DeepSeek’s new AI model, R2, has still not been released despite its expected release in May. The company has remained silent about this delay, citing a shift in China’s technology policies.
DeepSeek R2 model may be delayed
According to information obtained by the Financial Times, the Chinese government is now requiring AI companies in the country, including DeepSeek, to use Huawei’s domestic chips instead of Nvidia’s graphics processors.
In line with this request, DeepSeek switched to Huawei’s Ascend chips for the training process of its R2 model. However, this transition failed to achieve the expected technical compatibility. It is reported that the Huawei Ascend chips do not reach the same performance level as Nvidia’s H20 GPUs, and that the training process is experiencing serious issues.
Huawei sent a support team to DeepSeek to resolve the issues. However, internal assessments indicate that this intervention was not sufficient to completely eliminate the problems.
DeepSeek-R1, launched in January, attracted significant global attention and propelled the Chinese company into the international AI scene. The R2 model, to be announced following the R1, was touted as a system promising improved architecture and broader capabilities.
However, hardware issues have postponed the announcement of this model indefinitely. Recent information from China suggests that DeepSeek-R2 could be released in September. However, if the technical issues remain unresolved, the company may face another delay.
US sanctions weren’t the only factor in China’s decision. A recent report by the Cyberspace Administration of China suggested that Nvidia’s H20 models sold to China may have a remotely deactivated mechanism, a “kill switch.”
{{user}} {{datetime}}
{{text}}