DeepSeek’s R2 Model Hits a Snag: Huawei Chip Woes Force Delay

Tech giant DeepSeek slams the brakes on its R2 model rollout—blaming Huawei's chip supply chain chaos.

When silicon stumbles, AI stutters. The much-hyped R2 model—touted as DeepSeek's next-gen breakthrough—just got sidelined by Huawei's ongoing semiconductor struggles. No new launch date. No workarounds. Just another reminder that even cutting-edge AI kneels before hardware realities.

Meanwhile, crypto bros are probably thrilled—more time to shill 'AI-adjacent' tokens before the actual product arrives.

DeepSeek R2 AI model release delayed

The Chinese artificial intelligence company DeepSeek has postponed the launch of its latest AI model after encountering persistent technical challenges with Huawei’s Ascend processors.

DeepSeek had been encouraged by Chinese authorities to use Huawei’s chips instead of US-made Nvidia products after the successful release of its R1 model in January, but the company ran into major issues during the training phase of its R2 model.

According to people familiar with the matter, these problems forced DeepSeek to rely on Nvidia chips for training, while using Huawei’s Ascend chips for inference.

Industry insiders say Chinese chips, including Huawei’s, suffer from slower inter-chip connectivity, software limitations, and stability problems compared with Nvidia’s products.

Huawei deployed a team of engineers to DeepSeek’s offices to help adapt the model to its chips, but even with on-site support, the start-up could not complete a successful training run on Ascend hardware, sources said.

DeepSeek had planned to release the R2 model in May, but with the Ascend chip’s performance facing problems, the launch has been delayed.

Challenges to DeepSeek’s momentum

The delays to R2’s launch are worsened by a slower-than-expected progress in data labeling for the model’s updated capabilities, according to a person familiar with the matter.

Although some Chinese media reports suggest the new model could be released in the coming weeks, DeepSeek’s founder Liang Wenfeng has expressed dissatisfaction with its progress internally, urging the team to take the time needed to produce a model capable of maintaining the company’s competitive edge.

In the meantime, rivals have continued to gain ground. Ritwik Gupta, an AI researcher at the University of California, Berkeley, said that many developers in China have turned to Alibaba’s Qwen3 model, which he described as both powerful and flexible.

Gupta noted that Qwen3 adopted some of DeepSeek’s Core innovations, such as a training algorithm enabling reasoning capabilities, but it made them more efficient to deploy.

Gupta, who monitors Huawei’s AI ecosystem, said the Ascend platform is experiencing “growing pains” but predicted eventual improvement.

“Just because we’re not seeing leading models trained on Huawei today doesn’t mean it won’t happen in the future,” he said. “It’s a matter of time.”

Nvidia, a central player in US-China technology tensions, recently agreed to give the US government a portion of its China-related revenues in exchange for resuming sales of its H20 chips to Chinese customers. The company emphasized the importance of maintaining access to Chinese developers, warning that “surrendering entire markets and developers WOULD only hurt American economic and national security.”

KEY Difference Wire helps crypto brands break through and dominate headlines fast