NVIDIA Reportedly Halts H20 Chip Production Amid Chinese Market Uncertainty #
According to The Information, NVIDIA has notified major suppliers—including Samsung Electronics, Amkor Technology, and Foxconn—to halt production of its H20 AI chip. This move follows pressure from the Chinese government urging domestic companies to temporarily suspend purchases of H20 chips, creating an uncertain market outlook.
Neither Samsung, Amkor, nor Foxconn has commented publicly on the matter.
Why NVIDIA Halted H20 Production #
The H20 AI chip, designed specifically for the Chinese market, has faced repeated challenges:
- In April 2025, U.S. export control rules initially blocked H20 shipments to China.
- Although the U.S. later allowed NVIDIA and AMD to sell lower-end AI chips, the approval came with a controversial condition: companies must hand over 15% of revenue from related sales to the U.S. government.
- Despite this, demand remained weak due to China’s growing focus on domestic chip development and information security regulations.
NVIDIA has already taken a $5.5 billion impairment charge on its H20 inventory, and the production halt only deepens concerns about how the company will absorb this cost.
Supply Chain Impact #
If true, the suspension could significantly affect NVIDIA’s partners:
- Amkor Technology reportedly faces a backlog of semi-finished H20 chips.
- Market analysts at Bernstein predict NVIDIA’s Chinese AI chip market share could drop from 66% in 2024 to 55% in 2025.
- The slowdown may accelerate China’s de-Americanization strategy, giving domestic chipmakers more room to grow, though short-term supply challenges are likely.
NVIDIA’s Next Move: B30A and RTX 6000D #
Despite setbacks, NVIDIA isn’t abandoning China. Reports suggest the company is developing two new AI chips based on its Blackwell architecture:
-
B30A (for AI training)
- Single-die design (vs. dual-die B300)
- Around half the raw computing power of the B300
- Equipped with HBM and NVLink for high-speed data transfer
-
RTX 6000D (for AI inference)
- Targets mid-to-low-end market
- Memory bandwidth capped at 1.398 TB/s (just below the U.S. 1.4 TB export threshold)
- Uses GDDR memory, making it more affordable than the H20
Launch Timeline #
According to insiders:
- RTX 6000D is progressing faster, with small-batch deliveries expected in September 2025.
- B30A could launch between late 2025 and early 2026, depending on U.S. government approval.
Final Thoughts #
The rumored suspension of NVIDIA’s H20 chip production reflects the growing geopolitical and regulatory challenges of the global AI chip industry. While short-term disruptions may hurt NVIDIA and its suppliers, the company’s pivot to B30A and RTX 6000D shows its determination to adapt and maintain a foothold in the Chinese AI market.
Stay tuned as the story develops and new chips approach launch.