다섯 기업이 NVIDIA의 지배력에 대항하기 위해 차세대 연산 가속기를 개발하고 있다.

The artificial intelligence boom has created an almost insatiable demand for computing power. Behind every chatbot, image generator, and reasoning model lie specialized chips executing vast numbers of calculations within seconds.

Nvidia dominates much of this market, but it is no longer running the race alone. Huawei, Cambricon, Moore Threads, and Biren Technology are developing alternative processors as China seeks to reduce its dependence on foreign technology.

1. Nvidia

Nvidia remains the undisputed leader in the global AI chip industry. Originally developed for gaming, its graphics processing units (GPUs) have become the primary hardware for training and operating many of the world’s most advanced AI models. However, Nvidia’s greatest advantage extends beyond raw processor performance. Its CUDA software platform, high-speed networking technology, and expansive developer ecosystem make migrating workloads to competing hardware both costly and technically complex. This dominance remains especially evident in China, despite years of U.S. export restrictions. According to the SCMP, Chinese AI developers continue relying on Nvidia processors for advanced model training because replacing the entire hardware and software stack demands substantial engineering effort.

2. Huawei

Huawei has emerged as China’s most credible alternative to Nvidia. Its Ascend processors are engineered for AI training and inference, while its SuperNode systems integrate multiple chips to meet the escalating computational demands of increasingly complex models. The Shenzhen-based tech giant is simultaneously developing the software required to support its hardware. Its Compute Architecture for Neural Networks (CANN) aims to serve as a domestic counterpart to Nvidia’s widely adopted CUDA platform. Huawei recently validated this hardware-software integration through a collaboration with DeepSeek. The company’s Ascend 950PR and 950DT processors received immediate compatibility with DeepSeek V4, enabling the model to run inference across Huawei’s Ascend SuperNode infrastructure.

3. Cambricon Technologies

Cambricon Technologies is another major player in China’s rapidly expanding AI chip sector. Unlike diversified tech conglomerates, Cambricon focuses exclusively on developing processors and accelerator cards for AI applications across servers, data centers, and smart devices. Its machine-learning units are optimized for the parallel computations required to train models and generate outputs. Demand for these domestic chips has surged as export restrictions curtail Chinese firms’ access to Nvidia’s most advanced processors. Today, Cambricon and Huawei jointly lead the domestic expansion within China’s AI server market.

4. Moore Threads

Moore Threads is developing general-purpose GPUs capable of supporting artificial intelligence, graphics rendering, and scientific computing. This versatile architecture allows its processors to manage diverse workloads rather than functioning exclusively as specialized AI accelerators. Founded in 2020, the company has quickly become one of China’s most closely monitored semiconductor startups. It is positioning its GPUs as domestic replacements for organizations that previously relied on Nvidia and other foreign suppliers. Market demand for Moore Threads’ technology appears to be accelerating rapidly.

5. Biren Technology

Biren Technology is tackling one of the industry’s most formidable challenges: designing high-performance GPUs for AI training and high-performance computing. These workloads demand massive processing capacity, ultra-fast memory, and highly efficient interconnects linking thousands of individual accelerators. The company has encountered significant hurdles following U.S. restrictions that severed its access to advanced semiconductor manufacturing. Nevertheless, China’s drive for technological self-reliance is opening new avenues for Biren and other domestic chipmakers. Biren has projected that first-half revenue could surge by as much as 2,107 percent amid surging demand for locally produced chips. However, rapid sales growth does not automatically indicate that its processors have fully bridged the performance and software divide with Nvidia.

A Race Involving More Than Processing Power

Nvidia still maintains the industry’s most comprehensive suite of high-performance processors, networking infrastructure, and developer software. Consequently, its competitors must build complete computing ecosystems rather than simply release chips with strong benchmark scores. China’s massive appetite for AI infrastructure may still accommodate the growth of several domestic suppliers. While Huawei currently leads this domestic push, Cambricon, Moore Threads, and Biren Technology are also solidifying their positions as key players in an increasingly fragmented global semiconductor landscape.