NVIDIA's Vera Rubin has entered full production, with shipments officially slated to begin this fall. Foxconn has also confirmed that Rubin AI server racks will be delivered in the fourth quarter.

As per-rack power consumption keeps climbing, liquid cooling demand is rapidly expanding from mere “server cooling” to the entire rack-scale infrastructure. Supermicro's Vera Rubin NVL4 liquid-cooled solution reaches up to 362 kW per compute rack, and NVIDIA's third-generation MGX architecture now incorporates tray manifolds, rack quick-disconnect manifolds, and liquid-cooling busbars. This means a new wave of demand is being unlocked for liquid cooling components such as cold plates, quick disconnects, manifolds, and bellows.

Yet booming demand cannot mask the real difficulties on the supply side: insufficient processing yields for new materials, difficulty guaranteeing production consistency for core components, and back-end inspection efficiency becoming a capacity bottleneck. Manufacturing validation is emerging as the key gate before Rubin can ramp up smoothly.

Key Takeaways

• NVIDIA's Vera Rubin has entered full production — liquid cooling is expanding from server cooling to entire rack-scale fluid infrastructure.

• The real bottleneck is manufacturing, not design: new materials (diamond-copper, stainless steel bellows) and core parts (cold plates, UQD quick disconnects, manifolds) must be made repeatable at scale.

• Welding and helium leak testing are the true gatekeepers — finding a stable process window matters more than welding a single part.

• Equipment demand is shifting from single-process machining to a full chain: precision machining → micro-laser processing → welding → cleaning → leak testing → traceability.

edce5fd4f2deba80b7403d5275cdc5ad

Image source: Web

Liquid Cooling Demand Surges, Manufacturing Faces a Major Test

In the past, liquid cooling was mostly understood as a matter of a cold plate and a few pipes. But as AI server power keeps climbing, it is turning into a complete rack-level fluid infrastructure — and materials, components, and processes are changing along with it.

The manufacturing difficulty has shifted, starting with materials. Rubin's requirements for cooling efficiency and reliability are forcing an upgrade of the entire upstream material system.

On the chip side, one notable change is diamond-copper.

As pure copper increasingly struggles to handle extreme localized heat flux, diamond-copper — with thermal conductivity far beyond conventional copper — is being used in heat sinks and heat-spreading layers to target chip hot spots directly. Domestic players such as Ruiwei New Materials, Bozhi Jinzuan, and Sandi Precision have already moved into this space. But a material that works is not the same as a product that can be mass-produced.

Whether diamond-copper can retain stable thermal performance while achieving high processing yields after compositing, cutting, and precision machining remains a problem to solve for industrialization. This also means the importance of manufacturing capabilities such as ultra-precision machining, specialized cutting, and interface bonding is rising.

fe80fd9f2e3ef11e626436eef8973987

Image source: AI-generated

On the fluid side, a different change is taking place.

NVIDIA's Rubin platform emphasizes “no cables, no hoses,” with rigid metal manifolds and stainless steel bellows replacing traditional flexible-hose designs. This is not a simple material swap. Against demands for higher temperature, higher cleanliness, and long-term reliability, the forming, precision welding, inner-wall cleanliness, and corrosion treatment of stainless steel bellows all face much higher requirements.

Behind the material upgrade is, in essence, an upgrade of manufacturing requirements.

Core Components: The Gap From Prototype to Mass Production.The cold plate is the most direct embodiment of this change.

Early on, the industry tended to see a cold plate's value in terms of raw material costs such as copper and aluminum. But in the high-power server era, what really determines a cold plate's performance and production capability increasingly comes from flow-channel machining, welding, clean washing, performance testing, and batch-to-batch consistency.

Moreover, there is no single “one-size-fits-all” process route for cold plates.

dba894479d6efca4f7b6b79afbaec6cd

Compiled from public information

Tube-type cold plates are mature but have a ceiling on cooling capacity; stamped-and-brazed designs suit high volume but are constrained by tooling; CNC offers high flexibility but can be dragged down by cycle time; micro-channel and 3D-printed approaches push precision, cleanliness, and consistency to an even higher level.

What Rubin truly brings is not just growing cold-plate demand, but a repricing of high-end cold-plate manufacturing capability.

Across the global cold-plate supply chain, Cooler Master, AVC, BOYD, and Auras all have positions. NVIDIA's published Group A cold-plate suppliers include AVC, Cooler Master, Jentech, and Delta Electronics, among others. Auras expects to begin shipping Rubin cold-plate modules in Q3 2026, while Jentech supplies the vapor chambers.

9f3ac425ee2ec62cbfc0b0cd504654ff

Image source: Huizhou Yili Group

Quick disconnects and manifolds are moving from “single-machine components” to “rack-scale infrastructure.”

If the cold plate answers “how to cool the chip,” then quick disconnects, manifolds, and piping answer how to distribute liquid reliably across the entire rack.

Take the UQD quick disconnect: it looks like an unremarkable little part, yet its internal flow-channel precision can reach 20 microns.

This means the liquid cooling supply chain is starting to need large numbers of high-precision Swiss-type lathes, precision CNC machines, micro-hole machining, and high-consistency inspection equipment. Swiss-type lathe makers such as Tsugami, Star, and Citizen are direct beneficiaries of this demand wave. Industry research shows Tsugami received more than 2,000 liquid-cooling-related orders in the first five months of 2026, with a full-year forecast of 6,000 units, and equipment lead times have lengthened further.

6845d6e741bd71d2fd4e831993204561

In the manifold segment, demand is likewise being released rapidly along with the rack-scale architecture.

Readore, a subsidiary of Lingyi iTech, has entered the Rubin-architecture manifold ecosystem; Envicool has become a domestically certified NVIDIA NPN Tier-1 supplier; and Qiangrui Technology's Inner Manifold central distributor products are already used in Rubin servers. Cold plates solve “chip-level cooling”; quick disconnects, manifolds, and piping solve “rack-level liquid distribution.”

7177aeffd0ca8545f58d9245bfde5d0f

Image source: Web

As a result, the boundaries of liquid cooling manufacturing are expanding: from a single cold plate to an entire rack-scale fluid infrastructure. Manufacturing equipment is likewise extending from simple cutting to precision machining, laser processing, welding, clean processing, and inspection.

Component Volume Ramps Up, Equipment Demand Follows a New Logic

What truly changed in the liquid cooling supply chain is not just “more demand.” It's more components, finer processes, and longer manufacturing chains. From cold plates to manifolds, from quick disconnects to bellows, a single liquid cooling system already strings together cutting, stamping, brazing, laser welding, cleaning, and inspection.

In welding, the hard part is not “making a strong joint” — it's stabilizing the process window.

In cold-plate manufacturing today, vacuum brazing, friction stir welding, and laser welding coexist. But at mass-production scale, the real difficulty is not welding a single part — it's keeping tens of thousands or hundreds of thousands of parts consistent.

First, insufficient filler can cause incomplete penetration and weak joints, while excess filler can clog micro-channels. Second, vacuum brazing heats the whole part, so changes in the temperature profile, fixturing, and material thickness can all affect flatness and dimensional consistency. Third — and most easily overlooked — is cleanliness: oil, oxide layers, and residual particles can all compromise weld quality, and any foreign matter left inside a liquid cooling component can clog flow channels once in the system. So the core of liquid cooling welding is not welding a single part, but finding a stable process window and replicating it across mass production.

The rise of micro-channel cold plates and complex manifolds is also driving new processes such as micro-laser machining and laser welding. Metal 3D printing, meanwhile, further expands the design boundaries of cold plates, making complex 3D flow channels possible where milling, drilling, and brazing were once the limits.

76131af07e62727efeeef5c42ca6517f

Thus a new equipment demand chain is taking shape: precision machining → micro-laser processing → welding → clean processing → leak testing → automated traceability.

For inspection, the real mass-production bottleneck may lie in the back-end. Producing the part in the front-end is only the first step. For liquid cooling products, what truly determines reliable delivery is often welding and inspection — because a liquid cooling part is essentially a sealed fluid device that must guarantee structural strength, intact flow channels, and leak-free long-term operation. Welding determines whether a product carries hidden defects; helium leak testing determines whether a factory can catch them before they turn into batch scrap.

As a result, helium mass spectrometer leak detection is becoming a key inspection method for high-reliability liquid cooling products.

But as inspection precision rises, a new set of tensions emerges.

First, sensitivity versus cycle time. The finer the inspection, the longer the evacuate → helium-fill → stabilize → measure → vent sequence takes. Manual inspection can hardly match 24-hour continuous production, which is why automated helium leak testing is becoming increasingly important.

Second, per-piece inspection versus batch delivery. Suppose a cold-plate line produces 200,000 units a year with a helium-test first-pass yield of only 85% — that means about 30,000 units a year need rework or scrap. Inspection efficiency is, in itself, capacity.

Third, detecting a leak versus locating it. Truly mature production inspection doesn't just tell the factory “this part leaks” — it pinpoints whether the leak comes from a weld seam, a seal, a fitting, or the material itself. Only by feeding inspection results quickly back to the front-end process can a true manufacturing loop form. Companies such as Wanyi Technology and KYKY are already active in helium mass spectrometer leak detection and air-tightness testing for AI server liquid cooling systems.

As liquid cooling enters mass production, competition ultimately comes down to a few questions: Can precision machining efficiency improve? Can welding yield stabilize? Can helium leak testing evolve from a single inspection step into a true mass-production capability?

Can full-process traceability be built from machining to inspection? These are the questions the liquid cooling supply chain must answer as it moves from “technology validation” to “manufacturing validation.”

•These are exactly the manufacturing scenarios that ITES 2027 will focus on.

•Building on South China's liquid cooling industry cluster, ITES will connect upstream and downstream resources and leverage its strength in precision manufacturing, coordinating 500+ enterprises in metal materials, precision equipment, precision components, sheet-metal processing equipment, and welding and inspection equipment. Focusing on the liquid cooling server component manufacturing chain — and targeting the pain points in machining, welding, and inspection for AI server sheet metal, cold plates, UQDs, and manifolds — ITES will host a series of themed forums and set up an AI Server & Liquid Cooling Core Parts Machining Technology Zone to deeply link South China and overseas liquid cooling industry resources. Below is the effect rendering of the liquid cooling forum and the AI Server & Liquid Cooling Core Parts Machining Technology Zone.

•Forum topics:

•AIDC trend sharing

•Cold Plate Welding & Inspection Innovation Forum

•Manifold Welding & Inspection Innovation Forum

•Manifold & Liquid Cooling Distribution Machining Session

•Tube Processing & Piping Systems Session

•Shaft Parts (UQD) Machining Technology Session

•AI-Empowered Sheet Metal Manufacturing Innovation Launch, and more

•Who we serve:

•Liquid cooling total solutions, CDU units and equipment

•Micro-channel cold plates / heat exchangers / thermal structural parts

•AI server systems / ODM / OEM turnkey providers

•Liquid cooling manifold R&D, machining, and inspection vendors

Three ways to participate — attending, speaking, and sponsorship or exhibition booths in the liquid cooling zone — are now open for registration. We look forward to your participation!

As NVIDIA Rubin Ramps Up, the Liquid Cooling Supply Chain Races Toward “Manufacturing Validation”