Artificial intelligence news.
Straight from the original announcements.

Equinix opens liquid cooling test centers in five cities for enterprise AI

October 21, 2026, 13:45Source: Equinix

Equinix has launched its Enhanced Performance Innovation Centers (EPIC), liquid cooling environments inside live IBX data centers in London, Paris, Frankfurt, Osaka and Sydney. Enterprises can test liquid-cooled GPU and CPU platforms there with Equinix engineers and OEM partners before committing to a full deployment.

Key points
  • Equinix EPIC centers are live in London, Paris, Frankfurt, Osaka and Sydney, with more sites planned for 2027.
  • Each is a production-ready liquid cooling environment inside an active Equinix IBX data center.
  • Enterprises can test liquid-cooled GPU and CPU platforms from OEM partners with Equinix engineers before scaling.
  • Equinix says a rack of current-generation GPUs can draw more than 200 kW, versus 5 to 10 kW per rack in many older designs.
  • Criteo says it built confidence step by step before committing to scale liquid cooling.

Five sites live, more planned for 2027

According to Equinix, the EPIC centers are now live in five locations: London (LD5), Paris (PA10), Frankfurt (FR6), Osaka (OS3) and Sydney (SY5). Each is a production-ready liquid cooling environment inside an active Equinix IBX data center. The company says more sites are planned for 2027.

Enterprises can bring in the latest liquid-cooled GPU and CPU platforms from Equinix's OEM partners and run training and inference workloads under production conditions. They can also work through operational questions with Equinix engineers before rolling out at full scale. Site-by-site details are published in a product release note on Equinix Community.

Why the power gap matters

Equinix points to a sharp mismatch between current hardware and older facilities. A rack of current-generation GPUs can draw more than 200 kW, while most data centers built a few years ago were designed around 5 to 10 kW per rack.

The company says closing that gap is why liquid cooling has moved from a specialist topic to a standard line in enterprise AI plans. It is also, according to Equinix, why many infrastructure teams want to see the technology run before they commit capital to it.

How the cooling system is built

Equinix says every EPIC site follows the same basic design:

  • A primary loop carries heat from the coolant distribution unit (CDU) to the building's chillers.
  • A secondary loop carries coolant from the CDU to the racks, where manifolds and flexible hoses deliver it to each server through quick-disconnect couplings.
  • Sensors track temperature, flow and pressure along the way.
  • Leak detection with automatic shutoff valves sits underneath the system.

The announcement post includes photos from all five sites and a short video walk-through from the CDU to the rack. Equinix also links to a separate explainer on direct-to-chip liquid cooling by Lindsay Schulz and Sebastien Ye.

Criteo: see it, run it, then scale it

Equinix cites the ad-tech company Criteo as an example of why testing first matters. Criteo runs AI across a distributed footprint and needed liquid cooling to perform the same way at every site.

"Liquid cooling was new territory for us, so we built our confidence step by step," said Matthieu Blumberg, SVP of Engineering Infrastructure, Security, Internal IT & CISO at Criteo, in the Equinix blog post. "We saw how liquid cooling performed in practice, learned what it took to run it well, and then committed to scale."

Equinix describes that sequence as what EPIC is built for.

What it means for enterprise AI teams

For organizations planning high-density AI deployments, EPIC offers a way to trial liquid-cooled hardware in a live data center setting rather than relying on specifications alone. Teams can validate how the equipment behaves and what operating it requires before spending on a full rollout.

Interested companies can visit the EPIC program page to find the nearest site and request a visit, or contact their Equinix account team to arrange a session with the engineers.

Related news

Google makes GKE dynamic slicing generally available for TPU workloads
Chips and infrastructureNovember 9, 2026, 14:00

Google makes GKE dynamic slicing generally available for TPU workloads

Google Cloud has made GKE dynamic slicing generally available, allowing Kubernetes workloads to assemble and partition TPU slices without repeatedly creating and…

AWS SageMaker adds GPU partitioning for more efficient model hosting
Products and toolsNovember 6, 2026, 19:01

AWS SageMaker adds GPU partitioning for more efficient model hosting

Amazon SageMaker Inference now supports NVIDIA Multi-Instance GPU partitioning, allowing multiple models to share a physical GPU through hardware-isolated partitions.…

Flatkey launches dedicated NVIDIA B300 clusters for AI inference
Chips and infrastructureNovember 5, 2026, 19:30

Flatkey launches dedicated NVIDIA B300 clusters for AI inference

Flatkey has launched Flatkey Compute, a NeoCloud that builds and operates dedicated NVIDIA B300 clusters for AI inference. Each cluster is reserved for one customer…