NVIDIA Plans to Launch New China-Specific GPU at Significantly Lower Price Than H20

According to reports from the Financial Associated Press on May 26, sources indicate that NVIDIA intends to introduce a new AI chipset tailored for the Chinese market, priced substantially below the H20 model recently restricted by U.S. authorities. Mass production could begin as early as June.

Two sources revealed that the new GPU will be part of NVIDIA’s latest-generation Blackwell architecture AI processors, with a projected price range of $6,500 to $8,000—significantly lower than the H20, which sells for $10,000 to $12,000.

However, the reduced pricing comes with scaled-back specifications and manufacturing requirements. Sources noted that the chip will be based on NVIDIA’s RTX Pro 6000D and utilize conventional GDDR7 memory instead of high-bandwidth memory. Additionally, it will not incorporate TSMC’s advanced CoWoS packaging technology.

A spokesperson from NVIDIA stated the company is currently assessing its “limited” options and added, “Until we finalize new product designs and receive U.S. government approval, we are effectively shut out of China’s $50 billion data center market.”

Sources also noted that another Blackwell-based chip tailored for China is under development and could enter production as early as September.

Although NVIDIA’s market share in China has declined in recent years, the region still represents a significant portion of its business, contributing 13% of sales in the most recent fiscal year.

At the Computex 2025 event in Taipei, CEO Jensen Huang remarked that NVIDIA’s market share in China has dropped from 95% prior to 2022 to approximately 50% today.

Amid tightening U.S. export controls on advanced technologies to China, this development marks NVIDIA’s third effort to create custom GPUs for the market. After the H20 was effectively banned in April, the company initially explored a downgraded version of the model, though the plan did not come to fruition.

Huang also stated last week that the older Hopper architecture cannot be further modified to comply with current U.S. restrictions.

According to investment bank Jefferies, the revised U.S. export rules cap memory bandwidth at 1.7–1.8 TB/s, whereas the H20 delivers 4 TB/s. The forthcoming GPU using GDDR7 is projected to achieve around 1.7 TB/s, staying within the regulatory threshold.

Recent industry speculation suggests the new GPU may be named 6000D or B40.