Free Newsletter
HEAT AIR NEWS
Breaking News and Technical Articles for the HVAC Community, by EmailREFRIGERANTS  |  POLICY  |  DATA CENTERS  |  DEALS  |  TECHNICAL
Subscribe Free
EventsAboutContactNewsletter
Data Centers

Vertiv Opens ATLAS Lab to Test AI Factory Power and Cooling Before Chips Ship

Vertiv and Nvidia engineers cut the ribbon on Nov. 6. A Vertiv brief schedules full dynamic AI workload emulation for early 2027.

Key TakeawayVertiv's ATLAS Lab runs power, thermal and controls equipment together as a representative AI factory pod, using emulated Nvidia workloads to test infrastructure before the matching chips reach the market. Vertiv says designs proven at pod scale become reference designs for campus-scale builds.

Vertiv said Nov. 16 that it is testing the power, thermal management and controls infrastructure for AI data centers in a new system-level lab before the Nvidia hardware it will serve is commercially available. Vertiv and Nvidia engineering leaders held a ribbon-cutting and working sessions at the facility, the Vertiv ATLAS Lab, on Nov. 6.

ATLAS, short for Advanced Test Lab for Architecture and Systems, is set up as a representative pod of an AI factory, with power, cooling, controls and IT equipment in one test environment. Vertiv's release says most infrastructure is validated one product at a time, while an AI factory "succeeds or fails as a system." The lab combines workload emulation, physical testing and digital twin models built in Nvidia's DSX AI Factory platform, including Omniverse libraries, DSX Sim and SimReady assets, to reproduce the power and heat behavior of current and coming Nvidia architectures.

"ATLAS lets us find and solve integration issues before they reach a customer site," said Scott Armul, Vertiv's chief product and technology officer. "We design in simulation, prove it on the floor and test ahead of the silicon, so infrastructure keeps pace with the platform."

From One Pod to a Campus

Vertiv says a design proven at pod scale becomes a validated reference design for campus-scale deployments, so the infrastructure is not re-engineered for each site. The release describes the test chain as running from grid interconnection to the rack, and from liquid cooling at the chip through heat rejection and reuse.

Vertiv's AI factory readiness application brief adds a schedule the release leaves out. It lists row-level bulk load testing in 2026 and full dynamic AI server workload emulation in early 2027. The questions it says the lab is built to answer include whether the power train responds correctly when the thermal system changes operating mode, and whether the interface between the CDU and rack-level liquid cooling performs as modeled at full density.

Why This Matters to Contractors

Vertiv's ATLAS brief separates the lab's work from factory acceptance testing, which checks each unit against its own specification, and aims instead at the seams between subsystems where control timing and flow distribution go wrong. The same brief says lab data will feed commissioning sequences and control loop parameters for fielded systems, so part of the startup sequence on a site built to an ATLAS-validated design would come from lab runs made before the equipment ships.

Sources

  1. Vertiv Holdings Co, "Vertiv ATLAS Proves AI Infrastructure Ahead of Silicon," PR Newswire, Nov. 16, 2026. prnewswire.com
  2. Vertiv, "From Design Intent to Measured Performance: The Vertiv ATLAS (Advanced Test Lab for Architecture and Systems)," AI factory readiness application brief. vertiv.com
Data Centers > Testing & ValidationVertivNvidiaAI FactoriesLiquid CoolingCDUsControlsTestingData CentersAI