
Jibang Consulting: It is estimated that Blackwell will account for over 80% of NVIDIA's high-end GPU shipments in 2025

According to TrendForce's estimate, by 2025, Blackwell GPUs will account for over 80% of NVIDIA's high-end GPU shipments. With the stabilization of the AI server market, ODM manufacturers such as Oracle, Supermicro, and Quanta are expanding AI data centers, driving the shipment of related products. The application of liquid cooling solutions in high-end AI chips is also increasing and is expected to become a key development in the future
According to the latest survey by TrendForce, the overall Server market has recently stabilized, with ODMs focusing on the development of AI Servers. Starting from the second quarter, they have gradually ramped up production for NVIDIA (NVDA.US) GB200 Rack, HGX B200, and other Blackwell new platform products, while the updated B300 and GB300 series have entered the sample verification stage. Therefore, TrendForce estimates that this year, Blackwell GPUs will account for over 80% of NVIDIA's high-end GPU shipments.
Observing the recent dynamics of Server ODMs, North American CSP giant Oracle (Oracle) is expanding its AI data centers, which not only brings order growth for Foxconn (Foxconn) but also benefits players like Supermicro (Supermicro) and Quanta (Quanta). TrendForce expects Supermicro's main growth driver this year to come from AI Servers, and it has recently secured some GB200 Rack projects. Quanta has successfully expanded its GB200/GB300 Rack business, benefiting from collaborations with large clients like Meta, AWS, and Google, and has recently performed impressively in the AI Server field by securing Oracle orders. Wiwynn (Wiwynn Technology) continues to deepen its cooperation with Meta and Microsoft, which is expected to drive performance growth in the second half of this year. It focuses on ASIC AI Servers and is currently a major supplier for AWS Trainium AI models.
The expansion of AI data centers will be key to the scaling of the liquid cooling industry chain
TrendForce states that as the shipments of GB200/GB300 Racks expand in 2025, the adoption rate of liquid cooling solutions for high-end AI chips is continuously increasing. Compared to traditional air cooling systems, liquid cooling architectures involve more complex supporting facility constructions, such as the layout of cooling liquid pipelines at the cabinet and data center levels, cooling water towers, and fluid distribution units (CDUs). Therefore, many newly built data centers are incorporating the concept of "liquid cooling ready" in the early design stages to enhance overall thermal management efficiency and scalability.
Liquid cooling will not only become a standard configuration for high-performance AI data centers but will also significantly drive up the demand for cooling components, accelerating the shipping pace of suppliers. For instance, Fositek has officially shipped NV QD (quick connectors) specifically for the GB300 platform, designed to pair with the cold plates developed by its parent company AVC, for the GB300 NVL72 Rack system. The supply chain reveals that Fositek has mass-produced and shipped the quick connectors and floating quick connectors needed for AWS ASIC liquid cooling, and it is estimated that its supply ratio of quick connectors for this platform can compete with Danfoss Auras (Shuanghong) has also been actively laying out the data center liquid cooling market in recent years, with related businesses gradually becoming the core driving force for the company's growth. Its main customers include Oracle, Supermicro, and HPE (Huiyu), among other mainstream server brands. The product line covers cold plates and manifold modules. Auras has also begun shipping liquid cooling products to Meta, laying the foundation for entering the supply chain of the GB200 platform liquid cooling system, and is expected to join the second wave of core supply systems for cold plates from Meta and AWS in the future