- Deepseek delays the launch of R2 due to persistent technical difficulties with Huawei Ascend chips
- NVIDIA H20 systems are still more reliable for the training of national ancestry hardware
- Alibaba’s qwen3 exploits Deepseek delays, incorporating central algorithms while improving efficiency and flexibility
The Chinese giant of the Deepseek has apparently found unexpected delays in the release of its latest model, R2, after facing persistent technical difficulties with the Ascend chips of Huawei.
The company had been encouraged by Chinese authorities to adopt national processors instead of trusting NVIDIA H20 systems, which are generally considered more mature and reliable.
Although Huawei engineers are on site to help, Depseek could not complete a successful training race using Ascend chips, and as a result, the company was based on NVIDIA hardware for training while using Asce for inference tasks.
Technical challenges delay the development of R2
The launch of R2, originally scheduled for May 2025, was postponed due to these technical obstacles and a data labeling longer than expected for the updated training data set.
According to the reports, Deepseek founder Liang Wenfeng, expressed dissatisfaction with the progress of the model, emphasizing the need for additional development time to produce a model capable of maintaining Deepseek’s competitive advantage.
Meanwhile, competitors such as Alibaba’s QWEN3 were able to take advantage of this delay, since it has incorporated the basic training algorithms of Deepseek while improving efficiency and flexibility, showing how quickly the ecosystems of AI can evolve even when a single start fight.
The broader impulse of Beijing for the self -sufficiency of AI has pressed national companies to adopt local hardware.
In practice, however, this strategy has revealed gaps in stability, connectivity between chip and software maturity between Huawei chips and NVIDIA products.
Developers continue to play a crucial role in configuration of the success of IA: NVIDIA ecosystems has emphasized the maintenance of access to Chinese developers, it is strategically important, warning that restricting the adoption of technology could damage economic and national security interests.
Meanwhile, Chinese companies must balance government pressures with practical realities in the development and implementation of LLM.
Despite these setbacks, Depseek’s R2 model can still be launched in the coming weeks.
It is likely that the model in front of a scrutiny regarding its performance in relation to the rivals trained in more mature hardware, offering a clear example of the tension between political ambitions, technical capacity and the implementation of the real world.
Through Arstechnica
