Making every drop count in efficient liquid cooling systems
Matt McLaughlin, corporate account manager for Ecolab’s data centre business, talks to Capacity about the ins and outs of water and liquid cooling.
Pulling the wire
Desk
Technology
The long desk. Cooling, the silicon in the rack, the 800 VDC architecture, and rack density stay on the page. The other desks are the day’s file.
United States, the last day, plus 3 filings from this week still large enough to keep: a blockbuster, at least $250 million, or at least 300 MW. A foreign filing is here only as a blockbuster. A story sits under the chapter it concerns. The table is the same plant as Indexes.
Cooling
The water the rack is built for
Vera Rubin NVL72 is built to run fully liquid-cooled at a 45°C water inlet, on a liquid-cooled busbar rated to 5,000 A. A hall already on liquid cooling does not redesign the CDU plant to take it. GB200 NVL72 and GB300 NVL72 are liquid-cooled on the same MGX footprint. Kyber stands up to 18 vertical blades in the chassis. The inlet and the busbar are NVIDIA’s rack note of 16 March 2026. The blade count is the OCP note of 13 October 2025.
- Inlet
- 45°C liquid
- GB200 and GB300
- Liquid. MGX footprint.
- Kyber
- 18 vertical blades
Silicon
What is actually in the rack
GB200 NVL72 and GB300 NVL72 each put 72 GPUs in the rack. Vera Rubin NVL72 ships in the second half of 2026 with 72 Rubin GPUs and 36 Vera CPUs. Kyber doubles that rack to 144 GPUs. NVL576 is eight racks of 72, not one chassis. Feynman on Kyber is a scale-up to 1,152 GPUs, and NVIDIA has not dated that rack. This is the silicon the hall has to cool and feed. A foundry, a memory stack, and a supply contract stay on Chips & Supply.
- GB200
- 72 GPUs
- GB300
- 72 GPUs
- Vera Rubin
- 72 Rubin GPUs, 36 Vera CPUs
- Kyber
- 144 GPUs
- Feynman
- Scale-up to 1,152 GPUs
VDC architecture
Where the row stops being AC
An MGX 800 VDC power rack in the second half of 2026 takes the building’s existing AC and hands 800 VDC to the compute racks in the row. The building electrical system stays. The row power center, expected in 2027, is an overhead 800 VDC busway up to 2 MW per row. Kyber moves the in-rack parts from 54 VDC to 800 VDC. The same copper carries over 150 percent more power, which is how a 200 kg busbar comes out. A hall still on the drawing board is the one that takes medium voltage straight to 800 VDC. The 2 MW figure is the row bus.
- In the row
- 800 VDC power rack, H2 2026
- Row bus
- Up to 2 MW, expected 2027
- In the rack
- 54 VDC to 800 VDC
Microchip, Navitas Unveil 800V Power Reference Design for AI Servers
Microchip Technology and Navitas Semiconductor have unveiled a power-conversion reference design for artificial intelligence (AI) data centers.
ABB launches range of DC power products
Vendors looking to be ready for 1MW racks and 800vDC architectures
Rack density
What one rack asks of the hall
NVIDIA’s DGX GB rack guide states about 120 kW of rack consumption. The GB300 NVL72 reference states a full rack of up to 142 kW. The DSX facilities guide states a 330 kW cabinet TDP for Vera Rubin NVL72, and it scales that design point from an MGX Gen 1.1 cabinet at 198 kW. Each figure stays in the words its source used. A megawatt of that stated rack power is 8.3 GB200 racks, 7.0 GB300 racks, or 3.0 Vera Rubin racks. Kyber and Feynman have no stated kilowatt. NVIDIA’s DSX example site is a 972,500 square-foot building for a 250 MW-class IT load. That reading is 3,890 square feet of building per megawatt, and 2.57 megawatts per 10,000 square feet. It is the whole building in that reference design.
- Vera Rubin
- 330 kW TDP
- The building
- 3,890 SF per MW
The rack
Five generations, one table
Consumption, a full-rack requirement, and a cabinet TDP are not one series. Kyber and Feynman have no stated kilowatt. The same table is on Indexes.
Swipe to read across
The file
The rest of the systems wire
Networks, servers, and the other systems filings that are not a cooling, silicon, VDC, or density story.
Dell Expands AI Data Platform to Ground Agents in Enterprise Context
Dell AI Data Platform, the data foundation of the Dell AI Factory, expands with new data orchestration and storage capabilities ROUND ROCK, Texas, Oct.
Cisco Expands Validated SONiC Across Nexus N9000 for Open AI Fabrics
Cisco is expanding its validated SONiC network operating system across the Nexus N9000 portfolio, creating a supported open-networking model that spans both conventional leaf-spine networks and high-performance AI…
Vast uses tiered storage to ease AI agent memory demands
AI agent memory is creating new demands on infrastructure as agents run longer sessions and spread across the enterprise.
BattleBots builders turn to AI simulation to win fights before entering the box
AI infrastructure built on dense graphics processing unit clusters and high-speed AI networking is no longer reserved for frontier model labs.
LibreOffice says ‘no AI’ is now a software feature
The maker of the open source document editor says it has no plans to add AI to its software's default configuration, citing user privacy.
ALCF Expands Aurora’s DAOS Storage to 800 Servers
Among the world’s fastest storage systems, DAOS delivers high performance across thousands of Aurora nodes to support data-intensive workloads ranging from massive simulations to AI model training.
Anthropic opens its most powerful AI models to more security teams
Its partners under Glasswing, an initiative aimed at securing the world's most critical software, found at least 129,000 verified vulnerabilities between April and July.
Maryland nets 440 MW/1,760 MWh in first bulk energy storage procurement
The first round of awards under the Maryland Next Generation Energy Act fell short of the state’s 800-MW target as the PJM Interconnection’s capacity crunch continues.
Infleqtion and Partners Build Optical Cavity Prototype to Shrink Quantum Sensors
New prototype reduces footprint needed for laser stabilization, paving way for compact, fieldable quantum systems and chip scale manufacturing LOUISVILLE, Colo., Oct.
Asia-Pacific
India to Invest 500 Billion Rupees in BESS by 2033
India is stepping up investment in battery energy storage systems (BESS) to stabilize its power grid.
Building Resilient Data Centers for the AI Era with Scott Longhurst
TL;DR AI is reshaping data center operations, driving higher rack densities, increased power requirements, and a greater demand for liquid-to-chip cooling expertise.
When One Datacenter Is No Longer Enough
SPONSORED POST: Why training AI across multiple sites is creating a new networking challenge
Precisely launches unified AI data platform and AI Studio library of agents and apps
Data management software company Precisely Software Inc.
AOI Raises $588M to Fund Manufacturing Expansion
Applied Optoelectronics (NASDAQ: AAOI) has completed a previously announced $600 million at-the-market equity offering program, raising approximately $588 million in net proceeds as the optical component manufacturer…
HPE’s Rahim: Pace of change in the data center is ‘extraordinary’
Following its record-setting third quarter , HPE said it expects cumulative AI network orders for fiscal 2026 to top $3 billion, exceeding its previous forecasts.
HPE Targets 50%+ Data Center Networking Growth as AI Orders Top $3 Billion
HPE is setting aggressive growth targets for its Networking business as the company moves the former Juniper portfolio deeper into AI data centers, projecting low-to-high 50% compound annual growth for Data Center…
