H20 Resumes Sales in China: Market Reception and Its Impact on the GPU Ecosystem

On July 16, NVIDIA attracted significant attention at its first appearance at the Chain Expo, with many attendees seeking servers equipped with H20 chips. Although the company showcased switches, laptops integrated with NVIDIA GPUs, and several humanoid robots, computing servers were notably absent from the display.

In contrast, Lenovo Group’s exhibition area featured a WA76160 G5 server fitted with 16 H20 chips, as reported by The Paper. According to booth staff, NVIDIA had visited Lenovo's booth to film a vlog that specifically highlighted this server model.

On July 15, in an interview with China Central Television, NVIDIA Founder and CEO Jensen Huang announced two major updates: U.S. authorities have approved the export of H20 chips to China, and the company plans to launch the RTX Pro GPU.

The H20 chip series was specifically developed by NVIDIA for the Chinese market to comply with U.S. export regulations. Pre-orders began in February 2024. However, in April, the U.S. government imposed an indefinite ban on H20 chip exports to China.

With sales now resuming after several months, attention has turned to market response. A senior industry figure told The Paper that he has been overseeing servers utilizing H20 chips and confirmed strong market demand. “There were many outstanding orders for the H20 that could not be delivered due to the ban. Given the chip’s solid performance, some clients are proceeding with their purchases.”

Some market watchers have speculated that NVIDIA may look to deplete existing H20 inventories before introducing a successor model. The same source noted that NVIDIA maintains a strong ecosystem and is convenient for training applications. If demand remains high, production could continue, though he emphasized that it is currently too early to predict with certainty given the recent restart of sales.

On July 15, NVIDIA officially announced the resumption of H20 chip sales in China and plans to launch a new GPU that will be fully compatible and tailored for the Chinese market. According to a company statement, Jensen Huang had recently visited both Washington, D.C., and Beijing to promote artificial intelligence, underscoring its potential benefits for global business and society.
At the same expo, Inspur also presented its servers, which incorporate GPUs from multiple Chinese manufacturers. Staff from Inspur confirmed ongoing collaboration with various domestic GPU suppliers.

A product manager responsible for servers utilizing domestic GPUs commented during an interview with The Paper that he is closely monitoring the impact of the H20’s return to market on local manufacturers. He suggested that while there may be some influence, the primary buyers of the H20 are likely to be existing users. Clients who had previously remained undecided or had not adopted H20 are unlikely to switch now.

He added that domestic GPUs have been performing strongly, especially during the period when H20 sales were suspended, creating growth opportunities for local players. “Most domestic GPU models are compatible with NVIDIA’s CUDA ecosystem, which facilitates the migration of CUDA-based applications into domestic ecosystems,” the product manager explained.

Addressing speculation that domestic GPU manufacturers experienced a surge in demand during the H20 ban, he clarified that production capacity constraints and limited supply were the main reasons for tight availability, rather than an actual surge in orders.

“Each domestic GPU brand has its own specialization and is selected based on client-specific needs. It wouldn’t be accurate to claim that any one brand is comprehensively superior. Each has carved out its own niche,” he said. Due to production limitations, domestic GPU makers are currently focusing on serving large clients, with limited ability to meet the needs of smaller businesses.
A senior professional at another server manufacturing firm told The Paper that if the H20 chip is priced competitively against domestic alternatives, it could see a strong resurgence in the market. “Over the past few months, domestic GPUs have certainly gained ground.”

During a media roundtable on July 16, Jensen Huang commented on the performance of H20 orders since the announcement of resumed sales, stating, “Since yesterday’s announcement, I have not yet had a chance to meet with any clients.” He added that ramping up H20’s supply chain capacity will require time and that NVIDIA aims to accelerate the process in the coming months.
Huang also expressed an intention to offer more advanced chips to the Chinese market. “The H20 still performs very well today,” he said. “In the coming years, we will offer whatever products we are permitted to sell in China.”

Market intelligence firm TrendForce has also provided forecasts regarding H20 demand. The agency believes that with the policy reversal allowing resumed exports, there will be a rebound in demand from Chinese AI and cloud service providers. TrendForce anticipates that H20 will once again become a key high-end AI chip in the market, driving increased demand for High Bandwidth Memory (HBM).

NVIDIA is not the only company resuming exports. According to Singapore’s Lianhe Zaobao, as cited by Reference News, U.S. chipmaker AMD also plans to restart exports of its MI308 chip to China following U.S. regulatory approval. An AMD spokesperson said on July 15 that the U.S. Department of Commerce had informed the company that its export license application for the MI308 is under review. The MI308 is an AI accelerator developed specifically for the Chinese market.

TrendForce projects that if current trends continue, NVIDIA’s push to meet its original shipping targets could raise the share of China’s AI market reliant on chips from NVIDIA, AMD, and other foreign suppliers to 49%, up from a previous estimate of 42% under the former export restrictions.