Nvidia is making waves in the global AI industry, with the much-anticipated H20 chip making a highly anticipated comeback in China. On July 14, the company made the announcement that the US government has approved the sale of its H20 chip to China.
It has been almost four months since the ban on sales of Nvidia’s H20 chip was imposed during the trade war. This unexpected move with the Nvidia H20 chip marks a dramatic reversal and signals a more pragmatic approach to US-China relations.
What Makes the H2O Chip Special?
Nvidia’s H20 is a GPU specially designed for the Chinese market due to export regulations. The H20 chip is not the most powerful Nvidia chip; however, it is more powerful than any chip China can produce at scale.
The H20 was built on Nvidia’s Hopper GPU architecture, which is highly optimized for AI and high-performance computing (HPC) workloads. This feature provides it with significant processing power for complex AI tasks.
While the chip lacks the full capacity of the flagship H100 or H200, it features a substantial amount of High Bandwidth Memory (HBM3), typically 96GB, with an NVLink of 400GB/s. This is crucial for large language models (LLMs) and other AI applications that require rapid access to massive datasets.
Why Was the Trade Restricted in the First Place?
Trade restrictions on exports to China were first implemented in October 2022. This was a move by the US government to restrict China’s access to advanced computing and semiconductor manufacturing items.
Over the years the restrictions were expanded and increased multiple times, notably in October 2023 and December 2025. The most recent was the restriction implemented in April 2025 during the US-China trade war, where all exports of the H20 chips were halted by the US government as a show of power.
In essence, the restrictions have always been a strategic move by the US government to control and curb the level of China’s technological advancement, particularly in areas deemed critical for national security and geopolitical influence.
The Impact of the Trade Restriction
Impact on Nvidia
With Nvidia being a top player in the AI chip market and China being its top buyer, the company suffered lots of losses during the restriction period.
In its fiscal first quarter (May 28, 2025), Nvidia reported a $4.6 billion loss for unsold inventory and was barred from shipping an additional $2.7 billion worth of AI products to China.
The ban led to disruptions in Nvidia manufacturing plans and led the company to void existing orders and cancel manufacturing capacity with TSMC, which then reallocated those production lines.
Impact on the US
The trade restrictions undeniably cost companies like Nvidia and AMD lost revenue and potential market shares. These losses caused a setback in the country’s position in the AI chip market.
The restriction also pushed China to seek alternative ways to develop domestic alternatives to the H20 chip, which potentially led to a decoupled global technology ecosystem. This would have led to the US losing all form of control to regulate China’s technological advancement and future market opportunities.
Impact on China
The ban on H₂O exports served as a wake-up call to China to accelerate its development of domestically made AI chips.
Chinese companies poured resources into developing their own AI chips, design tools, and manufacturing processes. This led to the development of Huawei’s Ascend Series, particularly the Ascend 910B and Ascend 920.
What Does the H2O Comeback Mean for China’s Future?
While China has already begun the production of its domestically made chip, like Huawei’s Ascend series, their production levels cannot yet meet up with the demand of the AI chips.
The comeback of Nvidia’s H20 will provide a much-needed relief for the pressure of meeting up with the high demand for their domestically made chips, giving them time to further develop the technology and maybe in a much sooner timeframe meet up with/topple the US position in the global tech race.
Until then, the return of the H20 is a monumental win for China and will serve as a means for more AI advancements due to its high computing power and compatibility with existing AI frameworks.
