SiliconFlow: A Generative AI Computing Infrastructure Platform 137

SiliconFlow is a startup company focused on AI infrastructure (AI Infra), founded in August 2023. The company is dedicated to building a scalable, standardized, and highly efficient AI platform, providing a variety of AI model services to help developers and enterprises focus on product innovation. SiliconFlow has partnered with Huawei Cloud to launch the DeepSeek inference service based on Ascend Cloud, with performance comparable to globally leading GPU-deployed models. The founder, Yuan Jinhui, was the founder and former CEO of OneFlow. SiliconFlow has secured nearly 100 million RMB in an Angel+ round of financing.

Products and Services:

• DeepSeek R1/V3 Inference Service: The DeepSeek R1/V3 inference service launched by SiliconFlow in partnership with Huawei Cloud, based on Ascend Cloud. It offers performance comparable to globally leading GPU-deployed models and provides stable, production-grade inference services.

• SiliconCloud Platform: A one-stop cloud service platform launched by SiliconFlow, aggregating mainstream open-source large models and providing highly efficient, low-cost AI large model services.

Technical Advantages:

• Proprietary Inference Acceleration Engine: Through its self-developed inference acceleration engine, SiliconFlow achieves performance comparable to globally leading GPU-deployed models while reducing the cost of large model inference deployment.

• Multimodal Inference: In multimodal inference, SiliconFlow's open-source OneDiff inference acceleration engine delivers up to 3x performance improvement over comparable open-source products, achieving end-to-end image generation in hundreds of milliseconds.

What problems does SiliconFlow solve?

• Efficient Deployment and Inference of AI Large Models: Through its self-developed inference acceleration engine and partnership with Huawei Cloud, SiliconFlow provides high-performance, low-cost AI large model services, addressing the issues of low efficiency and high cost in large model training and inference.

• DeepSeek Service Interruptions: DeepSeek has frequently experienced service interruptions due to surging traffic and malicious attacks. SiliconFlow's partnership with Huawei Cloud to launch the DeepSeek R1 and V3 inference services based on Ascend Cloud has alleviated server pressure and provided more stable services.

• Lowering the Barrier to AI Technology Adoption: SiliconFlow offers free API call quotas and preferential pricing, supporting developers in achieving "Token freedom," reducing computing costs for enterprises and individual developers, and promoting the popularization and application of AI technology.

• Performance Enhancement for Multimodal AI Applications: SiliconFlow's multimodal inference engine OneDiff delivers up to 3x performance improvement, supporting end-to-end image generation in hundreds of milliseconds and enhancing AI performance in multimodal applications such as images and video.

• Promoting the Development of China's Domestic AI Ecosystem: The partnership between SiliconFlow and Huawei Cloud marks the rise of China's domestic AI ecosystem, providing fully domestic AI solutions, reducing dependence on foreign technology, and enhancing the country's competitiveness in the AI field.

Recently, SiliconFlow announced the launch of the DeepSeek-V3/R1 models, attracting over 3 million developers in the first week. In partnership with Huawei Cloud, it has launched inference services based on Ascend Cloud. Through this collaboration, SiliconFlow has achieved high-performance inference based on Ascend Cloud services, with performance on par with international top-tier standards.

SiliconFlow is committed to building a scalable, standardized, and highly efficient generative AI computing infrastructure platform, offering products such as the model cloud service platform SiliconCloud, the large language model inference engine SiliconLLM, and the high-performance text-to-image/video acceleration library OneDiff, helping enterprises and individual users deploy AI models efficiently and cost-effectively.