NVIDIA H100 AI GPUs See 10% Price Drop In Chinese Black Market Ahead of H200 Launch

NVIDIA H100 AI GPUs have seen a drastic price decline in China's black market as the launch of next-gen H200 GPUs is imminent.

With the harsh US sanctions, there was no way individuals could access NVIDIA's cutting-edge AI products, and this ultimately gave rise to black markets in the region. This led to the NVIDIA H100 being accessible at astronomical prices in China, with reports disclosing that a single unit retailing somewhere around 3 million yuan, or around $415,000, is a vast increase over what the H100 is initially available for. However, it is reported that with the expected debut of the new H200 AI GPU, prices have seen a slight cooldown.

Taiwan Economic Daily reports that spot prices of NVIDIA's H100 AI GPUs have dropped by around 10% as "black market" sellers try to clear off the current inventory, anticipating the upcoming NVIDIA H200 AI accelerator. The H100 currently retails in spot markets for around 2.6 million yuan in Hong Kong which is around $359,000 US, marking an average 10% drop in the region. Well, this development has clearly shown us that illegal markets are still highly prevalent in China, & despite seeing restricted access to next-gen technology, it has still made its way into the country.

Another interesting fact with this report is that the debut of NVIDIA's H200 AI GPU means we will see the next-generation Hopper architecture drop into Chinese markets, not just the official way. Although in small amounts, Chinese AI firms won't see a complete blockade of AI chips. Instead, they can access specific dealers to acquire cutting-edge components, and we might witness NVIDIA's H200s being again popular in China, just not how the H100s were, but the sentiment would be similar.

While next-generation AI chips are attractive, we have witnessed a massive decline in interest in globally-renowned AI chips, like the ones from NVIDIA in China. Local firms involved in the AI race have switched to domestic alternatives since they are more accessible, have much lesser costs, and are ideal for low-to-mid inferencing tasks, making them a viable alternative. With government subsidies planned in, we might see colossal adoption of in-house AI solutions, taking the likes of NVIDIA and Intel out of the equation.