- Deepseek V3 and R1 models are available via Huawei Cloud Ascend Service
- They are fed by the ACCSED 910X accelerators prohibited in the United States, the EU and the United Kingdom
- The price is much lower than that offered by Azure and AWS which began to test Deepseek
Deepseek recently disrupted the world markets with the launch of its Open Reasoning LLM, which was built and formed for a fraction of the cost of much larger American competitors models, although Optaai has since accused Deepseek developers to use its models to form theirs.
A new newspaper said that V3 LLM from Deepseek had been trained on a group of only 2,048 GPU NVIDIA H800 – paralyzed from H100 designed to comply with American export restrictions to China. The rumors around the new Deepseek reasoning model, R1, suggest that it may have been formed by up to 50,000 GPU “Hopper” Nvidia, including H100, H800, and the new H20, although Deepseek N ‘Not – and probably not – confirmed that – confirm this. If it is true, it raises serious questions about China’s access to Advanced AI equipment despite the current commercial restrictions, although it is not a secret, there is a flourishing black market for advanced equipment Nvidia ai.
Now, in a decision that will shake more Western companies, the Morning post in southern China The reports of the Huawei Technologies Cloud Computing Unit established a partnership with Siliconflow from Infrastructure AI Infrastructure based in Beijing to make Deepseek models available for end users at an incredibly low price.
Propelled by Huawei equipment
This collaboration, on which was worked during the Chinese New Year Holidays, offers effective and profitable access to the V3 and R1 models of Deepseek via the Huawei Cloud Cloud service, which is propelled by local Huawei solutions, including Controversial accelerators ascende 910X which, which is prohibited in the United States, the United Kingdom and Europe.
Huawei did not hide that he wanted to become the Chinese Nvidia, and Huawei Cloud says that his performance levels are comparable to those of the models carried out on premium global GPUs.
Siliconflow, which hosts the Deepseek models, was released by swinging with aggressive prices, offering it for 1 yuan (approximately 0.13 US) by 1 million input tokens and 2 yuan for output tokens with V3 , while access R1 is at the price of 4 yuan and 16 yuan.
Microsoft added Deepseek to its Azure foundry a few days ago, and Amazon quickly followed suit, adding the LLM to its service managed by the rocky substratum of AWS. AWS presented the AI model using an instance ML.P5E.48XLARGE, supplied by eight GPU NVIDIA H200 providing 1128 GB of GPU memory. However, it is early for both cloud offers, and they work much more than the super low Siliconflow prices.
The collaboration between Huawei, Siliconflow and Deepseek highlights China’s broader strategy to strengthen its national AI capacities while reducing dependence on Nvidia material.
THE Morning post in southern China Note: “The decision to launch Deepseek models on a local material dorsal thorn highlights China’s progress in reducing dependence on foreign technologies and strengthening its IA industry in the middle As the United States growing efforts to stifle high-end china access which, according to the United States government, could be used to advance military objectives. »»
