NVIDIA's successor to Blackwell B30 in China is its next-gen Rubin R30 AI GPU in 2028

NVIDIA's successor to the tweaked B30 AI GPU for China is reportedly the Rubin R30 AI GPU according to investment banking and capital market firm Jefferies.

In a note picked up by insider @Jukanrosleve, Jefferies expects the annual NVIDIA chip procurement by China in 2027 to account for 1 million H20 as well as 70,000 more powerful H100 AI GPUs. In 2025, they're estimating 1.8 million H20 AI GPUs and 50,000 H100 AI GPUs, while in 2026 they're expected to shift to 1.2 million new B30 AI GPUs and 50,000 of the more powerful Blackwell B200 AI GPUs, while the orders drop slightly in 2027 with 1 million B30 and 50,000 B200 AI GPUs.

However, Jefferies says things will change in 2028 with the next-gen Rubin R30 AI GPU which we haven't heard about until now. The headlines have been filled with the Hopper H20 and H100 AI GPUs, with the new tweaked-for-China B30 preparing for release in Q4 with performance said to be 10-20% slower than H20, but B30 will cost 30-40% less.

In recent news, NVIDIA has reportedly moved up the production of its next-gen Rubin AI GPUs, while recent rumors have NVIDIA exploring the use of CoWoP (Chip-on-Wafer-on-Platform) PCB packaging for its next-gen Rubin R150 AI GPUs. Rubin R100 and its faster SKUs will be using next-gen HBM4 memory, while we could see the made-for-China variants like the purported R30 use GDDR7 memory instead.

NVIDIA's last-gen H20 AI GPU used HBM3, but the cheaper B30 moves to GDDR7 memory -- the same memory modules that go onto the consumer + gaming GPUs like the GeForce RTX 50 series "Blackwell" gaming GPUs. NVIDIA will most likely do something similar, reserving HBM4 memory for its high-end Rubin AI GPUs and use GDDR7 or even GDDR8 memory on its R30 for China in 2028.