In an exciting development for the Chinese AI landscape, DeepSeek and Huawei have announced a strategic collaboration. This partnership has led to the launch of SiliconCloud, a fully domestic large-scale model cloud service platform. SiliconCloud's initial offering features DeepSeek-V3 and DeepSeek-R1 models, powered by Huawei Cloud's Ascend AI infrastructure.
The collaboration between DeepSeek and Huawei marks a significant step towards building a self-reliant AI ecosystem in China. SiliconCloud offers several key advantages:
All-Chinese Solution: SiliconCloud leverages Huawei's Ascend AI infrastructure and DeepSeek's advanced models, creating a completely domestic AI solution.
Cost-Effective Pricing: SiliconCloud offers competitive pricing, making advanced AI models accessible to a wider range of users. During its introductory period, DeepSeek-V3 costs just 1 RMB/M tokens for input and 2 RMB/M tokens for output, while DeepSeek-R1 is priced at 4 RMB/M tokens for input and 16 RMB/M tokens for output, mirroring DeepSeek's official discounts (valid until February 8th, 2025).
User-Friendly Deployment: Developers can easily integrate DeepSeek models into their applications by using the SiliconCloud API. This simplifies the deployment process, removing barriers for developers who want to leverage the power of large language models.
Stable and Production-Ready: SiliconCloud offers a reliable and scalable inference service, ensuring that DeepSeek models can be used in large-scale production environments to support commercial deployments.
The launch of SiliconCloud has generated immense interest, with users globally flocking to the platform in overwhelming numbers. According to SiliconCloud's founder and CEO Yuan Jinhui, the initial traffic exceeded the team's expectations, despite anticipating significant demand based on DeepSeek's popularity during the Spring Festival. As a result, SiliconCloud implemented traffic restrictions within the first 10 hours of its launch, demonstrating the huge interest in accessible, domestic AI solutions.
SiliconCloud emphasizes that its solution, driven by its proprietary inference acceleration engine and Huawei Cloud Ascend, enables DeepSeek models to perform on par with those deployed using high-end GPUs.
The platform also gives access to DeepSeek’s multimodal model, Janus-Pro-7B, with limited access.
A quick benchmark test was conducted, comparing DeepSeek-V3 and DeepSeek-R1 in terms of response speed and output quality. Both models exhibited similar response times, outputting approximately 15-16 tokens per second. DeepSeek-V3 provided more direct answers, while DeepSeek-R1 demonstrated a more comprehensive understanding of context and nuances.
This collaboration extends beyond simply hosting DeepSeek models on Huawei Cloud. Yuan Jinhui revealed that part of the collaboration will be integrated into Huawei software packages.
DeepSeek has rapidly gained recognition in the AI community, with its daily active users reaching 19.09 million in just 19 days, surpassing even ChatGPT. Furthermore, DeepSeek-R1 achieved first place on the City-Data creative writing leaderboard, dethroning Claude after a seven-month reign.
The DeepSeek-Huawei collaboration represents a major step forward for China's AI industry. By providing access to powerful AI models at affordable prices, SiliconCloud has the potential to empower developers and accelerate innovation across various sectors. Despite initial traffic limitations, the platform's launch signals a strong demand for domestic AI solutions and highlights the growing competitiveness of the Chinese AI ecosystem.
