Skip to research
stock3dby Fynca
Explore in 3D

COMPANY RESEARCH / 3324.TWO

Auras Technology

Cold plates, manifolds, quick disconnects and CDUs

Published by FyncaResearch checked How we research

Where it fits in an AI server rack

Its Aug 2026 investor presentation shows an “AI Liquid Cooling Solution” of cold plates, DIMM cold plates, internal, rack and AI manifolds, quick disconnects, liquid-to-liquid and liquid-to-air CDUs, an in-row CDU, a sidecar, a building manifold and a rack, beside heat-pipe, vapor-chamber and 3D-VC air-cooling modules. Servers were 78% of 2Q26 revenue (NT$8.70 billion, +64% y/y); NVIDIA named Auras among its Blackwell partners (Oct 2024).

What the evidence establishes

AI-rack evidence

Disclosed: Auras’s Aug 2026 investor presentation (MOPS) shows its AI liquid-cooling line-up and puts servers at 78% of 2Q26 revenue.

An explicit AI-server or AI-rack product, design-in, supply role, program, reference design or partnership, such as a part named for rack-scale AI systems or a product sold for AI data centers. For a rack builder, it means the company’s own AI servers or racks. This does not establish revenue, market share, volumes or future sourcing.

Component roleCold plates, manifolds, quick disconnects and CDUs
EvidenceDisclosed: Auras’s Aug 2026 investor presentation (MOPS) shows its AI liquid-cooling line-up and puts servers at 78% of 2Q26 revenue.
How we knowDisclosed The company’s own filing, release, annual report, product page or investor record describes the AI-server or AI-rack role. Where customers are unnamed, the listing says so.
Primary sourceAuras Technology — supporting disclosure
Company or official disclosure; vendor claims are attributed to their source.
Checked. This is the research review date, not the source publication or quote time.

Keep in perspective

Customers and liquid-cooling revenue are not disclosed. The same slide shows a 450 kW power shelf (DC48V/DC400V) that is not mapped, because its role is not described.

Specialist business exposure. This describes the breadth or role of the business; it is not an investment rating. An AI-server product, design-in or partnership does not by itself establish material AI-rack revenue, market share, or future returns.

Connections to the assembly

These links explain the technology relationship. The model is illustrative and does not represent an actual manufacturer’s bill of materials.

Cold plates

Copper plates with fine internal channels mounted on the GPUs and CPUs. Coolant flowing through the channels carries the chips’ heat away.

Scope of the company links: Cold-plate makers with documented AI-server products: AVC (large-scale AI liquid-cooling production; a reference designer for unnamed leading GPU makers), Motivair by Schneider Electric (cold plates for NVIDIA SXM5 and AMD Instinct MI300X), JetCool, a Flex company (SmartPlate for NVIDIA H100 and RTX PRO 6000 and Broadcom TH6 switches), CoolIT, owned by Ecolab since July 2, 2026 (5+ million cold plates shipped), Auras, Delta (microchannel cold plates), Envicool, Jentech (liquid-cooling modules), SUNON (shipping since 1H 2026) and Forcecon (catalog). Eaton added Boyd Thermal’s cold plates when it closed that purchase (Mar 12, 2026); Supermicro markets its own DLC-2 cold plates; Luxshare lists cold plates for AI racks; Furukawa is adding cold-plate capacity for data-center GPUs from January 2027. The cold-plate supplier in any given NVL72 tray is not disclosed. Cooler Master and ZutaCore are private, and Fositek’s “Cold Plate + UQD” offer relies on AVC, so Fositek is mapped to quick disconnects only.

Rack manifolds

Vertical supply and return pipes at the back of the rack that distribute cool liquid to every tray and collect the warm liquid.

Scope of the company links: Kaori Thermal’s MGX and Vera Rubin rack manifolds are marketed as compliant with NVIDIA’s NVL72 reference designs, and nVent’s manifolds are part of the GB200 NVL72 reference architecture it defined with NVIDIA. Auras (rack, AI and internal manifolds), AVC (manifold systems), Envicool, Delta, Motivair by Schneider Electric (in-rack manifolds), CoolIT (Ecolab), JetCool (Flex), Nidec (coolant distribution manifolds) and Forcecon (catalog) also document manifolds. NVIDIA’s GB200 NVL72 design uses a blind-mate liquid-cooling manifold (NVIDIA, Oct 2024); which supplier builds it for a given rack is not disclosed. Eaton (Boyd Thermal), Supermicro and Luxshare also list rack manifolds for liquid-cooled AI racks.

Quick disconnects

Dripless couplings at each tray connection. They seal both sides when unplugged, so a tray can be removed without draining the loop.

Scope of the company links: Fositek says its UQDs entered the GB200/GB300 supply chain and passed Rubin testing (2025 annual report); Envicool says NVIDIA listed its UQDs (OCP 2024) and its UQDs and MQDs (OCP 2025) among MGX ecosystem partners; Auras shows quick disconnects in its AI liquid-cooling line-up; Amphenol markets UQD, UQDB, BMQC and MQD couplings for AI data-center liquid cooling; Dover’s CPC is an NVIDIA-named Blackwell partner and Dover’s 10-K cites connectors for liquid cooling of server racks; JetCool (Flex) offers manifolds and quick disconnects. Parker was not verified as a quick-disconnect supplier in our sources, and Nidec’s quick couplings are made for its own CDUs. Stäubli and Danfoss, both NVIDIA-named, are private. Luxshare lists blind-mate quick connectors for AI racks.

Coolant distribution unit

Pumps, a heat exchanger, filters and controls that circulate the rack’s clean coolant and pass its heat to the building’s water loop.

Scope of the company links: In-rack CDUs like the illustrated one sit at the bottom of the rack: Kaori (4U in-rack CDU), Auras, SUNON (in-rack RPU/CDU, mass-production ready in 1H 2026), JetCool (Flex; SmartSense rack-mounted CDU), Envicool (rack CDU validated by Intel) and Forcecon (80 kW, catalog). Row- and facility-scale CDUs serve many racks: nVent (in its GB200 NVL72 reference architecture with NVIDIA), Delta (40–144 kW liquid-to-air; 1.5 MW liquid-to-liquid), Motivair by Schneider Electric (MCDU-70 in the Schneider–AMD Helios reference design; WCDU up to 3.5 MW), CoolIT (Ecolab), Vertiv (Liebert XDU), AVC, Nidec, Modine’s Airedale (400 kW–2 MW+) and Lite-On (a 600 kW in-row CDU announced in 2024). Sanyo Denki supplies a liquid-cooling pump, a CDU component rather than a CDU. LiquidStack is private. Eaton’s Boyd Thermal markets CDUs for GB200 NVL72 deployments, Supermicro’s GB300 NVL72 rack carries its own in-rack CDU of up to 250 kW, and Luxshare lists CDUs.

Air heat sinks

Finned heat sinks and vapor chambers on network chips, power parts and other components that are not on cold plates.

Scope of the company links: Heat sinks, heat pipes and (3D) vapor chambers for servers: AVC, Delta (two-phase vapor-chamber heat sinks for AI GPU/ASIC chips and power racks), Auras (heat-pipe, vapor-chamber and 3D-VC modules), SUNON, Nidec and Forcecon (server thermal modules, catalog). In a liquid-cooled tray these cover the parts left on air (NICs, drives, power stages); the supplier for any given rack is not disclosed. Jentech’s heat spreaders are package lids, not heat sinks, so they are not mapped here. Furukawa began new data-center heat-sink output in July 2026.

Related research

Companies with overlapping sector exposure. Their products and evidence may differ; this is not a list of equivalent investments.

Suggest a correction · Research and market-data disclosures