Nvidia is preparing to launch a new, lower-cost graphics processing unit (GPU) tailored specifically for China, sources familiar with the matter told Reuters. The new chip will be part of Nvidia’s latest Blackwell architecture, but unlike its powerful H20 predecessor, which was effectively banned by US export regulations, it comes with more modest specifications and a price tag to match.
Expected to sell for between $6,500 and $8,000, the upcoming chip is significantly cheaper than the $10,000 to $12,000 H20 model, which was pulled from the Chinese market following new US export rules targeting high-performance AI hardware. The cost savings stem from the chip’s use of conventional GDDR7 memory rather than high-bandwidth memory and the absence of advanced packaging technologies from Taiwan Semiconductor Manufacturing Co. (TSMC), sources revealed.
While the official name of the chip hasn’t been confirmed, Chinese brokerage GF Securities speculates it may go to market as either the “6000D” or the “B40”.
No Green Light Yet, But Plans Are in Motion
Although mass production could begin as early as June, Nvidia is still treading carefully around US export rules. “Until we settle on a new product design and receive approval from the US government, we are effectively foreclosed from China’s $50 billion data center market,” an Nvidia spokesperson said.
The company, once dominant in China’s AI chip space, has seen its market share fall dramatically — from 95% before 2022 to about 50% today — due to ongoing US trade restrictions. CEO Jensen Huang acknowledged the slide while speaking to reporters in Taipei, adding that continued curbs may push more Chinese buyers toward homegrown options like Huawei’s Ascend 910B chips.
Nvidia’s pivot to the new Blackwell-based GPU reflects the company’s urgency to stay in the Chinese game, even if it means sacrificing performance. The new chip is expected to max out around 1.7 terabytes per second of memory bandwidth — just under the limit imposed by US controls, and far below the H20’s 4 terabytes per second.
The Long Game: CUDA Ecosystem May Be Nvidia’s Trump Card
Despite trimmed specs, Nvidia retains one powerful asset: its CUDA platform, which remains the backbone of AI model development for many engineers and developers. “Domestic Chinese technologies like Huawei are expected to catch up with the computing performance of downgraded versions within one to two years,” said Nori Chiou, investment director at Singapore-based White Oak Capital Partners. “Nvidia’s remaining edge lies primarily in its ability to integrate AI clusters with its CUDA platform.”
This software advantage could help Nvidia keep its remaining foothold in the Chinese market as it faces a new generation of domestic rivals. CUDA’s widespread adoption and developer ecosystem remain central to Nvidia’s strategy, particularly as its hardware faces increasing scrutiny from US regulators.
Meanwhile, Nvidia is reportedly working on a second China-specific Blackwell GPU variant, potentially ready for production by September. However, specifics remain under wraps. After scrapping earlier plans to modify its Hopper-based H20 to meet new export limits, the company had to write off $5.5 billion in inventory and forgo an estimated $15 billion in sales.
As the global tech cold war continues, Nvidia’s recalibrated approach signals a delicate balancing act — one that hinges on strategic compromises, regulatory approvals, and the enduring pull of its software ecosystem.
