www.ptreview.co.uk
22
'26
Written on Modified on
NVIDIA Partners Accelerate Sustainable AI Factory Deployments Across Europe
European ecosystem partners deploy the NVIDIA DSX platform to build water-efficient, grid-responsive AI factories optimized across power and cooling domains.
www.nvidia.com
Accelerating computational demands for generative artificial intelligence and frontier foundation models have placed acute pressure on global utility grids and regional water resources. Conventional hyperscale facilities rely heavily on evaporative cooling towers, consuming millions of liters of potable water daily to reject heat from dense computing rows. Simultaneously, rigid facility power provisioning caps limit the number of active graphic processing units that can be energized within fixed substation interconnects, slowing the speed at which cloud builders can turn capital expenditure into productive capacity. Furthermore, regional energy regulators and grid operators increasingly mandate that multi-megawatt computing facilities function as flexible loads capable of shedding power during peak demand periods. Resolving these operational constraints demands codesigned reference architectures that unify accelerated computing clusters with facility power distribution, closed-loop thermal rejection, and dynamic demand-response telemetry.
Closed-Loop Thermal Architectures and Water Conservation
To mitigate critical infrastructure footprints across Europe, leading design, energy, and cloud providers have adopted the NVIDIA DSX AI Factory Platform. Built as a full-stack reference architecture, the platform codesigns compute, networking, storage, software, power delivery, and thermal systems from initial site blueprinting through live runtime operations.
A primary technical pillar of the architecture is the elimination of evaporative water consumption. Conventional cooling towers reject thermal loads through continuous water evaporation, requiring ongoing makeup water feeds. The DSX reference designs transition facilities to closed-loop liquid cooling paired with exterior dry-cooler radiators. Engineered for the NVIDIA Vera Rubin architecture, the system supports 100 percent direct-to-chip liquid cooling with facility coolant supply temperatures operating up to 45 degrees Celsius.
This high-temperature threshold allows outdoor dry coolers to transfer heat directly into ambient air without requiring evaporative phase changes, reducing facility cooling water consumption to near zero. Under baseline modeling, a 2-gigawatt deployment using DSX architecture eliminates up to 54 million liters of water consumption daily compared to traditional cooling tower configurations, equivalent to the daily municipal water usage of 180,000 households.
Dynamic Power Optimization and Grid-Responsive Workload Coordination
To maximize throughput within strict power allocations, the platform incorporates DSX MaxLPS for dynamic power budgeting and DSX Flex for bidirectional utility grid interaction:
- DSX MaxLPS: Rather than maintaining static worst-case power margins across computing nodes, the software dynamically allocates power within provisioned thresholds, enabling operators to generate equivalent token output with up to 30 percent less electrical power. In early proof-of-concept testing across NVIDIA HGX B200 systems, cloud provider Lambda demonstrated a 24 percent increase in cluster token throughput alongside a 19 percent expansion in managed computing capacity within a fixed power budget. At Nscale's Icelandic facility, deploying DSX MaxLPS across four NVIDIA GB300 NVL72 racks yielded a 49 percent improvement in inference throughput and a 37 percent expansion in managed cluster capacity under a constant 264-kilowatt power budget.
- DSX Flex: The telemetry module coordinates cluster consumption with utility dispatch needs, spot electricity pricing, and demand-response signals. By synchronizing grid feeds, on-site battery energy storage systems, and local renewable generation, the platform transforms compute clusters into dispatchable loads. In pilot evaluations conducted with Emerald AI, EPRI, Nebius, and the United Kingdom National Grid on an NVIDIA Blackwell Ultra GPU cluster, the system achieved power draw reductions of up to 40 percent during simulated grid stress events while sustaining non-deferrable mission-critical compute tasks.
Simulation-Ready Digital Twins and Modular Infrastructure Software
To accelerate site commissioning, the platform couples simulation tools with modular infrastructure management software. Through DSX Sim and the newly launched SimReady Central Library, engineering teams evaluate validated digital assets from partners such as Vertiv and PTC to simulate structural layout, power distribution, fluid dynamics, and acoustic profiles before breaking ground. Nscale is leveraging these digital twin capabilities to de-risk facility rollouts across Iceland and Portugal.
At the operating system layer, NVIDIA introduced three modular, open-source DSX OS components in developer preview to automate runtime environments:
To accelerate site commissioning, the platform couples simulation tools with modular infrastructure management software. Through DSX Sim and the newly launched SimReady Central Library, engineering teams evaluate validated digital assets from partners such as Vertiv and PTC to simulate structural layout, power distribution, fluid dynamics, and acoustic profiles before breaking ground. Nscale is leveraging these digital twin capabilities to de-risk facility rollouts across Iceland and Portugal.
At the operating system layer, NVIDIA introduced three modular, open-source DSX OS components in developer preview to automate runtime environments:
- NVIDIA Config Manager: Automates multi-tier AI factory networking topologies.
- NVIDIA Rack Management Service: Consolidates rack-level inventory tracking, power routing, firmware flashing, and switch telemetry.
- NVIDIA Cluster Readiness Engine: Executes distributed synthetic workloads to validate node-level hardware and interconnect fabrics, isolating faults prior to production onboarding.
These modules work in tandem with NVIDIA Infrastructure Controller, NVIDIA AI Cluster Runtime, and NVSentinel to establish automated root-cause isolation and fleet lifecycle governance for cloud operators, including CoreWeave, Nebius, Crusoe, and Firmus.
Cross-Ecosystem Industrial Integration
The European buildout is supported by coordinated engineering deployments across industrial technology providers:
Cross-Ecosystem Industrial Integration
The European buildout is supported by coordinated engineering deployments across industrial technology providers:
- Facility Infrastructure: Schneider Electric is delivering power and liquid cooling designs for Vera Rubin infrastructure alongside AVEVA lifecycle digital twins. Siemens is integrating DSX Sim for acoustic and thermal modeling, while Dassault Systèmes and Vertiv are deploying model-based systems engineering models for Vertiv OneCore modules. ABB is developing SimReady components to evaluate source-to-rack pathways while advancing 800-volt direct current architectures via its Infinitus platform. Procore is integrating its construction management suite with DSX Sim to streamline field execution.
- Pre-Engineered Power and Cooling: Trane Technologies and Eaton collaborated on integrated grid-to-chip designs combining Eaton Beam power platforms with Trane Continuum cooling systems. Trane also released seven thermal management reference designs, including three 250-megawatt blueprints.
- Qualified Hardware: Equipment verified under the NVIDIA DSX Ready program includes the Trane Technologies LiquidStack GigaModular and Vertiv 2.3-megawatt CoolChip coolant distribution units, alongside Hitachi Energy battery energy storage systems. Grid connection planning is being advanced in tandem with GE Vernova, Hitachi Energy, and Siemens Energy.
Additional Context
This section details technical specifications and competitive benchmarking not included in the original product announcement.
In high-density data center engineering governed by ASHRAE TC 9.9 guidelines, thermal management systems are categorized by facility water supply temperatures, ranging from Class W1 (17 to 27 degrees Celsius) to Class W5 (greater than 45 degrees Celsius). Standard direct-to-chip liquid cooling loops traditionally operate within W2 or W3 thresholds (up to 32 degrees Celsius), which necessitates mechanical chillers or adiabatic misting systems when ambient outdoor air temperatures rise during summer peaks. By engineering cold plates and internal fluid paths on the Vera Rubin architecture to support inlet water temperatures reaching 45 degrees Celsius (Class W4/W5 compliance), DSX enables continuous economizer dry cooling across nearly all European climatic zones, completely bypassing mechanical refrigeration compressors and eliminating evaporative plume losses.
Within the accelerated computing infrastructure landscape, the DSX platform establishes an integrated framework competing directly with open hardware specifications, such as the Open Compute Project Open Rack Wide and Open Accelerator Infrastructure standards, as well as proprietary turnkey pods like Google Cloud TPU hypercomputer architectures and AWS UltraClusters. While traditional cloud facilities treat mechanical cooling distribution, medium-voltage switchgear, and workload scheduling as decoupled systems governed by static building management system protocols (such as BACnet or Modbus TCP), DSX Exchange binds facility sensors directly to orchestrator layers through telemetry APIs. This enables dynamic thermal throttling, rack-level power capping, and automated workload migration based on localized thermal headroom and grid frequency response, achieving power usage effectiveness and water usage effectiveness metrics that establish a benchmark for multi-gigawatt European AI hubs.
Edited by Romila DSilva, Induportals Editor, with AI assistance.
This section details technical specifications and competitive benchmarking not included in the original product announcement.
In high-density data center engineering governed by ASHRAE TC 9.9 guidelines, thermal management systems are categorized by facility water supply temperatures, ranging from Class W1 (17 to 27 degrees Celsius) to Class W5 (greater than 45 degrees Celsius). Standard direct-to-chip liquid cooling loops traditionally operate within W2 or W3 thresholds (up to 32 degrees Celsius), which necessitates mechanical chillers or adiabatic misting systems when ambient outdoor air temperatures rise during summer peaks. By engineering cold plates and internal fluid paths on the Vera Rubin architecture to support inlet water temperatures reaching 45 degrees Celsius (Class W4/W5 compliance), DSX enables continuous economizer dry cooling across nearly all European climatic zones, completely bypassing mechanical refrigeration compressors and eliminating evaporative plume losses.
Within the accelerated computing infrastructure landscape, the DSX platform establishes an integrated framework competing directly with open hardware specifications, such as the Open Compute Project Open Rack Wide and Open Accelerator Infrastructure standards, as well as proprietary turnkey pods like Google Cloud TPU hypercomputer architectures and AWS UltraClusters. While traditional cloud facilities treat mechanical cooling distribution, medium-voltage switchgear, and workload scheduling as decoupled systems governed by static building management system protocols (such as BACnet or Modbus TCP), DSX Exchange binds facility sensors directly to orchestrator layers through telemetry APIs. This enables dynamic thermal throttling, rack-level power capping, and automated workload migration based on localized thermal headroom and grid frequency response, achieving power usage effectiveness and water usage effectiveness metrics that establish a benchmark for multi-gigawatt European AI hubs.
Edited by Romila DSilva, Induportals Editor, with AI assistance.
