The lifting of the ban on NVIDIA's H20 chips is a "tactic in the open".
It has to be said that Jensen Huang's public relations campaign in China was quite successful. However, the future market performance of the H20 still remains to be tested by time.
From April 15, 2025, when NVIDIA's H20 chips were banned from being exported to China, to July 15, when NVIDIA was approved for an export license and attempted to aggressively "invade" the domestic market, just as Jensen Huang said, "We've always been expanding in China." However, the re - entry of the H20 chips may seem like a loosening of controls on the surface, but actually hides a deeper meaning.
A Technological Siege
I once elaborated on a similar phenomenon in the article "The Branding Path of Domestic Software": "From chips to software and terminals, the United States has mobilized its entire nation to conduct multi - dimensional blockades against Chinese technology giants..." Now it seems that from Microsoft to Intel, from software to hardware to the ecosystem, the United States has never given up its blockade plan. On May 14, 2025, the U.S. Department of Commerce issued an announcement, including Huawei's Ascend series of chips in the control scope of "General Prohibition 10", banning any global enterprise from using Huawei's Ascend chips. Otherwise, they will face criminal penalties such as up to 20 years in prison. One can clearly feel the intensity of the Sino - U.S. technology war.
U.S. Treasury Secretary Bessent even admitted that the lifting of the ban was due to "China having developed chips with comparable performance." Jensen Huang also praised Huawei in front of the media, saying that "Huawei has strong independent capabilities in system engineering and cloud services." This reveals a fact: the return of the H20 is not a technological favor but a microcosm of the Sino - U.S. high - end technology game. The cost - effectiveness of NVIDIA's H20 chips in large - model training and inference makes them a strong competitor to domestic similar chips. This "conspiracy" of "hoping that China will be more dependent on U.S. technology" is unfolding, and there are bound to be more stories in the future.
The next question is whether the technological and ecological routes of domestic chips can break through the siege?
An Independent Breakthrough
Currently, Huawei's computing cards have evolved from ASIC (Application - Specific Integrated Circuit) to GPGPU (General - Purpose Graphics Processing Unit), building a new platform different from the CUDA architecture. Through middleware compatible with CUDA, it realizes better application transfer to build an independent new ecosystem. This is similar to the HarmonyOS, which aims to "provide the world with an additional choice." Through compatibility design, it reduces the user migration cost and forms a key fulcrum for ecological breakthrough.
Specifically, the Ascend 910B chip uses Huawei's self - developed Da Vinci architecture in its design, which can be sized according to different scenario requirements and cover full - scenario applications. It has achieved independent research and development from IP to the instruction set. In terms of chip manufacturing process, although Huawei faces many restrictions, by using SMIC's 7nm (N + 2) process, it has achieved an energy - efficiency ratio significantly better than that of the H20, solving the problem of balancing computing power density and energy consumption. In terms of packaging, it cooperates with domestic enterprises. Through optimized design and manufacturing processes, it has achieved independent control of the domestic supply chain and improved chip performance.
Looking further ahead, with technological development, if the globally leading domestic optical quantum technology is applied as the next - generation computing infrastructure, its underlying innovation path is expected to reconstruct the technical architecture of chips.
Optimization of software algorithms is another key for Huawei to achieve high - performance chips. Ren Zhengfei once proposed the concept of "using mathematics to compensate for physics", adopting cutting - edge technical means such as sparse computing, model quantization, and pruning. The mode of collaborative optimization between software and hardware enables the Ascend chips to achieve efficient computing results even under relatively low - end manufacturing conditions. At the same time, Huawei's cluster computing mode allows its chips to rival top - level GPUs in overall computing power and support the training of its Pangu large model. In general, Huawei's Ascend 910B and NVIDIA's H20 chips are actually at the critical point of a performance game.
For some domestic private enterprises, the H20 may be an optimal choice after this lifting of the ban. From an industrial perspective, the H20 can effectively relieve the AI computing power pressure that some enterprises urgently need to solve, thereby promoting the training and growth of domestic large models and improving business efficiency. On the other hand, China's determination to achieve independence in the chip industry cannot be ignored. Although NVIDIA chips still have advantages in ecosystem and cost - effectiveness, they cannot stop the expansion of domestic chips' advantages in performance ratio and local adaptation.
"Short - term recovery but long - term pressure" will be the test that NVIDIA is about to face. Jensen Huang provided strong evidence for this: being replaced is "just a matter of time."
It can be predicted that in the future, the U.S. export policy for high - performance chips to China will only become stricter. Against this background, a clear route differentiation will gradually emerge in the chip field.
This article is from the WeChat official account "Jiong Jiong Time", author: Zhang Xiansheng. It is published by 36Kr with authorization.