- Deepseek delayed launch R2 due to persistent technical difficulties with Huawei Ascend Chips
- Nvidia H20 systems remain more reliable for IA training than interior ascend equipment
- Alibaba Qwen3 uses Deepseek delays, incorporating basic algorithms while improving efficiency and flexibility
The Chinese AI giant Deepseek apparently encountered unexpected delays in the publication of his latest model, R2, after having faced persistent technical difficulties with the Huawei Ascend chips.
The company had been encouraged by the Chinese authorities to adopt national processors instead of relying on H20 systems in Nvidia, which are generally considered more mature and reliable.
Although Huawei engineers are on site to help, Deepseek could not finish successful training using the Ascend fleas – and consequently, the company relied on Nvidia equipment for training while using Ascence for inference tasks.
Technical challenges delay the R2 development
The R2 launch, initially scheduled for May 2025, has been postponed due to these technical obstacles and longer data labeling than expected for the set up to date.
The founder of Deepseek, Liang Wenfeng, would have expressed his dissatisfaction with the progress of the model, stressing the need for additional development time to produce a model capable of maintaining the competitive point of Deepseek.
Meanwhile, competitors like Alibaba Qwen3 were able to take advantage of this delay, as it incorporated the main Deepseek training algorithms while improving efficiency and flexibility, showing how fast IA ecosystems can evolve even when a single start-up.
Beijing’s wider thrust for AI self -sufficiency has exerted pressure on national companies to adopt local equipment.
In practice, however, this strategy has revealed gaps in the stability, inter-hue connectivity and the maturity of software between Huawei fleas and Nvidia products.
The developers continue to play a crucial role in the formation of the success of the ECOC ecosystems – NVIDIA has emphasized the maintenance of access to Chinese developers is strategically important, warning that the restriction of the adoption of technologies could harm the interests of economic and national security.
Chinese AI companies, on the other hand, must balance government pressures with practical realities in the development and deployment of LLMS.
Despite these setbacks, the R2 model of Deepseek could still be released in the coming weeks.
The model is likely to do a meticulous examination concerning its performance compared to competitors trained in more mature equipment, offering a clear example of tension between political ambitions, technical capacity and the deployment of real world.
Via Arstechnica
