The AI is putting the US power grid in trouble. And Google has already taken a measure that shows the magnitude of the problem

Google data centers work 24/7, processing searches, videos and now also AI models. But not everything can grow at the same pace. In several areas of the United States, electricity begin to notice the pressure: Energy demand is accelerated and In some places already exceed capacity forecasts.

Given that scenario, Google moves: It will reduce the consumption of your data centers when there are peaks, prioritizing the essential and postponing what you can expect. The novelty is the focus: Machine Learning charges.

Artificial intelligence progresses. The electricity grid notice. The expansion of AI is going so fast that companies receive more connection requests than they can meet in certain areas. The consequence is no longer only technique: there is an energy restriction that conditions the deployment.

It’s not about turning off machines, but moving loads. The “demand response“It consists of adapting consumption to what the network can supply at all times. In practice: displace or reduce non -urgent loads – like the processing of programmable videos or tasks – outside critical hours. It is a tool used in intensive industries and cryptocurrency mining, now applied to data centers with AI.

The system has clear limits. This type of flexibility is not applicable in all centers or in all situations. Google recognizes it clearly: there are services that you just can’t expect. Platforms such as Search, Maps or the cloud for critical sectors – such as health or emergencies – require continuous availability, without margin for load settings. There are no “non -urgent” tasks that can be postponed.

Therefore, although the response to demand is a valuable tool, its implementation will remain partial and selective. It requires planning, previous agreements and an infrastructure designed to absorb that type of reorganization. Not all centers can do it. But where it is possible, it becomes a real way to relieve pressure on the network without compromising the essential.

There is already experience, and now. It is not theory. Google tested this flexibility With the public electric of Omaha and reduced demand associated with Machine Learning in three network events last year. The next step are formal agreements with Indiana Michigan Power (Fort Wayne) and with Tennessee Valley Authority: in Indiana it will be integrated from the beginning of the new center, and in Tennessee it will be applied coordinated with the operator.

From experiment to strategy. What began as a pilot becomes operational policy: Managing demand flexible helps stabilize the network and accelerates the connection of large loads without waiting for new lines or centrals. It is not a magical solution, but it wins time while the infrastructure is reinforced.

Images | Xataka with Gemini 2.5 Flash | Andrey Metelev

In Xataka | Google has signed the largest hydroelectric agreement in history. You no longer know where to get more energy to feed your AI