Chinese startup DeepSeek’s upcoming R2 large-scale language model is facing problems due to a shortage of Nvidia H20 server chips. In addition, DeepSeek’s CEO is still not satisfied with the new model’s performance.
Chinese startup DeepSeek’s upcoming R2 large-scale language model is facing problems due to a shortage of Nvidia H20 server chips. In addition, DeepSeek’s CEO is still not satisfied with the new model’s performance.
China’s AI GPU shortage has worsened after the recent US ban on Nvidia’s H20 chips, which were specifically designed for the Chinese market, Investing reports with reference to The Information. Most companies currently using DeepSeek’s popular R1 model are currently running on these H20 chips.
Although engineers have been working intensively on the new model for several months, DeepSeek CEO Liang Wenfeng is not yet satisfied with its performance. The company has not yet determined when the R2 will be released.
The startup’s engineers are currently continuing to refine R2 and will only release it after receiving approval from Liang. If R2 outperforms current open-source models upon release, demand is expected to outpace Chinese cloud providers, which are already facing a shortage of Nvidia chips.
DeepSeek R2 is the long-awaited successor to DeepSeek R1, the AI model that made a splash in the tech industry over the winter. R2 is expected to offer improved coding capabilities, multilingual reasoning, and increased computational efficiency.
DeepSeek R2 is expected to be a multimodal model, meaning it will be able to process not only text but also images, audio, and possibly even video, making it a more versatile AI system. Details about its exact architecture remain unknown, but industry insiders speculate that it will have expanded training datasets for reinforcement learning (RL).
The release of DeepSeek R2 was planned for May 2025 to build on the momentum created by R1, but the model has not yet been released.
