The US has indeed embarked on an adversarial course against the wishes and ambitions of China for semiconductor activities. Those are the words of the US Secretary of Commerce Howard Lutnick, and they leave no room for doubt: the US export laws greatly interfere with China's own plans of producing advanced chips.
The chip battle has been going on for a long time since ever since it gained traction during the Trump era, and at this moment the chip war can be truly said to be on. It is not a purely business affair but one of national security. There have been export bans, fresh tariffs—all aimed at making sure the US stays on top. Secretary Lutnick has further described the results of these measures, warning that domestic production of high-end chips in China—essential for AI and modern smartphones—is a little fraction of what its own market requires. He estimates that China may be able to produce only about 200,000 of such advanced chips, a drop in the ocean compared to what it needs.
The US is going after the ecosystem, not just the production itself. The new laws around "AI Diffusion" puts companies making use of Huawei's Ascend AI chips in jeopardy with US authorities. Restrictions on EDA (Electronic Design Automation) software, vital for chip design, along with a major crackdown on the illicit sale of AI equipment into China, apply further tightening. The message is clear: The US wants to secure this technological front, and the current administration will appear to do anything to bring this about.
In its all-out effort to cripple China, the US simultaneously initiated the return of chip-making back home. TSMC, the prime example, is the largest chip maker in the world, yet it's an injection of funds into the creation of new US-based facilities and R&D, turning many heads. Some reports indicate those production lines are already booked; this, being part of worldwide chip production focus shifting more and more towards the US.
Huawei Has Trouble Competing in the AI Chip Sector
Huawei pinned its hopes to its Ascend 910C GPUs being able to combat NVIDIA's AI monopoly in China. All attempts at breaking NVIDIA's hold under largely the protection of its CUDA software ecosystem have proved to be a near insurmountable obstacle. More difficulties are coming Huawei's way from its own internal problems.
Reports suggest that prominent tech companies in China, including ByteDance (the parent company of TikTok), Alibaba, and Tencent, are hesitant to place large orders for Huawei's AI chips. Various reasons account for this reluctance:
- Ecosystem Lock-in: Most of the Chinese tech giants have made heavy investments in NVIDIA's CUDA platform. Transitioning to Huawei's contender, CANN (Compute Architecture for Neural Networks), considered the less mature product, would need considerable time and resources. These firms allegedly want Huawei to fit into their existing platforms, not the other way around.
- Competitor Concerns: Due to Huawei being a competitor with many of these tech firms, there is an inherent reluctance to become too reliant on the offerings of their rival.
- Reliability: Supposedly, the Ascend 910C chips have had some overheating issues from time to time, which have raised questions regarding their reliability among China's tech crowd.
- Existing NVIDIA Stock: Most large Chinese tech companies have stockpiled plenty of NVIDIA GPUs over the years. Until their stocks run down, the urgency to switch Huawei sources will not be higher.
US Export Controls are Still Lingering: The May announcement from the Department of Commerce all but branded Huawei's chips as "toxic." Any unauthorized user of the chips is feared to be infringing on US export controls, which creates a major concern for Chinese companies with international operations.
Despite these headwinds, Huawei's Ascend 910C is a competent chip and is said to pack quite a punch equal to NVIDIA's H100 GPU in terms of raw computing power. With the CloudMatrix 384, Huawei even unveiled a system bundling up to 384 Ascend chips in a direct challenge to NVIDIA supercomputer offerings. However, it lacks native support for essential memory-efficient calculation formats such as FP8, relying instead on a rather poor translation tool.
While NVIDIA is apparently prospering, finding its way through the absence of any very strong winds from China, for one, it has recently described its visibility on "tens of gigawatts" worth of AI infrastructure projects. Analysts are speculating that should even a fraction of this pipeline prove to be true, the revenue could well be in the hundreds of billions of dollars-a-year for NVIDIA's AI-centric data center segment alone. The demand for NVIDIA's AI technology appears to be all-but insatiable, creating a firm foundation for the company atop the market.
The worldwide chip landscape remains a dynamic and fierce battleground. While the US actively works to maintain its technological lead, China strives for self-reliance, and tech industries all over the world feel the ripple effect.
