DeepSeek hits brakes on R2 AI model's release: Here's why
What's the story
Chinese artificial intelligence (AI) start-up, DeepSeek, has delayed the launch of its highly-anticipated R2 model. The decision comes as CEO Liang Wenfeng expressed dissatisfaction with the model's performance, The Information reported. The R2 is a successor to DeepSeek's hugely successful R1 reasoning model and was originally scheduled for a May release.
Model advancements
Engineers have been fine-tuning the model for months now
The R2 model is expected to improve coding capabilities and support languages other than English. However, DeepSeek's engineers have been fine-tuning the model for months now, waiting for Liang's approval to launch it. The delay in the release could also be attributed to potential challenges in adoption due to limited capacity of Chinese cloud providers.
Chip shortage
Lack of NVIDIA server chips could affect R2's adoption
The rapid adoption of the R2 model could be hindered by a lack of NVIDIA server chips in China, owing to US export restrictions. This was revealed by employees at leading Chinese cloud companies that provide DeepSeek's models to enterprise customers. The report further stated that DeepSeek has been in talks with some Chinese cloud firms, giving them technical specifications for hosting and distributing the model from their servers.
Trade restrictions
Trump's export curbs complicate chip availability in China
The Trump administration's new export curbs imposed in April have stopped NVIDIA from selling its H20 chips in the Chinese market. These were the only AI processors that could be legally exported to China at the time. The restriction further complicates DeepSeek's plans for a successful launch and widespread adoption of its R2 model.