Dell shifts to fully liquid-cooled racks to tap scattered power for AI workloads
Dell is packaging liquid-cooled networking and storage into single racks while targeting stranded energy sources to handle massive AI inference demands.
Artificial intelligence infrastructure is growing too large for traditional server rooms. As models expand and distributed inferenceInferenceRunning a trained AI model to get an answer. Open in glossary spreads into production, hardware setups must stretch across many facilities and tap into small, scattered sources of electricity. Dell Technologies is responding by shifting its focus toward pre-integrated systems instead of individual parts.
Liquid cooling across the whole rack
Modern computing clusters now link tens of thousands or hundreds of thousands of graphics processing units on a single network fabric. To manage the extreme heat produced by these systems, Dell is moving liquid cooling beyond just processors.
The company is rolling out what it describes as the first entirely liquid-cooled networking setup. This system includes network switches, optical components, and cables that Dell tests in its factory before shipping them together in a single rack. Dell previously introduced liquid-cooled storage for AI hardware. Unifying compute, networking, and storage under identical cooling and electrical designs lets operators construct one combined facility instead of separate spaces for processing chips and network gear, cutting deployment times from weeks to hours.
Harvesting stranded energy for inference
Finding enough electricity in one place to run massive data centers has become a major roadblock. As a result, businesses are looking to run AI inference across up to 1,000 smaller locations that hold unused power capacity.
This distributed setup relies on software that acts as a single control planeControl planeThe administrative software layer that manages and directs how computing resources run across a network. Open in glossary across mixed processors. By distributing workloads, companies can take advantage of unusual or stranded energy sources, ranging from extra power at retail outlets to captured methane emissions. Connecting these small pockets allows organizations to run workloads without waiting years to construct massive centralized facilities.
How does this story land?
One tap per reader.
Quick poll
Will distributed AI inference across scattered sites replace mega data centers?
One tap. No sign-up.
Ask this story
Still have a question?
Answers use only what is in this story. They are written by AI and can be wrong.