NVIDIA Outshines Chinese Government’s Efforts to Avoid Its AI Chips as Local Companies Demand GPUs, According to Broker Research

NVIDIA Outshines Chinese Government’s Efforts to Avoid Its AI Chips as Local Companies Demand GPUs, According to Broker Research

The following content is for informational purposes only and does not constitute investment advice. The author does not hold any financial positions in the stocks discussed here.

Insights into China’s AI GPU Market

Recent analysis from US brokers highlights several key trends in the competitive landscape of artificial intelligence (AI) graphics processing units (GPUs) in China. While domestic manufacturers are emerging vigorously, established Western companies like NVIDIA face increasing regulatory challenges.

Key Factors Shaping the AI GPU Landscape

Broker reports have identified four crucial factors influencing China’s AI GPU ecosystem:

  1. SMIC’s Production Challenges: The yield and capacity of SMIC’s 7nm node process continue to raise concerns. Notably, it is reported that most of Huawei’s Ascend 910C GPU units utilize TSMC’s 7nm dies, which Huawei acquired through complex transactions involving third-party routing.
  2. Strategies of Cloud Service Providers: Various Chinese cloud service providers (CSPs) are adopting different strategies to secure access to AI GPUs, particularly given US legislative efforts, like the Remote Access Security Act, aimed at limiting access to advanced Western AI resources.
  3. NVIDIA’s B40 AI GPU: The introduction of NVIDIA’s B40 chip, specifically designed for the Chinese market, is the latest development in the region’s evolving AI narrative. While the Trump administration has permitted NVIDIA to resume shipments of older H20 models to China, this chip is experiencing increasing scrutiny in Chinese policymaking.
  4. AI Capital Expenditure: China is pursuing complete self-reliance in AI computing, which necessitates substantial investments from leading industry players, thereby reshaping capital allocation strategies across the sector.

Technological Developments and Market Preferences

The recently launched DeepSeek’s DeepGEMM AI model, trained on NVIDIA GPUs, is written in CUDA and can be adapted by several domestic AI GPU manufacturers utilizing the UEBMO FP8 memory calculation format. Conversely, Huawei’s CloudMatrix 384, which integrates up to 384 Ascend chips, lacks native support for memory-efficient calculation formats like FP8. Although Huawei has developed a workaround to enable compatibility, expert reviews suggest this solution remains less than optimal.

Meanwhile, Alibaba is in the process of developing its own AI GPU, while Chinese firm Cambricon is currently experiencing a surge in market interest due to the impressive sales figures for its Siyuan 590 GPU. Despite the rise of domestic alternatives, many analyses still indicate a dominant preference for NVIDIA’s GPUs.

NVIDIA’s GPUs are particularly favored for their robust software ecosystem, notably through the CUDA platform, which significantly enhances clustering performance due to the NVLink interconnect technology.

Importantly, NVIDIA’s RTX Pro 6000D systems utilizing the B40 chip are exempt from needing additional licensing for sales in China, as these products utilize standard memory configurations primarily intended for inference applications rather than foundational model training. Consequently, these chips are expected to achieve rapid sales once they are accessible to Chinese enterprises.

For more detailed insights and updates on China’s evolving AI GPU landscape, visit here.

Related Articles:

New PlayStation 5 Digital Edition Update: Smaller SSD Design Features
Discover Amazing NVIDIA RTX 50 Gaming GPU Deals: Save Over $200 on Complete RTX PCs and Find MSRP Prices on 5060 and 5070 Models!

Leave a Reply

Your email address will not be published. Required fields are marked *