NVIDIA CEO Jensen Huang warned the United States' House Foreign Affairs Committee that the country faces being left behind by a rapidly-moving surge in AI development, calling on the country to maximize and accelerate the adoption of American AI worldwide.
"We are at an inflection point: The United States needs to decide if it is going to continue to lead the global development and deployment of AI or if we are going to retreat and retrench," Huang told the committee on Thursday, May 1.
"America cannot lead by slowing down. If we step back, others will step in. And the global AI ecosystem will fragment — technologically, economically, and ideologically."
When it comes to long-term prosperity in the high-tech world, it's all about setting standards. Intel once set the standard with x86, PCIe, and USB and now the vast majority of devices use these technologies in one way or another. Nvidia now enjoys its investments in the CUDA ecosystem and is setting the standard in AI compute in general.
To a large degree, Nvidia's efforts made the U.S. industry the leader in AI. But containing AI hardware in the U.S. will provoke rapid development of competing AI ecosystems that can eventually outperform the one developed in America, the company now warns.
Nvidia is everywhere - for now
For now, Nvidia and its CUDA ecosystem set the standards for AI applications across the world, both for training and for inference. With products like GB200/GB300 NVL72 Nvidia can address companies that need on-premise AI deployments for their AI applications. Given the ubiquity of CUDA, such deployments are easy and relatively inexpensive.
Nonetheless, Nvidia has domestic rivals, including traditional competitors like AMD and Intel as well as newcomers like D-Matrix and Tenstorrent. Most of their efforts are aimed at inference though, as Nvidia is the de facto king of AI training thanks to CUDA.
Thanks to the omnipresence of CUDA, Nvidia leads in AI not only in the U.S., but also in Europe and China. The vast majority of China's high-tech giants — Alibaba, ByteDance, Tencent, just to name a few — use Nvidia hardware and virtually all European companies use Nvidia hardware.
Meanwhile, when it comes to China, Nvidia has major rivals both on the hardware and platform sides. On the hardware front, Nvidia has competitors like Biren Technology, InnoSilicon, and Moore Threads.
These companies are quite formidable competitors, even though for now their market share is negligible. All three companies use PowerVR GPU IP developed by the U.K.-based Imagination Technologies and have loads of experience with GPU development, according to Jon Peddie, the head of Jon Peddie Research.
The founder of Moore Threads, Zhang Jianzhong (also known as Zhang Jian Zhong), previously worked at Nvidia: he was the general manager of Nvidia's operations in China.
The founder of Biren Technology, Li Bing, previously worked at Huawei and also had experience at other tech companies. Co-CEO of Biren Technology is Allen Lee (also known as Li Xinrong), who used to be vice president and general manager of AMD's China R&D Center. He joined the startup in 2021 as co-CEO, and now he oversees organizational management and product design.
"Li Bing's background and expertise in the tech industry likely influenced his vision for Biren Technology, which focuses on developing high-performance GPUs for various applications," Peddie told Tom's Hardware.
The importance of Huawei
But while Biren, InnoSilicon, and Moore Threads have rather good hardware, for now they lack an ecosystem that is comparable to Nvidia's CUDA. However, there is a company in China that can compete with Nvidia not only on the hardware side of matters, but also on the platform level: Huawei. Huawei has its Cloud Matrix 384 system, which it claims can outperform Nvidia's GB200 NVL72 rack-scale machine for AI. Perhaps more importantly, the company has its own AI-oriented, heterogeneous Compute Architecture for Neural Networks (CANN) platform designed specifically to use the potential of Huawei's HiSilicon Ascend AI processors.
Just like Nvidia's CUDA, Huawei's CANN offers a complete suite of development resources such as runtime systems, model-building tools, and compilers. It works with both Huawei's MindSpore platform and widely-used AI libraries like TensorFlow and PyTorch, making it flexible for developers. The framework includes a broad range of tuned computational components to speed up model execution and is also compatible with ONNX Runtime, allowing it to run ONNX-based models efficiently on the company's Ascend accelerators for AI.
Although CANN is a key part of Huawei’s AI infrastructure, it has drawn criticism for being difficult to work with, mainly due to unstable performance, inadequate documentation, and reliability issues that complicate model training and deployment. Huawei has acknowledged these problems and is actively working to strengthen the platform and improve its usability. For now, imperfection of Huwei's CANN and significant efforts that are required to port programs from CUDA to CANN (several months and hundreds of developers) limit success of Huawei's hardware.
Nvidia's dominance may not last forever
However, if Nvidia's GPUs will be unavailable for Chinese and European buyers, they will at least consider Huawei, or perhaps Biren, Innosilicon, or Moore Threads hardware. This will not only decrease Nvidia's revenues by tens of billions every year and its market capitalization by hundreds of billions, but could also eventually make its competitors from China as trend setters in the AI segment, the company believes.
"Regardless of how one feels about DeepSeek’s open-source R1 model, it is a clear indication that innovation is moving rapidly around the world, with or without leading U.S. tech," the statement by Nvidia reads. "If U.S. platforms are absent, companies will turn to strategic competitors like Huawei to fill the gap. This is why leadership in AI depends not just on what we restrict — but on what we enable. Ecosystems are essential. It is not just about who can build the biggest data center or train the most advanced model. […]. One of Nvidia's key strengths is our global network of 6 million developers who build on our platforms. If we lose that ecosystem to our competitors, it will be almost impossible to get it back."
AI diffusion rule to be implemented from May 15
The new U.S. export rules for compute GPUs — known as the AI Diffusion Rule — come into effect on May 15. Under the Biden administration's AI Diffusion framework, unrestricted access to high-end AI chips like Nvidia's H100 is reserved for companies in the U.S. and a select group of 18 allied countries classified as 'Tier 1.' Companies in 'Tier 2' nations are subject to an annual limit of approximately 50,000 H100-class GPUs, unless they secure verified end user (VEU) approval. They can still import up to 1,700 units per year without a license, and these do not count toward the national quota. However, countries listed as 'Tier 3' — including China, Russia, and Macau — are essentially blocked from receiving such hardware due to arms embargo restrictions. The Trump administration is now reviewing this tier system to make it more straightforward and enforceable, and is rumored to make limitations for Tier 2 nations even stricter.
Not only will Nvidia cease to be able to sell its GPUs to China, which is one of its largest markets, but its Chinese customers will be forced to either use its GPUs in the cloud, or switch to processors developed in China, such as those designed by Huawei or one of the aforementioned companies. While this will slow down development of China's AI sector in the short term, it will give a strong boost for its AI hardware ecosystem in the mid and long-term future.
Huawei already spends tens of billions of dollars on R&D every year. Once Huawei and others increase sales of their AI hardware, they will be able to invest more in development of their AI ecosystems, which will get more competitive against those developed by Nvidia and other American companies (such as AMD and Intel) than they are today.
Huawei can become a global AI leader
Having China as a fortress and being able to sell its hardware elsewhere, Huawei and other Chinese companies will compete against Nvidia and other American entities for European and Middle-East AI hardware markets. What's more important, they will be able to set standards of the AI market and that will not only reduce Nvidia's influence on such standards, but it will greatly reduce American influence on AI development. AI leadership is about more than market share, it is about strategic control over future governance models.
"Today, the U.S. semiconductor industry is being pushed out of China, the world's largest market," the letter by Nvidia concludes. "On May 15th, if the AI Diffusion Rule comes into effect without significant changes, we will be forced to similarly retreat from the rest of the world."
The U.S. has already seen the consequences of ceding technological leadership, when Huawei gained a dominant foothold in global 5G deployments by offering cheaper and faster-to-deploy infrastructure. This serves as a cautionary example of how losing control over foundational standards can shift both market power and geopolitical influence. Nevertheless, whether the current administration has learnt from similar past mistakes remains to be seen.
