AIbaseSource publication

Musk will order 20,000 NVIDIA racks next year, becoming the largest buyer of Vera Rubin

GF Securities analyst Jeff Pu expects SpaceX to become the largest buyer of NVIDIA Vera Rubin AI racks in 2027. The company originally intended to obtain 30,000 sets of NVL72 racks, but now it is expected to get only 20,000 sets.

GF Securities analyst Jeff Pu expects SpaceX to become the largest buyer of NVIDIA Vera Rubin AI racks in 2027. The company originally intended to obtain 30,000 sets of NVL72 racks, but now it is expected to get only 20,000 sets. However, based on next year’s market demand, it will still be the top single purchaser of the next generation system.

The market figure provided by Jeff Pu is as follows: Vera Rubin NVL72 racks will reach 73,000 units by the end of 2026, and further increase to 96,000 units by the end of 2027. SpaceX’s 20,000 units account for approximately 20.8% of the total next year. The six major new cloud service providers are expected to purchase a total of 34,000 units, accounting for about 35% of the year, indicating that the demand for high-performance infrastructure in AI remains growing.

Previously, Musk revealed that SpaceX has raised $40 billion through debt financing, which is specifically used for purchasing equipment such as NVIDIA GPUs. This money is also used to continuously expand data centers and provide computing power for its own AI business.

Foxconn Secures Nearly Half of the Assembly Orders

Unlike single-purchase GPUs, rack-level systems also involve high-speed interconnects, power supply, cooling, and overall system integration. Delivery depends not only on the chips but also on memory supply and assembly capabilities. At one end of the supply chain, Jeff Pu expects Foxconn to secure approximately 45% of the final assembly share in SpaceX’s data center orders. As the computing power race shifts from “building chips” to “assembling systems,” manufacturing and delivery capabilities are becoming the new decisive factors.


Original source

AIbase

Content notes

Original publication and rights belong to the source.

Machine translation · Refer to the original