Emerald AI leads alliance with Google and Nvidia, demand-response software enables 100 gigawatts for data centers

Emerald AI, a company specializing in electricity grid software, partnered with Google and Nvidia to establish the Artificial Intelligence Energy Management Alliance (AEMA), a move aimed at using the startup’s software to secure vacant capacity on power grids to accommodate additional data centers. The founding alliance was joined by Anthropic, along with a group of electric utility companies including AES, Constellation, National Grid, and NRG Energy.The alliance seeks to make demand response a core pillar in the development of computing infrastructure.It asserts that suspending non-critical computing tasks and temporarily shifting some processing loads are sufficient to enable power grids to connect additional data centers with a total capacity of up to 100 gigawatts without the need to build immediate feed lines of the same size.

The demand-response concept relies on an operational reality long used by utilities, namely that electricity grids are engineered to handle a projected peak consumption, while actual demand for most of the year is far below the maximum capacity. Traditionally, demand-response programs involved large industrial consumers such as factories, which agree to reduce their consumption during sharp peak hours in exchange for generous financial incentives paid by the utilities, either by temporarily shutting down production lines or by running backup generators. Today, AI data centers typically rely on these programs by operating polluting backup diesel generators, whereas Emerald AI offers a software alternative that coordinates utilities’ requests directly with data centers to suspend deferrable computing operations or shift loads to other centers whose local grids have available surplus.

The company is not alone in developing these tools; Google builds its own internal technologies, and Enel X enables data centers to use uninterrupted power supply (UPS) systems to mitigate peak loads. A Goldman Sachs study released last year shows that lowering grid usage to 90 percent for a few hours at a time can free up to 76 gigawatts of capacity.Emerald AI’s advantage lies in linking utilities directly to data centers programmatically.This gives facilities an instant response capability comparable to that of large electrical batteries. To support this expansion, the company recently raised $150 million in a Series A round led by Energise Capital and DCVC, while the new alliance commits to help technology and utility companies identify suitable sites for future computing centers.

Despite these software promises, smart solutions do not eliminate the root of the supply crunch, as Aisha Kushkoun, chief scientist at Emerald AI, explains that the company’s technology eases the tech sector’s need for new electricity generation sources but does not remove that need entirely. The volatile nature of AI workloads allows capacity to be increased or decreased rapidly, yet expanding total capacity ultimately requires actual power flow through wires to run the processors.

This shift reshuffles infrastructure priorities for those operating data centers and cloud services in the Gulf, Egypt, and the Levant. In the Gulf environment especially, where electricity grids face a severe seasonal peak in the summer months due to extreme cooling loads, relying on software flexibility offers an alternative to waiting for new power plants to be built to connect super-computing complexes. In practice, the negotiation model with national electricity authorities changes, allowing operators to commit to reducing instantaneous draw rather than reserving fixed, costly capacities that raise operating bills. The approach also compels engineering teams to develop architectural skills capable of segregating deferred training loads from real-time inference loads, turning electric-load management into an engineering standard that precedes the purchase of computing hardware itself.

The partnership confirms that the AI expansion battle has shifted from processor abundance to managing every megawatt on distribution lines.The wager is no longer limited to algorithmic efficiency but to the software system’s ability to maneuver with power grids at the moment loads reach their peak limits.