The Chinese AI start-up DeepSeek has reportedly postponed the launch of its new model after technical issues with Huawei Technologies’ chips forced a reliance on U.S.-made Nvidia Corp.'s NVDA processors.
Huawei Chips Stumble In Critical Training Phase
DeepSeek encountered persistent technical problems while attempting to train its R2 model using Huawei's Ascend processors, reported the Financial Times, citing sources familiar with the situation.
The difficulties forced the company to use Nvidia chips for training while relying on Huawei hardware only for inference tasks.
The launch, originally set for May, was delayed due to these technical hurdles, causing DeepSeek to fall behind competitors in the AI race, the report said.
Chinese Government's Push For Domestic Chips Faces Limits
The setback underscores the limits of China's drive to reduce dependence on U.S. technology. Beijing has reportedly encouraged companies to favor domestic AI chips and is scrutinizing orders of Nvidia's H20 processors to promote local alternatives.
Industry insiders note that Chinese chips still lag in stability, software support and inter-chip connectivity compared with Nvidia.
Huawei even dispatched engineers to DeepSeek's offices to assist with training, yet the company could not achieve a successful run with Ascend chips. DeepSeek continues to work with Huawei to make the model compatible for inference, the report added, citing sources.
Revenue, Security And Geopolitics Collide As US-China Tensions Rise
Nvidia and Advanced Micro Devices, Inc. AMD earlier this month agreed to remit 15% of their China chip sales to the U.S. government to secure export licenses.
Meanwhile, Chinese authorities have expressed security concerns over Nvidia's H20 chips, warning companies to avoid using them in government projects or sensitive infrastructure.
"Just because we're not seeing leading models trained on Huawei today doesn't mean it won't happen in the future. It's a matter of time," Ritwik Gupta, an AI researcher at the University of California, Berkeley, told the publication.
DeepSeek R1 Model Had Triggered A $600 billion Drop In Nvidia’s Value
Founder Liang Wenfeng reportedly pushed for more time to advance R2, which also faced delays due to extensive data labeling for the updated model. Chinese media suggests the launch could occur in the coming weeks.
DeepSeek sparked a $600 billion wipeout in Nvidia's market value with the debut of its R1 model in January.
At the time, Nvidia acknowledged DeepSeek's R1 AI model, stating that the company's work demonstrates how new models can be developed using test-time scaling, leveraging widely available models and compute that fully comply with export controls.
Read Next:
Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.
Photo courtesy: mundissima on Shutterstock.com
Edge Rankings
Price Trend
© 2025 Benzinga.com. Benzinga does not provide investment advice. All rights reserved.
