- Deepseeks V3 and R1 models are available via Huaweis Ascend Cloud Service
- They are driven by the ASCEND 910X -accelerators banned in the US, the EU and UK
- Prices are much lower than offered by Azure and AWS that have started trying Deepseek
Deepseek recently massively uncontrolled global markets with the launch of its open reasoning LLM, which was built and trained for a fraction of the cost of models from much larger US competitors, although Openai has since accused Deepseeks developers of using its models to train their.
A new paper had claimed Deepseeks V3 LLM was trained in a cluster of only 2,048 NVIDIA H800 GPUs – paralyzed versions of the H100 designed to comply with exporting restrictions to China. Rumors around Deepseek’s newer reasoning model, R1, suggest that it may have been trained on as many as 50,000 nvidia “jumps” GPUs, including H100, H800 and the newer H20, although Deepseek has not – and probably not – confirms This. If this is true, the serious question of China’s access to advanced AI hardware is raising despite ongoing trade restrictions, though it is no secret that there is a flowering black market for advanced NVIDIA AI hardware there.
Now, in a step that will shake western companies further, South China Morning Posts Reports Huawei Technologies’ Cloud Computing Unit has collaborated with Beijing-based AI-Infrastructure Start Silicon Flow to make Deepseeks models available to end users at an incredibly low price.
Driven by Huawei -hardware
This collaboration that was worked on during the Chinese lunar holiday from Lunar provides effective, cost-effective access to Deepseeks V3 and R1 models through Huawei’s Ascend Cloud Service operated by Huawei’s own homework solutions, including the controversial Ascend 910x accelerators, which is banned in the US, UK and Europe.
Huawei has made no secret that it wants to become the Chinese Nvidia, and Huawei Cloud claims its benefits are comparable to those from models running on Premium Global GPUs.
Siliconflow, hosting the Deepseek models, has come out swinging with some aggressive prices offering it to 1 yuan (about US $ 0.13) per day. 1 million input -tokens and 2 yuan for output -tokens with V3, while R1 access is priced at 4 yuan and 16 yuan.
Microsoft added Deepseek to his Azure AI foundry a few days ago, and Amazon followed quickly, adding LLM to his AWS ‘Bedock Managed Service. AWS showed the AI model using an ML.P5E.48XLARGE body, powered by eight Nvidia H200 GPUs, delivering 1128 GB GPU memory. However, it is early days for both cloud offers, and they prepare much more expensive than the Silicon Flow’s super-low price fixing.
The collaboration between Huawei, Siliconflow and Deepseek highlights China’s wider strategy to strengthen its domestic AI capabilities while reducing the addiction of Nvidia hardware.
The South China Morning Posts Remarks, “The move to launch could be used to promote military goals.”
