Recently, domestic tech giants have invested 2 billion to procure NVIDIA’s H20 computing power, accelerating the large-scale implementation of AI applications. The policy has classified computing power as a new type of productive force, with the State Council aiming for a computing power scale exceeding 300 EFLOPS by 2025. Shenzhen is promoting “training vouchers” to accelerate intelligent computing infrastructure.
IDC predicts that the market size will exceed 300 billion by 2027, with an annual growth of 18.9%. The tariffs between China and the US have extended the procurement cycle, making domestic substitution urgent; technological independence and a complete ecosystem are driving the industry from substitution to leadership, opening up a trillion-dollar market.
This issue summarizes the six most authentic companies in computing power chips and domestic substitution:
Rank 6·Hengshuo Co., Ltd.
Focusing on integrated storage and computing AI chips andNOR Flash storage technology, its integrated storage and computing chip based onNOR Flash supports facial recognition, voice keyword recognition, and other terminal inference scenarios;7nm process AI chips achieve 8K real-time video processing, with machine learning training efficiency leading competitors.Core Advantage: The first domestic company to mass-produceNOR Flash integrated storage and computing chips, jointly promoting SRAM digital storage and computing technology research and development with the University of Science and Technology of China, covering both analog and digital technology paths.
Rank 5·Anlu Technology
Domestic FPGA chip leader, the 22nm Phoenix series supports 5G base station beamforming and AI acceleration, with 55nm/28nm products filling the high computing power market gap.Core Advantage: Self-developedPhoenix compiler and EDA toolchain closed loop, improving customer development efficiency by over 50%; products have been bulk procured by leading communication manufacturers such as ZTE and FiberHome, maintaining a top-tier market share in domestic FPGA.
Rank 4·Jingjia Micro
GPU domestic leader, the JM9 series supports 4K graphics rendering and OpenGL 4.0, with performance comparable to NVIDIA’s GTX 1050; the JM11 series is adapted for cloud-based AI training, significantly improving computing power density.Core Advantage: The only domestic listed company to achieve large-scale commercial deployment of GPU, with products entering the Loongson and Feiteng ecosystem, and the next generation of high-performance GPU benchmarking against international mainstream architectures.
Rank 3·Haiguang Information
x86 CPU and DCU dual-track breakthrough, the Deep Computing No. 3 DCU supportsLLaMA/GPT large model training, with performance benchmarking against NVIDIA’s A100; self-developed DTK software stack reduces the migration cost of Xinchuang servers, with ecological adaptability leading in the country.Core Advantage: The only domestic manufacturer to mass-producex86 CPU and DCU, with a continuously expanding market share in the Xinchuang market.
Rank 2·Lanqi Technology
One of the global memory interface chip giants, DDR5 chip rates reach 9000MT/s, with single server usage increasing 30%; the PCIe 5.0 Retimer chip empowers AI server GPU interconnection.Core Advantage: Deeply involved in the international DDR5 standard formulation, with full-generation solutions covering major global server manufacturers.
Rank 1·Cambricon
Cloud-based AI chip pioneer, the Siyuan 370 (7nm) INT8 computing power reaches 256TOPS, compatible with LPDDR5 memory; the Siyuan 590 (5nm) training cluster computing power reaches 2.3EFLOPS, supporting large models with hundreds of billions of parameters.Core Advantage: The first domestic company to achieve mass production of cloud-based AI chips, jointly customizing chips with ByteDance and Alibaba Cloud, with a technology path covering the entire chain from training to inference.
Sharing is not easy, remember to star, follow, like, and share as motivation to persist!
Like to watch, earn a million a month