- Deepseek V3 and R1 models are available through the Huawei Ascend Cloud Service
- They are driven by Ascend 910x accelerators in the United States, the EU and the United Kingdom
- The price is much lower than the one offered by Azure and AWS that have begun to try Deepseek
Deepseek recently restless in global markets with the launch of its open reasoning LLM, which was built and trained by a fraction of the cost of much larger American competitors models, although Openai has accused Deepseek developers to use their models for train yours.
A new article had claimed V3 LLM of Deepseek was trained in a group of only 2,048 GPU NVIDIA H800: paralyzed versions of the H100 designed to comply with US export restrictions to China. The rumors about the new Deepseek reasoning model, R1, suggest that it may have been trained in up to 50,000 gpu “hopper” Nvidia, including H100, H800 and the new H20, although Deepseek has not done so, and probably will not do so, Confirm this, confirm this. . If it is true, it raises serious questions about China’s access to advanced AI Hardware despite the ongoing commercial restrictions, although it is no secret that there is a prosperous black market for advanced hardware of NVIDIA there.
Now, in a movement that will shake even more to Western companies, the Post in the morning of southern China The reports of the Huawei Technologies cloud computing unit has been associated with the beginning of the AI -based infrastructure based on Beijing, Siliconflow to make available to Deepseek models to end users at an incredibly low price.
Huawei Hardware driven
This collaboration, in which it worked during the Chinese lunar New Year holiday including controversial accelerators ascend 910x that are prohibited in the United States, the United Kingdom and Europe.
Huawei has not hidden that he wants to become the Chinese Nvidia, and Huawei Cloud states that his performance levels are comparable to those of the models that are executed in GPU Global Premium.
Siliconflow, which houses the Depseek models, has been balance A price of 4 yuan and is 4 yuan and is priced at 4 yuan 16 yuan.
Microsoft added Depseek to his Azure Ai Foundry a few days ago, and Amazon quickly followed his example, adding the LLM to the service managed by the rock bed of his AWS. AWS showed the AI model using an instance ML.P5E.48xlarge, fueled by eight NVIDIA H200 GPU that delivers 1128 GB of GPU memory. However, it is early for both cloud offers, and work much more expensive than super low siliconflow prices.
The collaboration between Huawei, Siliconflow and Deepseek highlights China’s widest strategy to strengthen their domestic abilities while reducing Nvidia hardware dependence.
He Post in the morning of southern China He points out: “The movement to launch Depseek models in a trunk network of crop hardware own China’s progress high of the United States to drown China’s access to high -end chips that the United States government said it could be used to advance military objectives. “
