NVIDIA responds to rumors that manufacturers cannot place orders for H20 chips: No comment on rumors
NVIDIA responded to rumors about the H20 chip, stating that they do not comment on rumors. Some manufacturers claim that they are unable to place orders for the H20 chip, but there are also reports indicating that a large number of H20 chips have arrived, with shipments exceeding expectations for the year. The H20 chip is seen as a smaller version of the H100, with high performance and low power consumption characteristics. Due to US export controls, NVIDIA's revenue from Chinese customers has significantly decreased
There are reports that some manufacturers are unable to place orders for NVIDIA's custom H20 chips in China. It has also been reported that some manufacturers claim to have received a large batch of H20 chips recently, exceeding the annual shipment expectation of about 400,000 units.
In response to these rumors, on the evening of September 20th, a Titanium Media App editor sought a response from NVIDIA, to which they replied, "No comments on rumors."
It is reported that in December last year, NVIDIA CEO Jensen Huang stated at a conference in Singapore that NVIDIA will continue to provide compliant chip products for the Chinese market. The "special supply" chips are expected to include HGX H20, L20 PCle, and L2 PCle, based on NVIDIA's Hopper and Ada Lovelace architectures.
Public information indicates that the H20 AI GPU is considered a smaller version of the H100 GPU, with a memory capacity of 96GB, a speed of up to 4.0 Tb/s, and computing power of 296TFLOPs. Compared to the H100, the H20 excels in performance density (TFLOPs/Die size) at 2.9, while the H100 is at 19.4. Additionally, the total power consumption of the H20 GPU is 400W, with an 8-way configuration, retaining a 900 GB/s NVLINK connection, and providing 7-way MIG functionality. According to related reports, the H20 has been in mass production since Q2 of this year, but detailed information has not yet been listed on the official NVIDIA website.
As early as October 17th last year, the U.S. Department of Commerce's Bureau of Industry and Security (BIS) issued new export control regulations for chips, imposing new export controls on semiconductor products including NVIDIA's high-performance AI chips; the restrictions took effect on October 23rd. NVIDIA's filing with the U.S. SEC shows that immediately banned products include A800, H800, and L40S AI chips.
It is worth noting that due to the upgraded export controls by the United States, NVIDIA's revenue from Chinese customers in the data center business has decreased from 19% in the 2023 fiscal year to the mid-single digits percentage in the 2024 fiscal year.
During the August earnings call, NVIDIA CFO Colette Kress mentioned that revenue from the data center business in China increased quarter-over-quarter, and the Chinese market remains an important contributor to NVIDIA's data center business. However, the proportion is lower than before the export controls, and NVIDIA faces intense competition in China.
Jensen Huang emphasized that NVIDIA will continue to "fully comply" with U.S. trade regulations. He pointed out that the chip industry supply chain is still globalized, and U.S. chip manufacturers are at least 10-20 years away from supply chain independence
