DeepSeek Delays New Models Due to Issues with Huawei Chips

• DeepSeek has delayed the release of model R2 due to issues with Huawei chips.
• For training the new models, the company switched to NVIDIA chips.
• The use of foreign Huawei chips is facing technical difficulties.

This is reported by Mezha, Ars Technica.

The Chinese company DeepSeek has delayed the release of new models due to difficulties in their training, caused by technical problems with Huawei's foreign-made chips. As reported by Reuters, the company was using Ascend chips for training its large language models, but due to problems with them, it started using NVIDIA chips.

Earlier, DeepSeek had planned to release model R2 in June, but issues with Ascend forced the company to postpone the deadline. Ascend chips are now being used for generating responses to user queries, which require less computational power.

The situation with DeepSeek underscores the general difficulties of Chinese companies in using foreign-made chips. With active support from Huawei's chips on the part of the Chinese government, most developers of artificial intelligence continue to choose NVIDIA.

Currently, NVIDIA is under pressure from both the US and China, as the Chinese government is concerned about the potential risks of using foreign chips, while the US government is looking to limit the export of advanced chips to China. Recently, former US President Donald Trump announced restrictions on the export of H20 AI chips to China following a meeting with NVIDIA's CEO Jensen Huang.

DeepSeek gained popularity in the world in June 2025 with the release of its flagship model DeepSeek-R1, which was considered competitive with foreign analogs, but was significantly cheaper in training. Model R2 was supposed to be the next flagship model of the startup, but technical problems with Ascend hinder this.

Tags: Technology/AI