Archive — CN — technode
DeepSeek AI model faces delays amid US chip export restrictions
DeepSeek's AI model development is delayed due to US export restrictions on NVIDIA H20 chips, which are scarce due to US trade policies. The shortage also affects the existing DeepSeek-R1 model. CEO Wenfeng Liang is dissatisfied with DeepSeek-R2's performance, which has not yet been deployed amidst broader US efforts to restrict China's access to advanced AI hardware.
Key points
- DeepSeek-R2 development delayed due to US export restrictions on NVIDIA H20 chips.
- H20 chips are designed for the Chinese market but have limited availability.
- Previous model, DeepSeek-R1, utilized 50,000 NVIDIA Hopper GPUs, with 30,000 being H20s.
- H20 shortage impacts the deployment and performance of DeepSeek-R1.
- CEO Wenfeng Liang is pushing for improvements in R2, which is not yet ready.
Why it matters
- Current US policies restrict high-performance chip exports to China, affecting tech development.
- Similar restrictions have historically targeted technology transfer in US-China relations.
- By 2025, trends indicate a growing need for alternative chip sourcing and AI model innovation in China.
The delays in DeepSeek's AI model highlight the significant impact of US export controls on Chinese tech advancement, reflecting rising tensions and market challenges amid a global race for AI supremacy.