It is reported that NVIDIA's new RTX6000D chip for the Chinese market has encountered a lukewarm response, with mainstream companies showing little interest

Zhitong
2025.09.16 10:21
portai
I'm PortAI, I can summarize articles.

NVIDIA's new RTX6000D artificial intelligence chip launched for the Chinese market is experiencing weak demand, with some mainstream technology companies choosing to hold off on orders. Sources indicate that the performance and pricing of the RTX6000D do not match, resulting in insufficient cost-effectiveness. Although JP Morgan and Morgan Stanley had previously predicted that the chip's production would reach 1.5 to 2 million units in the second half of the year, the market response has been contrary to expectations. NVIDIA has initiated the shipping of the RTX6000D, and a spokesperson stated that they will continue to provide high-quality products

According to the Zhitong Finance APP, two sources familiar with the procurement negotiations revealed that NVIDIA (NVDA.US) has developed a new artificial intelligence chip, the RTX6000D, specifically for the Chinese market. Currently, market demand is sluggish, and some mainstream technology companies have even chosen to hold off on placing orders.

These two sources stated that the RTX6000D is primarily used for artificial intelligence inference tasks, but considering its performance, the pricing is relatively high, and the cost-performance ratio is insufficient.

Additionally, sources earlier this month indicated that Chinese tech giants such as Alibaba, Tencent, and ByteDance are still waiting for a clear response regarding whether their orders for the NVIDIA H20 chip can be processed normally. Although the American company regained sales permission for the H20 chip to China in July, shipments have yet to resume.

Moreover, these companies are also hoping that the NVIDIA B30A chip, which significantly outperforms the H20, can receive approval from the U.S. government. The three chips (RTX6000D, H20, B30A) are all "downgraded" versions launched by NVIDIA in response to U.S. export restrictions—these measures were implemented by the U.S. government to curb China's technological development and maintain its leading position in the field of artificial intelligence, prompting NVIDIA to adjust the performance of overseas versions of the chips to comply with the restrictions.

The lukewarm reception of the RTX6000D sharply contrasts with the optimistic expectations of sell-side analysts. JP Morgan predicted in a report last month that the production of the RTX6000D could reach 1.5 million units in the second half of this year; Morgan Stanley estimated in July that the inventory of NVIDIA RTX6000D would reach 2 million units.

One of the sources revealed that NVIDIA has initiated the shipment of the RTX6000D this week. A spokesperson for NVIDIA stated in a press release: "Market competition is fierce—we are always committed to providing the best products within our capabilities."

From a technical parameter perspective, the RTX6000D is based on NVIDIA's latest Blackwell architecture, equipped with traditional GDDR memory, and has a memory bandwidth of 1398GB/s, slightly below the 1.4TB/s threshold set by the restrictions implemented by the U.S. in April.

The development of this chip is, to some extent, aimed at filling the market gap left by the H20 chip—sales of the H20 chip to China were banned in April this year, and although the ban has since been lifted, market supply has yet to recover.

The H20 chip is priced between $10,000 and $12,000 and uses the older Hopper architecture, but its memory bandwidth is more advantageous, reaching 4TB/s.

However, the H20 chip has yet to resume shipments, and there are multiple reasons behind this, including NVIDIA needing to address legacy issues related to a recent agreement—under this agreement, NVIDIA must remit a portion of its sales revenue from the Chinese market to the U.S. government.

As for whether the planned B30A chip will receive approval from the U.S. government, it remains uncertain. Reports this month indicated that the B30A is also based on the Blackwell architecture, with performance expected to reach six times that of the H20, while its price is only double that of the H20