The Government Circuit: News on Defense Electronics, Europe, and Sustainability
American Made Advocacy: Taking the Fight to Capitol Hill
Defense Speak Interpreted: SWaPing Nanosatellites for Defense Systems
NVIDIA Blackwell Platform and ASIC Chip Upgrades to Boost Liquid Cooling Penetration to Over 20% in 2025,
September 23, 2024 | TrendForceEstimated reading time: 2 minutes
TrendForce’s latest reports reveal that the launch of NVIDIA’s Blackwell platform, expected in 4Q24, is set to significantly drive the adoption of liquid cooling solutions. Liquid cooling penetration is projected to grow from around 10% in 2024 to over 20% in 2025. This shift will be driven by rising global ESG awareness and the accelerated deployment of AI servers by CSPs, prompting a shift from air cooling to liquid cooling systems.
NVIDIA continues to remain the dominant supplier in the global AI server market in 2024. Specifically, in the GPU AI server market, NVIDIA commands an overwhelming lead with a nearly 90% market share, while AMD follows at a distant 8%.
TrendForce notes that initial shipments of NVIDIA’s Blackwell platform are limited this year, as the supply chain continues to finalize product testing and validation testing, including high-speed transmission and cooling design optimizations. The high energy consumption of the new platform—especially for the GB200 full-rack solution—will require more efficient cooling, thereby driving the penetration of liquid cooling solutions. However, the current adoption of liquid cooling within existing server ecosystems remains low, and ODMs will need to navigate a learning curve to address challenges like leakage and insufficient cooling performance.
TrendForce forecasts that in 2025, Blackwell’s share in the high-end GPU market could exceed 80%, attracting power supply manufacturers and cooling solution providers to enter the AI liquid cooling market, thereby creating a new competitive landscape in the industry.
Taiwanese suppliers expected to provide quick disconnects by 1H24, as Google leads liquid cooling deployment
In recent years, major CSPs, including Google, AWS, and Microsoft, have accelerated the deployment of AI servers, primarily utilizing NVIDIA GPUs and custom-designed ASICs. TrendForce reveals that NVIDIA’s GB200 NVL72 racks, with a TDP of approximately 140 kW, will require liquid cooling solutions to address heat dissipation, with Liquid-to-Air (L2A) technology expected to become the mainstream approach. Other Blackwell server architectures, such as HGX and MGX, have lower density and will continue to rely on air cooling solutions.
Google is the most proactive in adopting liquid cooling solutions among CSPs developing their own AI ASICs, using both air and liquid cooling for its TPU chips. BOYD and Cooler Master are the primary suppliers of cold plates for Google. In China, Alibaba is aggressively expanding its liquid-cooled data centers, while other Chinese CSPs continue to use air cooling solutions for their AI ASICs.
TrendForce highlights that CSPs are specifying key suppliers for liquid cooling components in NVIDIA’s GB200 racks. Currently, Asia Vital Components and Cooler Master lead in providing cold plates, while Cooler Master and Auras supply manifolds, and Vertiv and Delta Electronics provide coolant distribution units. Quick disconnect (QD) components—critical for preventing leakage—are primarily supplied by international companies such as CPC, Parker Hannifin, Denfoss, and Staubli. However, Taiwanese companies like LOTES and Fositek are in the validation phase, and by the first half of 2025, they are expected to join the list of QD suppliers to help alleviate the current shortage.
Share on:
Suggested Items
Bransys Acquires Two REHM VisionXP+ Nitro 3850 Reflow Ovens
10/14/2024 | BransysBransys Group, a leading provider of comprehensive PCB design and assembly services, is pleased to announce the addition of two REHM VisionXP+ Nitro 3850 Type 834 reflow ovens to its state-of-the-art manufacturing facility.
Jabil Acquires Mikros Technologies
10/03/2024 | BUSINESS WIREJabil Inc., a global leader in design, manufacturing, and supply chain solutions, announced the successful acquisition of Mikros Technologies LLC, a leader in the engineering and manufacturing of liquid cooling solutions for thermal management, completed October 1.
NVIDIA Blackwell's High Power Consumption Drives Cooling Demands; Liquid Cooling Penetration Expected to Reach 10% by Late 2024
07/30/2024 | TrendForceWith the growing demand for high-speed computing, more effective cooling solutions for AI servers are gaining significant attention. TrendForce's latest report on AI servers reveals that NVIDIA is set to launch its next-generation Blackwell platform by the end of 2024.
Molex Releases Report on Thermal Management Challenges and Opportunities for I/O Modules
05/22/2024 | PRNewswireMolex, a global electronics leader and connectivity innovator, has published a report that examines thermal management pitfalls and possibilities as data center architects and operators strive to balance high-speed data throughput requirements with the impacts of growing power density and the need for heat dissipation on critical servers and interconnect systems.
The Chemical Connection: Reducing Etch System Water Usage, Part 1
04/05/2024 | Don Ball -- Column: The Chemical ConnectionWater conservation has become an important component of the overall system design for most manufacturing operations today. Changing climate conditions and increasing populations are beginning to strain the freshwater supplies in many areas of the country. As a result, as equipment suppliers, we see an increasing number of requests for options that reduce water usage in proposed wet processing systems. Etching systems tend to use more water because of the need for close temperature control to maintain steady etch rates and the rinsing requirements for complete removal of corrosive etchants from the surface of the product before the next process step. This column contains some simple suggestions for reducing water usage in etch systems that won’t strain the budget too much.