreements nationwide, according to Tyler Norris, Google’s head of advanced energy market innovation, who discussed the initiative with reporters.
That provides an example of the type of flexibility AEMA wants to expand across the industry.
The concept is particularly relevant as AI companies build increasingly power-intensive facilities. Instead of treating every new data center as a fixed electricity requirement, the alliance wants grid operators to consider whether a facility can respond when power supplies become constrained.
The potential benefit is faster access to electricity capacity without requiring every new AI project to wait for large grid expansions.
NVIDIA and Emerald AI Have Already Tested the Model
The new alliance builds on work NVIDIA and Emerald AI have already been doing with utilities, grid operators and data center companies.
Emerald AI has developed software designed to coordinate AI computing workloads with electricity-grid conditions. Its work with NVIDIA has included demonstrations in several locations, including Arizona, Illinois, Virginia, Oregon and London.
In 2025, Emerald AI and NVIDIA also announced a power-flexible AI factory project involving a 96-megawatt facility in Virginia. The project was designed to demonstrate how an AI data center could dynamically adjust its electricity consumption while continuing to run demanding AI workloads.
In June 2026, Emerald AI announced a commercial deployment with NVIDIA and Silicon Valley Power in California, using Emerald’s Conductor platform alongside NVIDIA DSX Flex technology to allow an AI data center to respond to utility signals.
Those projects provide a technical foundation for the broader industry framework AEMA is now promoting.
AEMA Wants Common Rules for Flexible Data Centers
A major part of the alliance’s proposal is standardization.
Rather than requiring a particular battery, software platform, generator or computing architecture, AEMA says its framework will be technology-neutral and performance-based.
That means facilities would be evaluated based on measurable capabilities such as:
- How quickly they can change electricity consumption
- How long they can maintain a reduction
- How predictable their response is
- How they behave during grid emergencies
- Whether their flexibility can be independently verified
The alliance also proposes clearer requirements for ride-through capabilities, curtailment and emergency response before a facility connects to the grid.
Standardized technical requirements and operational data could give utilities and grid operators more information about how an AI facility will behave under stressed conditions.
The Goal Is Faster AI Infrastructure Without Ignoring the Grid
The launch comes as electricity availability has become a major constraint on the expansion of AI infrastructure in the United States.
A separate report from Broadband Breakfast noted that federal proceedings launched in June are examining issues surrounding large data-center connections, including transmission costs, interconnection procedures and flexible demand.
The pressure is not only coming from technology companies. Communities and regulators are also increasingly focused on how rapidly expanding data centers could affect electricity costs and local infrastructure.
Axios reported that 84% of Americans surveyed in new AP-NORC/University of Chicago Energy Policy Institute polling said they were concerned about data centers’ impact on local electricity prices.
That makes the flexibility argument important beyond the AI industry itself.
If large data centers can reduce their electricity demand during relatively limited periods of grid stress, utilities could potentially use existing infrastructure more efficiently instead of immediately building additional capacity for the highest-demand periods.
What Happens Next
AEMA plans to bring together AI companies, data center operators, technology providers, utilities, power producers and regional grid operators to develop technical approaches and advocate for policies recognizing grid-responsive demand.
The alliance’s success will depend on whether utilities and regulators accept flexible data-center loads as a reliable resource and whether operators can consistently deliver the promised reductions when the grid needs them.
For AI companies, the stakes are significant. Faster access to electricity could remove one of the bottlenecks slowing the construction of new AI infrastructure.
For utilities and communities, the question is whether flexible data centers can genuinely reduce pressure on existing power systems while maintaining reliable service.
The launch of AEMA signals that the next stage of the AI infrastructure race may involve not just building more computing capacity, but building data centers capable of working with the power grid instead of simply demanding more from it.
