SpaceX Turns Away from Dell? Reports Suggest AI Server Procurement Shift to Direct Purchases from Taiwan, with Four Major Manufacturers Vying for VR200 Orders

Reports indicate that SpaceX and CoreWeave are adjusting their AI server procurement approach, moving away from purchasing through Dell and other branded vendors and instead opting for direct procurement from Original Design Manufacturers (ODMs) in Taiwan. According to a recent supply chain investigation by U.S. investment bank Jefferies, four major Taiwanese manufacturers are currently on the shortlist for negotiations, with the possibility of securing new orders as early as 2027. The initial deployment could involve NVIDIA’s next-generation VR200 rack based on the Vera Rubin architecture. If this partnership materializes, not only could these four Taiwanese manufacturers capture a significant wave of AI server business, but Dell’s previously dominant order share may also be redistributed, reshaping the landscape of the next-generation AI server supply chain.

SpaceX and CoreWeave Eye Direct Procurement – Four Taiwanese ODMs in the Spotlight

Jefferies’ latest channel survey reveals that SpaceX and CoreWeave are evaluating a shift in their AI server procurement path, planning to reduce reliance on branded vendors like Dell and instead collaborate directly with ODMs. The Taiwanese manufacturers named in market speculation include Foxconn, Quanta, Wiwynn, and Wistron. This model—where large-scale cloud or AI compute providers procure directly from ODMs—is not unprecedented in the industry. Hyperscalers such as Google, Meta, and Microsoft have long adopted this approach. What makes this development noteworthy is that both SpaceX and CoreWeave are themselves major players in AI infrastructure demand. If they increase their direct ODM procurement ratio, it will significantly boost opportunities for Taiwanese server manufacturers to engage directly with large-scale compute clients.

VR200 Emerges as Potential Deployment Model – Unit Value Soars

Jefferies notes that if the new procurement model takes shape, deployment could begin as early as 2027, starting with the NVIDIA Vera Rubin architecture-powered VR200 rack. Compared to the current Blackwell generation, the Vera Rubin platform offers enhanced AI rack specifications and a substantially higher unit value. Jefferies estimates that the average selling price (ASP) of VR-series racks could reach approximately $8–9 million, far exceeding the current GB200 NVL72’s ~$3 million. This indicates that the AI server market is being driven not only by shipment volume growth but also by rising per-rack value due to next-gen platform upgrades. However, higher server ASPs do not automatically translate into proportionally higher ODM gross margins. With rising costs for high-end GPUs, liquid cooling, power supplies, and other components, actual profitability will depend on product mix and cost management.

Foxconn’s GB300 Shipments Accelerate – AI Server Performance in Second Half of 2026 Under Scrutiny

Beyond the next-generation Vera Rubin platform, Jefferies also expresses optimism about Foxconn’s current momentum in Blackwell server shipments. The firm estimates that Foxconn could ship between 15,000 and 20,000 GB300 racks in the second half of 2026, with continued growth expected into 2027. Based on the estimated pricing of GB300 NVL72 racks, this shipment scale would provide meaningful support to Foxconn’s AI server business. As a result, while the market closely watches the potential VR200 orders in 2027, it is also tracking Foxconn’s AI server delivery progress from the second half of this year onward.

Quanta, Wiwynn, and Wistron Also in Focus – Order Allocation to Be Key

In addition to Foxconn, Quanta, Wiwynn, and Wistron are also highlighted by Jefferies as Taiwanese ODMs under consideration. Wiwynn has long specialized in large data centers and cloud servers, Quanta is one of the world’s leading server manufacturers, and Wistron has steadily expanded its AI server-related initiatives in recent years. Since discussions are still ongoing, the final order allocation among the four manufacturers remains uncertain. For the market, the real point of interest isn’t simply that “all four benefit,” but rather which suppliers SpaceX and CoreWeave ultimately select and the distribution of order shares among the ODMs.

Dell Faces Pressure on Order Share – But AI Demand Remains Strong

For Dell, the gradual shift by major AI customers toward higher ODM direct procurement ratios could indeed pose some risk of server order loss. However, the core of this supply chain change is not a decline in AI server demand, but rather a transformation in procurement models. Even as customers begin increasing direct ODM purchases, it doesn’t mean they will immediately cease all orders with Dell. Large AI clients typically employ multi-supplier strategies to mitigate risks related to capacity, delivery timelines, and supply chain disruptions. Therefore, the focus going forward will be on observing the actual procurement ratios adopted by SpaceX and CoreWeave.

AI Server Supply Chain Reshuffling – Can Taiwanese ODMs Secure Direct Orders?

As AI data centers continue to expand, competition within the server supply chain has evolved beyond mere manufacturing capability to encompass customer relationships, capacity allocation, and end-to-end system delivery performance. The reported evaluation of ODM direct procurement by SpaceX and CoreWeave, if formally implemented, would further elevate the strategic importance of Taiwanese ODMs in the AI server ecosystem. While the four Taiwanese manufacturers currently in the spotlight—Foxconn, Quanta, Wiwynn, and Wistron—are well-positioned, the extent of their actual benefits will depend on the final supplier selection, VR200 mass production timeline, and order distribution outcomes.