Nvidia Unveils Budget Blackwell AI Chip for China Amid U.S. Export Curbs

• Reuters reports that Nvidia (NVDA) plans to launch a new Blackwell-architecture AI chipset for China, priced between $6,500 and $8,000, with mass production starting as early as June, designed to comply with U.S. export restrictions capping GPU memory bandwidth at 1.7-1.8 terabytes per second.
• The chip, based on the RTX Pro 6000D and using GDDR7 memory, is a response to the April ban on Nvidia’s H20 model, which led to a $5.5 billion inventory write-off and $15 billion in lost sales, as the company’s market share in China dropped from 95% to 50%.
• Facing competition from Huawei’s Ascend 910B and ongoing U.S. restrictions, Nvidia is also developing another Blackwell-based chip for China, with production targeted for September, to maintain its 13% sales foothold in the $50 billion Chinese data center market.

Nvidia (NVDA) is navigating a complex landscape in China, a critical market that accounted for 13% of its sales in the past financial year, by developing a new artificial intelligence chipset tailored to comply with stringent U.S. export restrictions. This upcoming graphics processing unit (GPU), based on Nvidia’s latest Blackwell architecture, is designed, as reported by Reuters, to meet the demands of China’s $50 billion data center market while adhering to new regulatory constraints. Sources familiar with the matter told the publication that mass production of this chip is slated to begin as early as June, with pricing set between $6,500 and $8,000. This represents a significant discount compared to Nvidia’s recently restricted H20 model, which sold for $10,000 to $12,000, reflecting the new chip’s simpler manufacturing requirements and reduced specifications.

The new GPU, potentially named the 6000D or B40 according to a note from Chinese brokerage GF Securities, is derived from Nvidia’s RTX Pro 6000D, a server-class graphics processor. Unlike higher-end models, it will utilize conventional GDDR7 memory instead of advanced high-bandwidth memory and will forgo Taiwan Semiconductor Manufacturing Co’s (TSM) sophisticated Chip-on-Wafer-on-Substrate (CoWoS) packaging technology. These design choices align the chip’s performance with U.S. export limits, particularly the new cap on GPU memory bandwidth set at 1.7 to 1.8 terabytes per second, a critical metric for AI workloads. GF Securities estimates the chip will achieve approximately 1.7 terabytes per second, a sharp reduction from the H20’s 4 terabytes per second, ensuring compliance but limiting its computational prowess compared to unrestricted models.

Nvidia’s strategic pivot comes in response to escalating U.S. restrictions aimed at curbing Chinese technological advancements. The April ban on the H20, which utilized Nvidia’s older Hopper architecture, forced the company to abandon plans for a downgraded H20 variant, as CEO Jensen Huang noted that the architecture could no longer be modified to meet export controls. The ban resulted in a $5.5 billion inventory write-off and $15 billion in lost sales, underscoring the financial stakes of operating in China. Huang highlighted the company’s challenges on the Stratechery podcast, emphasizing the shrinking window of opportunity in a market where Nvidia’s share has dropped from 95% before 2022 to 50% today, largely due to competition from Huawei’s Ascend 910B chip. Huang warned that continued export curbs could further drive Chinese customers toward domestic alternatives like Huawei.

Despite these challenges, the report notes that Nvidia is doubling down on China by developing another Blackwell-based chip, with production targeted for September, though its specifications remain unclear. An Nvidia spokesperson indicated that the company is still evaluating its “limited” options pending U.S. government approval, highlighting the uncertainty surrounding its ability to compete in China’s data center market. This marks Nvidia’s third attempt to tailor a GPU for China, reflecting the delicate balance between innovation, market access, and geopolitical constraints. As Nvidia adapts its cutting-edge Blackwell architecture to meet regulatory demands, its ability to maintain a foothold in China will depend on delivering cost-effective, compliant solutions in a market increasingly dominated by local competitors.

WallStreetPit does not provide investment advice. All rights reserved.

• Bulenox: Get 83% OFF ... Use Discount Code: JJT3A
• Risk Our Money Not Yours | Get 50% to 90% OFF ... Use Discount Code: MMBVBKSM