Google, Nvidia and Anthropic Join Effort to Free Up 100 GW for AI Data Centers

The next constraint on artificial intelligence infrastructure may not be chips or servers, but the electricity grid. Grid software company Emerald AI has launched the AI Energy Management Alliance, or AEMA, with Google, Nvidia, Anthropic and utilities including AES, Constellation, National Grid and NRG Energy. The coalition says its approach could help secure up to 100 gigawatts of grid capacity for new data centers.

Key points

• Demand response becomes part of data-center planning. Utilities have used demand response for decades. Large industrial customers agree to reduce consumption during peak periods, often in exchange for payments.
• AI workloads offer more flexibility. Data centers can pause noncritical jobs or move some computing to facilities with more available grid headroom. Emerald AI is building software that connects utility requests directly with data-center operations so those changes can happen quickly.
• The aim is to reduce diesel use. Some data centers currently participate in demand-response programs by running backup generators. Emerald AI is promoting a different model based on the fact that many AI workloads can be delayed, shifted or ramped down.
• The approach is not unique. Google is developing its own tools, while Enel X allows data centers to use uninterruptible power supplies to smooth demand peaks. Emerald AI recently raised a $150 million Series A led by Energize Capital and DCVC.

Demand response works because the grid is built to handle its highest loads, even though demand remains below that maximum for much of the year. A Goldman Sachs study published last year estimated that limiting peak grid use to 90% for several hours could free about 76 gigawatts for data centers. The broader implication is that avoiding simultaneous peak operation may be faster and cheaper than immediately building new power plants for every proposed facility.

Why it matters—and what it cannot solve

AEMA could help existing data centers reduce their peak demand, but its ambitions also extend to site selection. By giving technology companies and utilities a common way to evaluate flexible loads, the coalition hopes to identify locations where new facilities can connect more easily. For utilities, controllable data-center demand could function as a large, fast-responding flexible resource.

The limits are equally important. Ayse Coskun, Emerald AI’s chief scientist, told TechCrunch that the company’s technology can reduce the industry’s need for new generation, but cannot eliminate it. As AI workloads continue to expand, grid upgrades, new generation and transmission investment will remain necessary. AEMA is best understood as a way to relieve near-term congestion and improve the use of existing infrastructure—not as a substitute for long-term energy construction.

Source: TechCrunch AI