Two sources familiar with the procurement negotiations revealed that demand for NVIDIA's (NVDA.US) new artificial intelligence chip, the RTX 6000D, specifically designed for the Chinese market, is currently weak, with some major tech companies even choosing to hold off on orders for now.

The two sources indicated that the RTX 6000D is primarily intended for AI inference tasks, but considering its overall performance, its pricing is relatively high, resulting in insufficient cost-effectiveness.

Another source mentioned earlier this month that Chinese tech giants such as Alibaba, Tencent, and ByteDance are still awaiting clear confirmation on whether orders for NVIDIA's H20 chips can be processed normally. Although the U.S. company regained approval in July to sell the H20 chips to China, it has not yet resumed shipments.

Additionally, these companies are pinning their hopes on NVIDIA's B30A chip, which far outperforms the H20, receiving approval from the U.S. government. All three chips—the RTX 6000D, H20, and B30A—are "downgraded" products developed by NVIDIA in response to U.S. export restrictions. The U.S. government implemented these export control measures to curb China's technological advancement and maintain its own leading position in the field of artificial intelligence, prompting NVIDIA to adjust the performance of its overseas chips to comply with the requirements.

The lukewarm reception of the RTX 6000D contrasts sharply with the optimistic expectations of sell-side analysts earlier. Last month, J.P. Morgan predicted in a report that the production volume of the RTX 6000D could reach 1.5 million units in the second half of this year, while Morgan Stanley estimated in July that NVIDIA's inventory preparation for the RTX 6000D would amount to 2 million units.

One of the sources disclosed that NVIDIA began shipping the RTX 6000D this week. An NVIDIA spokesperson stated in a declaration: "The market competition is fierce—we are always committed to providing the best products within our capabilities."

From a technical parameter perspective, the RTX 6000D is based on NVIDIA's latest Blackwell architecture, equipped with traditional GDDR memory, offering a memory bandwidth of 1398 GB/s, slightly below the 1.4 TB/s threshold set by the U.S. restrictions announced in April.

The development of this chip was, to some extent, intended to fill the market gap left by the H20 chip. In April, the H20 chip was banned from being sold to China, and although the ban was later lifted, market supply has never resumed.

The H20 chip is priced between $10,000 and $12,000, utilizes the older Hopper architecture, but boasts a superior memory bandwidth of up to 4 TB/s.

However, the failure to resume shipments of the H20 chip so far can be attributed to multiple reasons, including NVIDIA's need to address legacy issues related to a recent agreement. According to this agreement, NVIDIA is required to 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, also based on the Blackwell architecture, is expected to deliver six times the performance of the H20, while being priced at only twice that of the H20.