Global AI Accelerator Chip Market Overview and Emerging Chinese Vendors (2023)

The AI accelerator chip market is currently dominated by overseas leaders, with Nvidia holding an 82% share in 2022 data‑center AI acceleration, followed by AWS (8%), Xilinx (4%), and smaller shares for AMD, Intel and Google. Intel's share has slightly declined but it remains a major player.

In the GPU segment, Nvidia, AMD and Intel together control the global market. Nvidia's independent graphics cards captured 84% of the Q1 2023 market, while AMD and Intel held 12% and 4% respectively. Nvidia’s GeForce RTX 40 series, based on the Ada Lovelace architecture and 5 nm process, features 760 billion transistors and delivers a 70% increase in core count over Ampere.

Domestic Chinese firms are accelerating their GPU and GPGPU development. Companies such as HaiGuang (DCU series), TianShu ZhiXin (7 nm GPGPU "TianHai"), BiRui Technology (BR100/BR104), Moore Threads (MTT S2000/S3000), and MuXi (MXN100) have announced products with specifications ranging from 1.5 TFLOPS FP32 performance to 160 TOPS INT8 throughput, often leveraging 7 nm or 5 nm processes, chiplet designs, and high‑bandwidth memory.

In the GPGPU space, Nvidia and AMD remain the primary vendors, with Nvidia’s CUDA ecosystem dominating academic AI research. Google’s TPU v4, based on low‑precision matrix‑multiply optimizations, offers high efficiency for transformer training but lacks the general‑purpose flexibility of GPUs.

ASIC solutions are also gaining traction. Cambricon’s MLU series (MLU100, MLU290, MLU370) and Huawei’s Ascend series (310, 910) provide high‑performance AI inference and training capabilities, with peak INT8 performance exceeding 1,000 TOPS and FP16 performance up to 320 TFLOPS, respectively.

The analysis concludes that while Nvidia’s lead is secure, Chinese manufacturers are narrowing the gap through ecosystem compatibility (e.g., CUDA support) and differentiated hardware designs, positioning themselves for significant opportunities as large‑model AI workloads continue to expand.