DeepSeek is grappling with significant challenges in accessing NVIDIA’s advanced AI chips due to US export controls, which are slowing China’s progress in AI development. These export restrictions have specifically hindered the flow of NVIDIA’s H20 AI accelerators to China, stalling the anticipated release of DeepSeek’s R2 model.
Since the inception of these controls during President Trump’s administration, the goal has been to ensure the US maintains an edge in AI technology. As a result, DeepSeek’s CEO, Liang Wenfeng, has expressed dissatisfaction with the large language model’s (LLM) performance. Chinese cloud service providers (CSPs) are unable to adopt the R2 model effectively due to the chip shortage.
Even though China has attempted various “workarounds” to secure AI chips, the recent restrictions have proved to be a major obstacle for DeepSeek. The company’s previous model, the R1, significantly impacted NVIDIA’s market value, increasing anticipation for the R2 model. Unfortunately, the launch of R2 in China remains uncertain as domestic CSPs struggle to gather the necessary computing power.
Reports suggest that the existing inventory of NVIDIA’s H20 GPUs in China is primarily utilized by cloud customers running the R1 model. This scarcity of high-end chips makes it incredibly challenging for DeepSeek to deploy the R2 model at scale. The export controls have effectively curtailed China’s AI ambitions.
Previously, DeepSeek was rumored to have ties with the Chinese military, allegedly accessing AI chips through “shell companies” across Asia. Despite these efforts, acquiring NVIDIA’s top-tier chips has proven difficult. While alternatives like Huawei exist, they may not provide the scale needed for a mass rollout of the R2 model in domestic markets.
