AI chips race heats up as DeepSeek gives Chinese chipmakers a boost

The development of AI models by Chinese company DeepSeek marks a significant shift in China‘s artificial intelligence chip landscape, particularly in relation to inference-based computing tasks. As U.S. export restrictions limit access to advanced chips, Chinese manufacturers are seeking alternative paths to AI development.

The big picture: DeepSeek’s focus on inference-based AI processing provides Chinese chipmakers an opportunity to compete more effectively in their domestic market, despite lagging behind U.S. competitors in raw processing power.

• DeepSeek’s models prioritize computational efficiency over pure processing capability, allowing Chinese manufacturers to partially bridge the performance gap with U.S. chips
• Major Chinese tech companies including Huawei, Hygon, EnFlame, Tsingmicro, and Moore Threads have announced support for DeepSeek models
• The model’s open-source nature and low fees are expected to accelerate AI adoption across Chinese industries

Market dynamics and adoption: Chinese companies across various sectors are embracing DeepSeek’s technology, while established players like Huawei are finding new opportunities in the inference computing space.

• ByteDance has recognized Huawei’s Ascend 910B chip as particularly suited for inference tasks, which involve using trained AI models for predictions and operations
• Numerous Chinese companies, from automotive to telecommunications sectors, have announced plans to integrate DeepSeek’s models into their operations
• The focus on inference workloads aligns well with Chinese AI chipset vendors’ current capabilities, according to Omdia chief analyst Lian Jye Su

Nvidia’s continuing dominance: Despite Chinese advancements, Nvidia maintains significant advantages in both training and inference capabilities.

• Nvidia’s CUDA platform remains a crucial component of its market leadership, providing developers with extensive tools for general-purpose computing
• The company can still sell less powerful training chips to Chinese customers for inference tasks, despite U.S. export restrictions on advanced AI chips
• Nvidia has emphasized the growing importance of inference time in AI applications, asserting that its chips are essential for making reasoning models more effective

Technical challenges: Chinese manufacturers face significant hurdles in attempting to compete with established U.S. technology.

• Huawei’s CANN platform, designed to compete with Nvidia’s CUDA, struggles to convince developers to switch from the well-established CUDA ecosystem
• Chinese AI chip companies’ software capabilities lag behind Nvidia’s extensive library and diverse software offerings
• Bernstein analyst Lin Qingyuan notes that while Chinese AI chips are cost-competitive for inferencing, their appeal remains largely limited to the domestic market

Looking ahead: While DeepSeek’s emergence represents a meaningful step forward for China’s AI chip industry, the path to true technological parity with U.S. competitors remains challenging. Success will likely depend on continued investment in software development and the ability to build robust developer ecosystems around Chinese platforms.