TrendForce’s newest investigations discover that the general server market has not too long ago stabilized, with ODMs concentrating efforts on AI server growth. Beginning in Q2, shipments of recent Blackwell-based platforms—such because the NVIDIA GB200 rack and HGX B200—have progressively ramped up, whereas next-generation B300 and GB300 sequence merchandise have entered the sampling and validation section.

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

Oracle’s growth of AI information facilities in North America has generated important order development for Foxconn, whereas additionally benefiting suppliers like Supermicro and Quanta. Supermicro’s main development driver this yr is its AI server enterprise, having not too long ago secured a number of GB200 rack initiatives. 

Quanta, leveraging its longstanding partnerships with main purchasers comparable to Meta, AWS, and Google, has efficiently expanded its GB200/GB300 rack operations and not too long ago gained further orders from Oracle, boosting its AI server efficiency notably.

In the meantime, Wiwynn is deepening its collaborations with Meta and Microsoft and is anticipated to see development within the second half of the yr. Wiwynn’s AI server technique is centered on ASIC-based fashions and is presently a key provider for AWS’s Trainium programs.

Enlargement of AI information facilities to catalyze scale-up for liquid cooling provide chain
TrendForce notes that as shipments of GB200/GB300 racks broaden via 2025, the adoption fee of liquid cooling options for high-end AI chips is steadily rising. In comparison with conventional air cooling programs, liquid cooling requires extra complicated infrastructure, together with cabinet- and facility-level coolant piping, cooling towers, and CPUs. In consequence, new information facilities are more and more designed from the outset to be “liquid cooling prepared” to reinforce thermal effectivity and scalability.

Liquid cooling is quickly changing into the usual configuration for high-performance AI information facilities, considerably boosting demand for thermal elements and accelerating provider shipments. For instance, Fositek has begun delivery NVIDIA QD couplings particularly designed for the GB300 platform, used along with chilly plates developed by its mother or father firm, AVC, for the GB300 NVL72 rack system.

Provide chain sources point out that Fositek has additionally begun mass manufacturing of QDs and floating mount QDs of AWS’ ASIC-based liquid cooling programs, and its share of QD provide for this platform is anticipated to rival that of Danfoss.

Auras has likewise been increasing into the information middle liquid cooling house, with this section more and more changing into a core driver of firm development. Its main purchasers embrace main server manufacturers comparable to Oracle, Supermicro, and HPE.

Auras’ product choices span chilly plates and manifold modules. The corporate has additionally began delivery liquid cooling merchandise to Meta, laying the groundwork to affix the GB200 liquid cooling provide chain and positioning itself for potential entry into Meta and AWS’ second wave of chilly plate sourcing.

Supply: TrendForce, Taiwan.