Extrcool Industry Limited company News and Trends items

EXTRCOOL supplies liquid cooling infrastructure for data centers and intelligent compute scenarios, covering new builds and existing room retrofits, with a product line built on the cold plate route: single and double-row liquid cooling micro-modules for high-density deployment and dedicated liquid cooling racks. Piping and key components are assembled in the factory, so...

Rack power climbing past 100kW leaves air cooling exhausted beyond the 12kW critical point, and liquid cooling has moved from a better choice to a necessary condition: liquid carries 3,000 times the heat of air at equal volume, cuts noise by 10 to 15 decibels and saves about 30% electricity, with cold plate holding over 90% of the market while immersion reaches PUE 1.05 ...

Blackwell and Vera Rubin platforms push rack power past the hundred-kilowatt line, making liquid cooling a core layer of AI infrastructure, and a Rubin-class NVL72 rack carries roughly 140,000 to 160,000 dollars of liquid cooling value: the CDU at 80,000 to 90,000 as the largest item, cold plates at 35,000 with the highest technical barrier, and connectors and piping at ...

Container liquid cooling fuses liquid cooling with the modular data center: cooling system, IT equipment and infrastructure sit inside a standard container, adding fast deployment, flexible expansion and easy transport, with single racks running 10 to 20kW, PUE dropping to 1.05 on cold plate and 1.03 on immersion, deployable on rooftops, basements and remote sites. Cold ...

RuiSu's mobile liquid cooling compute pod uses a standard container as the carrier and integrates compute, cooling, power distribution, security and O&M into five factory-prefabricated systems, delivered whole and plug-and-play on site. The compute core runs cold plate or full immersion with PUE held at 1.1 to 1.15. Power supports 380V three-phase and 800V high-voltage D...

Seven ministries have jointly issued the implementation plan for digital-green synergistic transformation from 2026 to 2030, and the document sends a clear signal: liquid cooling heat dissipation is the big trend, backing innovation in liquid cooling, waste heat recovery and 800V high-voltage DC supply while promoting scaled deployment of single racks above a hundred kil...

Single GPUs have passed 1000 watts and air cooling is near its physical limit, pushing liquid cooling into the spotlight of compute infrastructure: CICC expects the 2026 global AI liquid cooling market to exceed 114.7 billion yuan at 273% growth, CCID puts the 2025 China market at 15.98 billion yuan up 45.2%, and orders queue to 2027 with single contracts of 2.7 billion....

Traditional data center construction spans 18 to 24 months against an AI chip iteration cycle of about 12 months, and rooms fall into the built-and-already-outdated trap. Yunjian's iMDC answers with full-stack productization: power distribution, cooling, management, fire protection and cabling integrate into one factory pre-installed and pre-commissioned system, and the ...

Midea Group announced that its Shunde liquid cooling manufacturing base will start production in August 2027 with over one billion yuan of investment, scaling natural cooling maglev chillers and CDUs in-house after a strategic agreement with Alibaba to co-develop liquid cooling temperature control for high-load AI clusters, the appliance giant's product-first shift towar...

GPU and NPU power keep climbing and NVIDIA's GB200 and GB300 generations have moved to rack-scale liquid designs, shifting the question from whether to adopt liquid cooling to how to plan it: rooms will run hybrid, high-density chips on liquid and storage, network and low-heat parts on air. The CDU holds the decisive position, connecting the IT-side loop with the facilit...

EXTRCOOL fuses the micro-module's agile architecture with liquid cooling to deliver a packaged solution with four claims: urgent-launch speed, gradual investment, over 50% electricity saving and lower O&M labor. Power, liquid cooling and monitoring systems are prefabricated, integrated and full-load tested in the factory, leaving only modular quick connection on site, sh...

AI training chips push single racks from a few hundred watts per server to tens of kilowatts per cabinet, a full AI rack hitting 40 to 50kW against the traditional 8 to 15kW design standard, and air carries roughly one 3,500th of the heat of liquid, so air-cooled rooms spend 0.4 to 0.6 kilowatt-hours on cooling for every one on compute while liquid cooling drops PUE to 1...

EXTRCOOL answers fast compute demand with a 45-foot fully prefabricated liquid cooling container data center: power distribution, compute, liquid cooling, automation and security integrate into one standard container with assembly, joint commissioning and full testing completed in the factory, and the site only connects water, electricity and network, reaching production...

JPMorgan forecasts China's AI inference token consumption rising from about 10 quadrillion in 2025 to roughly 3,900 quadrillion by 2030, a near 370-fold jump, with ByteDance's Doubao passing 50 trillion daily tokens, and token growth pushes data centers toward high density where air cooling cannot follow, leaving liquid cooling as the key support of AI compute infrastruc...

Blackwell architecture pushes whole-rack heat load past a hundred kilowatts, where air cooling hits the ceiling, and the liquid cooling compute pod, a prefabricated unit built on a standard container cabin, answers with cold plate or immersion cooling and PUE down to 1.08 against a national average of 1.34. Coolant conducts about 25 times better than air, carrying 60 to ...

ByteDance has set this year's AIDC tender core standards: 21kW as the starting threshold, 50kW as the mainstream high-end, and every rack above 21kW must take liquid cooling, with 95% of racks at 21kW and above. Liquid carries heat about 3,000 times more effectively than air, and past 20kW air cooling survives only on a fan spiral that pushes PUE up and ages hardware. Co...

Single-chip power in the hundreds of watts has pushed air cooling past its noise and energy ceiling, and the industry splits mainstream liquid cooling into three routes by how coolant touches the heat source: cold plate with indirect contact, immersion and spray with direct contact. Cold plate attaches a plate to the CPU and GPU surface without core hardware redesign and...

China's 2025 new cabinet demand reaches 750,000 units, with 380,000 above 10kW, putting the new liquid cooling market near 45 billion yuan. A fully liquid-cooled cabinet holds up to 160 CPUs at 100kW per cabinet, ten times a traditional data center, and a natural-cooling rear door exchanges 20kW for less than 2kW with COP beyond 10. Four-layer leak protection monitors ev...

Once a single rack passes 80kW and reaches 100kW or more, the immersion advantage becomes visible, and IDC puts the China liquid-cooled server market at 8.9 billion dollars by 2027 on a 55% compound rate. Liquid conducts heat over 1,000 times better than air, and the routes split into cold plate with about 90% of the market, immersion and spray. IT and communications tak...

Single-chip TDP heads toward 1000W with the B200, and single-phase open tank immersion leads large-scale deployment on reliability and cost, with GRC's ICEraQ setting the paradigm: Schneider alumni founded it in 2009 after air cooling proved a wasteful compromise, Texas mining farms proved servers run immersed with PUE dropping from 1.6 toward 1.03, and Samsung's Catalys...

EXTRCOOL has launched a 45-foot fully prefabricated liquid cooling data center for AI high-density scenarios, integrating power distribution, compute, liquid cooling, automation and security in a standardized cabin jointly tested at the factory. The site avoids heavy civil works and goes live within 72 hours after water, power and network connections, cutting the deploym...

Guizhou Grid has put live the province's first immersion cooling container data center in the Guian New Area: one retrofitted container holding four immersion racks, a closed cooling tower and black pipes, a plug-and-play compute node with no construction crews. Immersion drops the whole machine into insulating fluorinated fluid that contacts every heat source, and four ...

Per-rack power density has jumped from 5 to 8kW to 50kW and beyond 100kW, making liquid cooling the standard fit for new AI centers. Cold plate leads with over 90% share: coolant circulates in closed channels on the CPU and GPU with air assisting memory, supporting 40 to 100kW per rack in one to two days of retrofit. Single-phase immersion reaches PUE of 1.08 to 1.15, an...

XINKE's AquaPlex-45 turns the data center into a standardized prefabricated product: engineers integrate liquid-cooled racks, the CDU, power distribution, fire protection and an intelligent O&M platform inside a 45-foot container, completing wiring, joint commissioning and full-load pressure testing before shipment, and the site only connects water, electricity and netwo...

Behind every smart conversation sits a data center running at full speed, and a single AI chip draws 1400 watts with one rack giving off as much heat as a row of machine rooms used to, making liquid cooling standard for new AI computing centers. Stainless steel liquid cooling piping carries chip heat from CDU to cooling tower, the invisible blood vessel of the computing ...

Liquid cooling is a system reconstruction from chip to facility, pulling pumps, connectors, coolants and HVAC along. The GB200 NVL72 runs about 120kW per rack, and IEA projects data center electricity to about 950TWh by 2030. Three routes coexist: cold plate on high-power chips, immersion for HPC, and rear door units for retrofits. Global liquid cooling equipment heads f...

Data center energy use is expected to triple over the next decade, and the rear door heat exchanger offers rack-level cooling without major rework: replacing the cabinet's rear door with liquid cooling coils captures server exhaust before it enters the room, and the heat leaves through a closed liquid loop to a dry cooler or cooling tower, cutting mechanical cooling ener...

The most expensive part is the CDU, accounting for 30% to 40% of system cost. Three functions define it: heat exchange, flow drive and pressure stabilization, and a well-built unit holds approach temperature within 1°C, carrying 15% to 20% more heat. One operator saw pump impeller cavitation six months after delivery because saturation vapor pressure was ignored. Selecti...

A CDU transfers heat from one fluid source to another in a controlled way, separating the primary facility water loop from the secondary compute cooling loop. The pump is the core: centrifugal designs dominate, selected by operating point across efficiency, compactness and reliability. The brazed plate heat exchanger delivers the best duty per area, scaling to 2.5MW in i...

AI server racks have passed 1MW and the hub that decides whether cooling capacity reaches every chip precisely is the CDU, with rack-mounted, cabinet and air-liquid forms spanning 10kW to 2MW. Three hard indicators define the line: outlet temperature held within ±0.5°C steady state, flow regulation from 0 to 100%, plus dual-pump hot swap and online maintenance. The first...

Spec sheets lose comparability because parameters describe performance under each vendor's test conditions, and a COP of 5.0 at the design point may drop to 3.5 at part load while another unit at 4.5 design COP holds 4.0 and wins on annual efficiency. IT load averages 40% to 70% of nameplate, so part-load curves matter more than the full-load point, separating ±0.5°C fro...

The manifold splits one incoming flow into dozens of branches, one cold plate per outlet. The body is a solid block of copper or stainless. Nine in ten faults sit in machining: a hole off by 0.02mm or carrying a burr starves its cold plate. Every outlet carries an O-ring seal groove, since tool marks at the groove bottom weep after installation, and mounting face flatnes...

The manifold carries distribution, collection and regulation duties. Three roles define it: spreading CDU coolant evenly to 10 to 20 cold plates, collecting the return flow and raising branch flow as GPU load climbs, with domestic manifolds holding 16-branch deviation within ±3%. Sensors alarm below 80% of design flow, and modules stack from 8-16 to 32-64 branches. Mater...

A 1MW secondary CDU quotes about 180,000 yuan but settles around 400,000 at a northwest AI computing center, and the gap is not windfall profit: the 180,000 covers the CDU cabinet alone, while the rest layers channel fees of about 20%, EPC costs of 15% to 45% covering transport, technicians, commissioning and spares, plus settlement of 9% VAT, contingency and warranty de...

The CDU is the control hub between primary cold source and secondary IT equipment, running plate heat exchange with hydraulic isolation and leak detection with N+1 redundancy. Dongwu Securities puts the global CDU market at about $3.6 billion in 2026 and $33.4 billion by 2030, with the domestic market past RMB 9.2 billion in 2025; Envicool leads cold-plate delivery at 30...

The manifold sits at the rack rear between the CDU and the server cold plates. The CDU handles total flow; the manifold splits coolant evenly to every cold plate layer, gathers it after heat absorption and returns it to the CDU, closing the loop. Its jobs: even distribution preventing bias flow and GPU hot spots, collection back to the CDU for re-cooling, per-branch ball...