Google, Nvidia, Anthropic, and the network software provider Emerald AI jointly founded the AI Energy Management Alliance on September 16, 2026, along with 18 other partners. The coalition aims to unlock around 100 gigawatts of additional grid capacity for AI data centers in the USA by allowing data centers to flexibly reduce their power consumption during grid bottlenecks. This is expected to make new power plants partially unnecessary.
Emerald AI controls data centers like a battery
At the core of the alliance is the software from Emerald AI, which connects grid operators directly with data centers. The Emerald Conductor platform orchestrates AI workloads and on-site energy sources so that facilities can temporarily pause non-critical tasks or shift them to other locations during grid bottlenecks – similar to a battery that quickly responds to peak loads. The underlying principle is called demand response: large consumers temporarily reduce their consumption instead of permanently drawing peak loads.
Emerald AI itself completed a Series A round of $150 million in August 2026 at a valuation of $1.05 billion, led by Energize Capital and DCVC. According to company information, Nvidia is also among the investors through its venture arm NVentures. In total, the company has raised more than $220 million. Nvidia had previously invested three billion dollars in power supplier Lancium to secure the energy supply for its large customers. Emerald AI’s chief scientist Ayse Coskun stated to TechCrunch that the technology would dampen the industry’s need for new power plants, though not eliminate it entirely.
Grid operators and chip companies agree on common standards
The founding members include, in addition to Google, Nvidia, and Emerald AI, also Anthropic as an AI lab. The 18 partners also include the grid operators National Grid, AES, Constellation, and NRG Energy, as well as semiconductor manufacturer Analog Devices and data center operator Digital Realty. They aim to establish uniform technical standards for load reduction and performance metrics – so far, each operator has negotiated such conditions individually with its local supplier.
Emerald AI founder and CEO Varun Sivaram wrote in a guest article for Fortune that data centers which adapt their operations to the needs of the grid should be rewarded with faster access to new power. According to the alliance, US power grids operate at only about 50 percent of their capacity on average. Each ten percentage points of higher utilization could, according to a calculation by the Brattle Group, reduce electricity prices by about 3.4 percent. A study by Goldman Sachs estimates the short-term usable additional potential at up to 76 gigawatts if data centers limit their peak grid load to 90 percent – independently unverified.
Google brings its own experience, standards still missing
Google says it already operates a demand response portfolio of around one gigawatt and is bringing this experience into the alliance. Anthropic, meanwhile, primarily contributes its growing data center needs – the company recently signed several billion-dollar contracts for additional computing power. In Virginia, according to Emerald AI, the world’s first power-flexible AI data center is set to be built, with nearly 100 megawatts of capacity as a pilot facility for the new standards. Other tech companies are taking the opposite approach: Amazon is building its own gas power plant in Texas to remain independent of the public grid.
The alliance has not yet given a binding date for the new technical standards or a timeline for reaching the 100-gigawatt mark. There is no comparable program for Germany or the EU; the initiative initially applies exclusively to US power grids. A German energy company, Uniper, is pursuing a similar approach, however, renting out power plant sites with existing grid connections to data center operators.
What will be decisive is whether the alliance can make its standards binding enough that grid operators actually grant faster connection approvals in return – so far, it remains a voluntary commitment without regulatory pressure. If the pilot project in Virginia proves successful, the model could also gain relevance for the planned expansion of AI data centers in Europe, where scarce grid connections are already delaying projects.
