When AI consumes too much electricity, Google steps back in its data centers.

Tech giants devour electricity for their AI, to the extent that they invest in the American nuclear industry to power their very hungry data centers. But these increasingly important energy needs should not override those of humans, especially during periods when heating or air conditioning is required — essential for enduring scorching summers.

Read Why tech giants are going after nuclear power

It is in this context that Google has implemented a "demand response" solution, which involves adjusting or postponing certain electrical loads, such as machine learning tasks (for example, encoding a YouTube video), which are not urgent during periods of high network demand.

This mechanism helps reduce pressure on electrical infrastructure, speeds up the integration of new data centers, and avoids or delays the construction of new power plants.

The web giant has signed two new agreements with Indiana Michigan Power (I&M) and the Tennessee Valley Authority (TVA), to deploy these capabilities in data centers. These agreements are a first: energy optimization is now directly targeting loads related to machine learning.

Google had already experimented with this approach two years ago in the United States; the company also collaborates with partners such as Centrica Energy in Belgium or Taiwan Power Company. He intends to expand this model, particularly in regions where AI progress could risk saturating electricity production or transmission capacities.

👉🏻 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source : Google