【By Observer Net Columnist Chen Feng】
The list of products banned from export to China by the United States is getting longer, and the list of Chinese companies sanctioned is also getting longer, especially in the Biden era. During the Trump era, the rate at which the list was extended slowed down, and there seemed to be some signs of partial reversal, but China no longer cares, because U.S. export controls have forced China to develop its independent and self-reliant technology.
During the peak of the Sino-U.S. tariff war, the Trump administration halted exports of EDA, H20, and LEAP1C to China. After subsequent Sino-U.S. economic and trade negotiations, EDA and LEAP1C were unblocked, and now H20 has also been unblocked.
EDA is key software for chip design and manufacturing, long controlled by American Synopsys, Cadence, and German Siemens (which has close ties with American technology, and actually complies with U.S. export controls) in the world market. These software are not only design tools, but also process preparation tools for chip manufacturing, and are widely used in China's chip industry.
LEAP1C is a high-bypass ratio advanced turbofan engine jointly developed by General Electric of the United States and Safran of France, used to drive the C919 aircraft of COMAC. This is an optimized variant of the LEAP1 series specifically for the C919, and it is currently the only engine certified for the C919.
H20 is NVIDIA's GPU, used for AI computing. This is a degraded version of the high-performance H100 specifically optimized for China under the export restrictions imposed by the Biden administration, a China-specific model, which significantly reduces training power but enhances inference power. In the explosive growth of China's AI industry, H20 has played an indispensable role.
After the United States first targeted ZTE, then tried to strangle Huawei with the full force of the state, China's determination and measures for technological independence and self-reliance reached a new height. This was actually initiated from aerospace.
In the 1980s, there was a brief cooperation between China and the United States in aerospace, and the J-8II "Peace Pearl" project is a familiar story. After the end of the Cold War, the U.S. and Europe imposed strict military embargoes on China, but China relied on its own strength, seizing the historic opportunity of Sino-Russian cooperation after the collapse of the Soviet Union, successfully developing the J-10 fighter, greatly shortening the gap with the world's forefront.
One of the representative projects of the brief aerospace cooperation between China and the United States in the 1980s - the J-8II "Peace Pearl" project
At the turn of the century, the U.S. aviation community, based on the development history of the J-8, speculated that China would need 20 years to perfect the J-10, which was completely misreading the times. In reality, 20 years after the first flight of the J-10, the first-generation J-10A and the second-generation J-10B had already been phased out, and the third-generation J-10C had been mass-produced. Today, if it weren't for the outstanding performance of the J-10C in the 57th air battle between India and Pakistan, the J-10C would already be considered "unwatchable" among Chinese military fans.
The J-20 completely changed the status of Chinese fighters in the world: it stared at the U.S. F-35A in the East China Sea, and entered and exited the Tsushima Strait and the Bashi Channel like it was unnoticed. Importantly, the production of the J-20 has exceeded 300 units, and the production of the fully equipped version with the WS-15 engine and advanced radar is accelerating. The twin-seat J-20S is a fifth-and-a-half generation, pioneering a new concept of command and control fighter. North and South sixth-generation aircraft quietly remind the world that China is the leader in sixth-generation aircraft.
In space exploration, the Obama-era Wolf Act completely cut off Sino-U.S. space cooperation, and the "Long March" series of rockets launched by China to provide launch services for communication satellites for the United States were forced to stop, and even normal academic exchanges were hindered. Today, when NASA's manned space capabilities are increasingly "unimpressive" and rely on SpaceX, which has caused chaos, China's space station is quietly shining in the cold orbit, and lunar missions are progressing steadily.
China is racing ahead, but not in a breath-holding sprint. Everyone knows that holding one's breath can unleash more power, but it won't last long. When the breath runs out, you have to stop, take a deep breath, and regain your "human form." China is sprinting while building up its physical strength in development. This is a high-speed marathon. The growing comprehensive national strength is the key. It is in this marathon that demand-driven and technology-driven forces work together, creating an unstoppable momentum of development.
From "military for economic construction" to investing heavily in defense to protect development, from the "two bombs and one satellite" to exploring new dimensions of the orbital economy, the development of China's military high-tech has always taken a high-speed but sustainable path. Now it's the turn of civilian high-tech.
Civilian high-tech differs from military high-tech. For a long time, there was the possibility of introducing and digesting technology. It is certain that the West will "hold back," and the possibility of being "strangled" cannot be ignored; it is also a fact that they cannot exceed the Western ceiling. However, the huge role of imported technology is obvious. This has led to the long-standing debate between "trade, industry, and technology" versus "technology, industry, and trade."
Lenovo's booth at the World Artificial Intelligence Conference 2025
The "trade, industry, and technology" vs. "technology, industry, and trade" debate within Lenovo, represented by Liu Chuanzhi and Ni Guangnan, can be expanded to the entire civilian high-tech field. The fact that this debate has persisted for a long time itself indicates that both routes have their merits. But the truth has its time limit; short-sightedness and overambition are both wrong.
When starting from a low base and having the condition to introduce, "trade, industry, and technology" makes sense, as it can quickly improve technological levels and rapidly create economic benefits, which is beneficial for future independent development. When the baseline is already high and introduction is blocked, "technology, industry, and trade" becomes necessary, which is the foundation for sustainable development, not only for competing with the West, but also for surpassing and changing direction.
"Trade, industry, and technology" or "technology, industry, and trade" should not be mechanically adhered to. It depends on when to speak and what to do. These big principles are generally accepted, but the problem is that sticking to "trade, industry, and technology" for too long can lead to path dependence, making it difficult to switch to "technology, industry, and trade" without strong external pressure, and eventually hitting a ceiling early. On the other hand, pushing "technology, industry, and trade" forcefully may result in long-term failure to generate economic benefits, and technical challenges may not be the biggest issue; overcoming the initial "death valley" is the real challenge, and trying to grow seedlings in barren land will ultimately result in dead seedlings and barren land.
China's reform and opening-up are inseparable. Reform means political and economic system reform, and opening-up not only means introducing foreign capital and opening up foreign trade, but also means technology introduction. The achievements of reform and opening-up are evident. As China's economy moves to second place and aims for the first, China's strength in basic scientific research, engineering technology, and market size has also moved to second place and aims for the first. If we use the Nature Index as a reference, China has already become the world leader. China's leading position in the commercialization of scientific and technological inventions is decisive.
All of this determines that the conditions for "technology, industry, and trade" have been formed, and the gradually tightening embargo and increasingly clear intentions of the United States have become the driving force to force a shift.
Huawei, despite extreme blockades and suppression, has shown extraordinary determination and strength. Huawei continues to lead in communications technology through 5.5G and is actively expanding in intelligent driving. HarmonyOS is not just a mobile operating system to replace Android and iOS, but a full-scenario distributed operating system for all connected things, supporting mobile phones, tablets, computers, smart wearables, and smart screens, as well as cross-device operation. It is a platform providing application development and device development services, with a microkernel structure that is lightweight, agile, secure, reliable, and particularly excellent in deployability and scalability.
Huawei's EDA has already gone live, otherwise the design and manufacturing of Huawei's new chips would be impossible. Huawei's new smartphones and GPU chips are still 7nm, but the bottleneck is not in design tools, but in manufacturing capabilities. China keeps this secret, but the world feels the undercurrents.
Huawei's ERP has also been successfully launched, which is an advanced core business software for enterprises that is fully self-controlled. Based on Huawei's Euler operating system and GaussDB, among other root technologies, it is developed in collaboration with many partners using cloud-native architecture, multi-tenant metadata architecture, real-time intelligence, and other advanced technologies. Unlike the PPT project, Huawei's ERP not only gets built, but also gets used and works well.
Huawei's Ascend 910 series GPUs are also emerging as a strong competitor. According to Tom's Hardware, the performance of Ascend 910C is equivalent to 60% of NVIDIA's H100. However, with the support of advanced system architecture and system integration technologies, the CloudMatrix 384 super node built based on Ascend 910C directly competes with NVIDIA's top-tier GB200 NVL72 system. This is comparable to DeepSeek defeating OpenAI's "computational magic" with more "intelligent" software and algorithms in the hardware version.
CloudMatrix 384 contains 384 Ascend 910C GPUs, while GB200 NVL72 contains 36 Grace CPUs and 72 B100 GPUs. This is not simply about stacking hardware. It's like pulling a cart with horses: one horse can pull the cart, two horses can pull harder, four horses even more. But when increasing to eight, sixteen horses, it's not guaranteed. Ensuring that the horses' power is aligned is very difficult. Increasing to thirty-two, sixty-four horses, the challenge of ensuring the horses don't fight each other is enormous.
In engineering design, the principle of simplicity is emphasized, known as the KISS principle - Keep It Simple, Stupid. The incredible complexity of CloudMatrix 384 is a result of being forced, but it also proves the tremendous potential of Chinese ingenuity under pressure.
In the second quarter of 2025, Huawei regained the top position in smartphone sales in mainland China with 12.2 million units shipped
NVIDIA is not unaware of the benefits of stacking hardware. Previous DGX H100 NVL256 had a similar optical module communication architecture to CloudMatrix 384, but was abandoned due to excessive costs, high power consumption, and serious reliability issues. Whether it's a coincidence or not, the J-20 integrates a diamond-shaped nose cone (the sharp side edges are equivalent to leading edges), full-moving canards, leading edges, and full-moving V-shaped tail fins, with unparalleled aerodynamic complexity. The F-22 does not even have a full-moving vertical tail, and its aerodynamics are much simpler.
Huawei is not fighting alone. Cambricon's MLU (Machine Learning Unit) cloud-based intelligent chips are essentially GPUs, and have achieved application results second only to the Ascend series. Cambricon's Sima 590 can support thousand-card-level training for mainstream open-source large models such as DeepSeek-R1 671B, Llama-3, and Qwen-3, with actual computing density reaching 80% of NVIDIA's A100. Cambricon's "hardware-software-ecosystem" system is also beginning to take shape, and it collaborates with domestic mainstream large model developers, aiming to form a domestically produced computing matrix covering training and inference by 2025-2026.
In terms of AI large models, after DeepSeek went viral, Alibaba, ByteDance, and Tencent have been catching up, and the Chinese team has already firmly occupied a leading position in the world. The distinction between "Chinese cleverness" and "American brute force" has become a hallmark in the AI world, with far-reaching implications for the future development of AI.
More importantly, multimodal models are on the horizon, which means there is still a lot of room for innovation in chip infrastructure, potentially breaking the dominance of NVIDIA in the language model era. This is a new starting line, indicating a new opportunity for overtaking. There are precedents in the aviation world. When transitioning from propeller-powered to jet-powered aircraft, the MiG-15 emerged unexpectedly, and European aviation powers never caught up again. The reason the MiG-15 used the Rolls-Royce "Nene" engine is not the reason, as British fighters had greater convenience in using the "Nene," but at the same time, British fighters could not even catch the shadow of the MiG-15.
Chips and AI are the main areas of U.S. export control, and civil aviation engines are another area. Even during the Trump 1.0 era, the U.S. had considered restricting the export of LEAP1C, but did not act. Now it has acted, but then withdrew, because China's civil aviation engine breakthrough is closer than chips and AI.
Through the development of the WS-10 and WS-15, China has crossed the threshold of advanced aero-engine research and manufacturing. Civil aviation engines and military aviation engines have diverged, but materials, manufacturing, design, and core engine technology are common. For example, the core engine of CFM56 comes from General Electric's F110. In China, the WS-20, which corresponds to CFM56, has already been produced and is used in the Y-20B.
The CJ1000 engine, which is paired with the C919, has been under development for many years and is planned to pass certification by the end of 2025. This is a counterpart to the LEAP1C. The relationship between the CJ1000 and the WS-20 is unclear, but the LEAP series can be seen as a deep improvement of the CFM56, adopting composite fan blades, ceramic matrix turbine blades, and larger bypass ratios, but the design philosophy and basic structure are highly similar.
COMAC is currently developing the C929, roughly comparable to the Boeing 787, and the CJ2000 is also under development at COMAC. This means that the C929's reliance on imported components during development and production is further reduced.
It is worth noting that as a latecomer, China has the advantage of starting anew, building a more advanced related technology ecosystem. For example, HarmonyOS has freed itself from the burden of Android's macrokernel and does not have the problems of iOS's closed system. The 3.9-meter cabin width of the C919 is wider than the 3.7-meter width of the A320, let alone the 3.54-meter width of the Boeing 737, not only increasing comfort but also avoiding the inconvenience of narrow corridors when extending the fuselage. The native digital fly-by-wire system also avoids the inherent defects like MCAS that could cause crashes.
While vigorously promoting domestic alternatives, China does not exclude suitable import options. As long as they are of good quality, affordable, and reliable, China will choose them. For example, many existing large models in China are still trained in the CUDA ecosystem, and using H20 to drive the inference engine is a reasonable choice. After the U.S. resumes the supply of H20, China's orders are enthusiastic, and NVIDIA has placed a new order for 300,000 chips with TSMC.
On July 31, the Cyberspace Administration of China held a meeting with NVIDIA regarding the security risks of backdoors in H20 computing chips
China is not afraid of H20 flooding the Chinese market again. China's market is large and diverse, and the U.S. "whip" reminds the Chinese government and businesses of the importance of de-risking critical supply chains. But if NVIDIA sets "tracking, positioning, and remote shutdown" functions on H20 as some U.S. congressmen have requested, it would certainly raise serious concerns in China.
China welcomes competition, and competition promotes growth. China is also not afraid of extortion.
When Trump loudly launched the tariff war, he faced a calm but precise counterattack from China. Trump wanted to further pressure with the export ban on EDA, H20, and LEAP1C, but immediately realized it was like picking up a stone to throw at his own feet. Not only might it further accelerate China's relevant technological independence, but American companies would also lose valuable markets forever. NVIDIA's market share in China's GPU market has dropped from 95% four years ago to 50% today.
There are elements of fabrication in the embargo, but if it could kill China's development in one move, Trump and the U.S. deep state would not hesitate to let American companies bear some losses. For years, China has repeatedly proven through independent technological achievements that extortion and embargoes are useless. But there are always people who don't believe it, how can they know unless they try?
U.S. export controls on China also involve investment. U.S. capital has been prohibited from investing in China's high-tech fields such as semiconductors, AI, quantum, biotechnology, and aerospace. This is also ineffective. China is no longer in a desperate need for capital. What is lacking is not capital, but opportunities, especially low-risk, high-return, short-payoff opportunities. Semiconductors, AI, quantum, biotechnology, and aerospace are not low-risk, high-return, short-payoff investments, but they are still high-quality investments. With U.S. capital withdrawing, the profit margin is left for Chinese capital. It's fine.
Now there is another issue: China strictly controls the export of rare earth refining and battery technology, can the U.S. (and Europe, Japan, South Korea, Australia) break through China's technological barriers? Rare earth is not rare, and mining is not the biggest challenge, but economic-scale refining is. The challenges of battery technology are similar; it's not about 0 to 1, but about 1 to 100, especially since China has already "dominated" 99%.
Before the Stockholm talks, Lutnick mentioned discussing the sale of TikTok, and if China does not approve, it will be banned. China did not respond, and after the talks, neither side mentioned TikTok. Whether TikTok is sold is ByteDance's decision, but AI recommendation technology is on China's export control list, and China strictly controls its export. TikTok does not operate in China, and if the U.S. can use substitute technology to replace it, there is no obstacle to selling TikTok, but can the U.S. do it?
What results can China's export controls on the U.S. bring about is an interesting question.
This article is exclusive to Observer Net. The content of this article is solely the opinion of the author and does not represent the views of the platform. Unauthorized reproduction will result in legal liability. Follow the Observer Net WeChat account guanchacn to read interesting articles every day.
Original: https://www.toutiao.com/article/7533509644153618990/
Statement: This article represents the views of its author. Please express your opinion by clicking on the 【top/beat】 button below.