Blackwell to account for over 80% of NVIDIA’s high-end GPU shipments in 2025: Liquid cooling adoption continues to rise

TrendForce’s latest investigations find that the overall server market has recently stabilized, with ODMs concentrating efforts on AI server development. Starting in Q2, shipments of new Blackwell-based platforms—such as the NVIDIA GB200 rack and HGX B200—have gradually ramped up, while next-generation B300 and GB300 series products have entered the sampling and validation phase.

TrendForce projects that Blackwell GPUs will account for over 80% of NVIDIA’s high-end GPU shipments in 2025.

Oracle’s expansion of AI data centers in North America has generated significant order growth for Foxconn, while also benefiting suppliers like Supermicro and Quanta. Supermicro’s primary growth driver this year is its AI server business, having recently secured several GB200 rack projects. 

Quanta, leveraging its longstanding partnerships with major clients such as Meta, AWS, and Google, has successfully expanded its GB200/GB300 rack operations and recently won additional orders from Oracle, boosting its AI server performance notably.

Meanwhile, Wiwynn is deepening its collaborations with Meta and Microsoft and is expected to see growth in the second half of the year. Wiwynn’s AI server strategy is centered on ASIC-based models and is currently a key supplier for AWS’s Trainium systems.

Expansion of AI data centers to catalyze scale-up for liquid cooling supply chain
TrendForce notes that as shipments of GB200/GB300 racks expand through 2025, the adoption rate of liquid cooling solutions for high-end AI chips is steadily increasing. Compared to traditional air cooling systems, liquid cooling requires more complex infrastructure, including cabinet- and facility-level coolant piping, cooling towers, and CPUs. As a result, new data centers are increasingly designed from the outset to be “liquid cooling ready” to enhance thermal efficiency and scalability.

Liquid cooling is rapidly becoming the standard configuration for high-performance AI data centers, significantly boosting demand for thermal components and accelerating supplier shipments. For example, Fositek has begun shipping NVIDIA QD couplings specifically designed for the GB300 platform, used in conjunction with cold plates developed by its parent company, AVC, for the GB300 NVL72 rack system.

Supply chain sources indicate that Fositek has also begun mass production of QDs and floating mount QDs of AWS’ ASIC-based liquid cooling systems, and its share of QD supply for this platform is expected to rival that of Danfoss.

Auras has likewise been expanding into the data center liquid cooling space, with this segment increasingly becoming a core driver of company growth. Its major clients include leading server brands such as Oracle, Supermicro, and HPE.

Auras’ product offerings span cold plates and manifold modules. The company has also started shipping liquid cooling products to Meta, laying the groundwork to join the GB200 liquid cooling supply chain and positioning itself for potential entry into Meta and AWS’ second wave of cold plate sourcing.

Source: TrendForce, Taiwan.