China Achieves AI Independence: SiliconFlow Deploys DeepSeek V3 and R1 on Huawei Cloud, Bypassing Nvidia
In a major breakthrough for China’s artificial intelligence (AI) industry, SiliconFlow has successfully deployed the DeepSeek V3 and DeepSeek R1 models on Huawei Cloud’s Ascend AI infrastructure, eliminating reliance on Nvidia’s high-end GPUs. This marks a significant step toward AI self-sufficiency, especially as the United States tightens export restrictions on AI chips to China under President Donald Trump’s administration.
DeepSeek on SiliconFlow with Huawei Cloud vs. OpenAI: A Cost Comparison
China’s DeepSeek models running on Huawei Cloud are not only strategically important for AI independence but also significantly more cost-effective compared to OpenAI’s offerings. Below is a direct pricing comparison between DeepSeek and OpenAI models.
1️⃣ DeepSeek-V3 on Siliconflow vs. GPT-4o
Model | Input Cost (per 1M tokens) | Output Cost (per 1M tokens) |
---|---|---|
GPT-4o | $2.50 | $10.00 |
DeepSeek-V3 | $0.28 | $1.10 |
🔹 DeepSeek-V3 is approximately 9× cheaper than GPT-4o for both input and output tokens.
2️⃣ DeepSeek-R1 on Siliconflow vs. OpenAI o1
Model | Input Cost (per 1M tokens) | Output Cost (per 1M tokens) |
---|---|---|
OpenAI o1 | $15.00 | $60.00 |
DeepSeek-R1 | $0.55 | $2.21 |
🔹 DeepSeek-R1 is about 27× cheaper than OpenAI o1, making it an extremely cost-effective alternative.
The Bigger Picture: China's AI Ecosystem Gains Momentum
Beyond cost savings, China is making rapid progress in AI infrastructure deployment.
On February 2, Tencent Cloud announced seamless support for DeepSeek-R1 on its HAI high-performance computing service, enabling developers to deploy the model in just three minutes. Similarly, on February 1, Huawei Cloud and SiliconFlow launched a robust DeepSeek inference service on Huawei’s Ascend AI cloud platform, proving that China’s AI ecosystem can function without reliance on Nvidia chips.
This milestone comes as the U.S. government considers tightening AI chip export regulations, potentially restricting sales of Nvidia’s H20 AI chips to China. However, the successful deployment of DeepSeek models on Huawei Cloud signals China’s growing ability to develop and scale AI technology independent of Western hardware.
As China continues to accelerate AI development, DeepSeek’s competitive pricing, local cloud support, and independence from Nvidia GPUs could position it as a strong alternative to OpenAI models, shaping the future of AI in China and beyond.
Key Takeaways
- SiliconFlow now runs DeepSeek V3 and R1 on Huawei Cloud’s Ascend AI chips, bypassing Nvidia's high-end GPUs.
- Huawei’s Ascend AI cloud provides robust computing power, enabling stable and efficient AI inference.
- Tencent Cloud integrates DeepSeek R1 into its HAI platform, further accelerating AI application deployment.
- Trump administration is considering new export restrictions on AI chips, but China is now far less reliant on Nvidia hardware.
- Nvidia’s stock took a hit following DeepSeek’s rapid rise, as global markets reacted to China’s advancements in AI.
Deep Analysis: China’s AI Independence and the U.S. Response
For years, Nvidia’s AI chips, such as the H100 and H800, were considered indispensable for training and running advanced AI models. However, in 2022, the Biden administration banned the export of top-tier AI chips to China, prompting Nvidia to release the lower-performing H800 and H20 models for the Chinese market. Now, even these downgraded chips face potential export restrictions under Trump’s renewed focus on limiting China’s access to cutting-edge AI technology.
Despite these efforts, China’s AI industry has made remarkable progress in chip self-sufficiency. Huawei’s Ascend AI processors, developed under its Ascend series, now power a growing number of AI applications, including DeepSeek V3 and DeepSeek R1. The performance of these models on Ascend AI chips rivals their Nvidia-based counterparts like GPT 4o and GPT o1, proving that China has built a viable alternative to U.S.-supplied hardware and leading LLMs.
This shift is particularly symbolic given recent reports that Trump met with Nvidia CEO Jensen Huang to discuss new export controls. The U.S. government’s intention is to slow China’s AI progress by limiting access to Nvidia’s latest AI chips, but the successful deployment of DeepSeek models on Huawei Cloud suggests that China no longer depends on Nvidia as it once did.
Did You Know?
- DeepSeek’s rapid adoption has shaken the global AI landscape. The model’s launch triggered a $1 trillion loss in U.S. tech stock market value, with Nvidia’s stock price dropping 17% in one day.
- SiliconFlow provides direct access to DeepSeek models on Huawei Cloud. Users can test DeepSeek R1, DeepSeek V3, and Janus Pro 7B on the SiliconFlow website (cloud.siliconflow.cn).
- Huawei’s Ascend AI chips power more than just DeepSeek. The “Hundred Models, Thousand Applications” AI cloud initiative supports popular AI models like LLAMA and GLM, providing Chinese developers with diverse alternatives to Western AI models.
- DeepSeek models use large-scale reinforcement learning to achieve superior performance with minimal training data, making them highly efficient compared to other models in the same category.
The Bigger Picture
The rapid deployment of DeepSeek models on Huawei Cloud highlights a major shift in global AI power dynamics. With China moving toward full-stack AI self-sufficiency, U.S. efforts to restrict Nvidia’s AI chip exports may have less impact than expected. Instead of slowing down, China’s AI industry is accelerating, creating robust domestic alternatives to Western AI infrastructure.
This development raises significant questions about the effectiveness of U.S. technology sanctions. As China continues to build its AI ecosystem without Nvidia, the global tech industry will need to reassess its assumptions about AI leadership and innovation in the years ahead.