NVIDIA Denies H100, H200 GPU Shortage Amid Market Confusion

On September 2, the official social media account of NVIDIA released a statement addressing rumors about the shortage of H100 and H200 GPUs. The company asserted that these AI chips, based on the Hopper architecture, are in ample supply and can meet all current demands without delay. This clarification came in response to recent speculation about potential shortages in the market.

During the company's earnings call the previous week, the CEO had mentioned that both the H100 and H200 chips were sold out, and large cloud service providers were leasing capacity from other providers. The CEO also highlighted the high demand for these chips, particularly from AI-native startups looking to train and infer models. This statement seemed to contradict the recent clarification, leading to some confusion among observers.

Some commentators suggested that the current market environment could be misleading. While NVIDIA's efforts to downplay the GPU shortage might alleviate some concerns, it is important to remain vigilant and track the flow of orders. The company's actions could be aimed at protecting profit margins or setting a trap for liquidity issues. It is crucial for stakeholders to monitor the situation closely and assess the company's strategies in the context of the broader market dynamics.

In July, during a visit to China, the CEO revealed that the U.S. government had approved the sale of H20 chips to the Chinese market. The company also announced plans to launch a new AI chip specifically designed for the Chinese market, with an expected release date in September. However, there were subsequent reports indicating that the H20 chip, customized for the Chinese market, had limited stock and no plans for further production. This created uncertainty about the availability and future supply of these chips in the region.

In late July, the Cyberspace Administration of China held a meeting with NVIDIA to discuss security risks related to the H20 chip, including concerns about potential backdoors. Despite NVIDIA's denial of these allegations, the incident raised questions about the security and reliability of the company's products in the Chinese market. The situation highlighted the need for NVIDIA to provide convincing evidence of the chip's security to regain the trust of Chinese users and regulators.

In August, there were reports that NVIDIA was developing a new AI chip based on the Blackwell architecture, codenamed B30A, which would offer superior performance compared to the H20 model. The company aimed to begin testing this new chip in September. Additionally, there were indications that NVIDIA had informed some component suppliers to halt the production of the H20 chip, citing the need to manage the supply chain in response to market conditions. This move further complicated the supply situation and raised questions about the company's long-term strategy for the Chinese market.

In response to the restrictions on the H20 chip, the AI industry in China reportedly turned to second-hand NVIDIA A100 and H100 chips. This shift underscored the ongoing demand for high-performance AI chips and the challenges faced by the industry in securing reliable supplies. The situation highlighted the need for continued innovation and diversification in the AI chip market to meet the growing demands of the industry.