The launch of DeepSeek's next-generation AI model, R2, is expected to face delays due to a shortage of NVIDIA H20 chips in China.

These chips, designed specifically for the Chinese market following US export restrictions, are essential for running DeepSeek's highly optimised models.

The ban on H20 shipments in April has triggered widespread concern among cloud providers about the scalability of R2, especially if it outperforms existing open-source models.

CEO Liang Wenfeng has reportedly held back the model's release, expressing dissatisfaction with its current performance.

Engineers continue refining R2, but the lack of compatible hardware poses a deeper challenge. DeepSeek's reliance on NVIDIA architecture makes switching to Chinese chips inefficient, as the models are tightly built for NVIDIA's software and hardware ecosystem.

Some Chinese firms have begun using workarounds by flying engineers to Malaysia, where NVIDIA chips are still available in local data centres.

After training their models abroad, teams return to China with trained systems. Others rely on gaming GPUs like the RTX 5090, which are easier to access via grey markets despite restrictions.

While Chinese tech giants ordered 1.2 million H20 chips earlier in 2025 to meet demand sparked by R1's success, inventory is still unlikely to support a full R2 rollout.

Companies outside China may launch R2 more easily without facing the same export hurdles.