China's AI infrastructure is gaining momentum! Due to the rising fame of DeepSeek, Silicon-based Flow has completed a new round of financing worth hundreds of millions

Source: Hard AI

Author: Zhang Yaqi

After the explosive success of DeepSeek, there is more good news in the field of AI infrastructure in China.

On the 19th, Silicon Flow officially announced that the company has completed a Pre-A round financing of 100 million RMB by the end of 2024. This round of financing was led by Huachuang Capital, with PwC Capital participating, and existing shareholder Yaotu Capital continuing to over-invest, while Huaxing Capital acted as the exclusive financial advisor.

Silicon Flow was established in August 2023 and quickly attracted investor attention just months after its initial round of financing. Notably, prior to this, Silicon Flow had already brought in Meituan as a strategic shareholder, highlighting its important position in the AI infrastructure field.

This year, Silicon Flow launched the SiliconCloud platform, which integrates several mainstream large models, including Alibaba's Qwen2, Zhipu AI's GLM-4, and Huansheng Quantitative's DeepSeek-V2. The platform not only supports text generation but also includes advanced image generation models such as SDXL and PhotoMaker.

The company's founder, Yuan Jinhui, stated: "This financing will accelerate the upgrade and commercialization of Silicon Flow's AI cloud infrastructure. We will provide developers and enterprise clients with cost-effective, stable, and reliable product services, contributing to the implementation and prosperity of the global AI application ecosystem."

With the continuous development of AI technology and the expansion of application scenarios, the demand for computing power will continue to grow. CICC data predicts that various segments of AI infrastructure are expected to maintain a high-speed growth rate of over 30% in the next 3 to 5 years.

Silicon Flow: Solving the Computing Power Challenge of Large Models and Supporting the Explosion of Large Model Applications

DeepSeek, as a powerful AI assistant, has quickly gained market favor due to its outstanding performance in intelligent search, data analysis, and content creation. However, with the surge in user numbers, the "server busy" prompt has begun to appear frequently, severely affecting user experience.

According to the analysis by "Fat Treasure in the Melon Field," the user access volume of DeepSeek's servers has increased by 50% over the past year. This reflects both the strong appeal of large models like DeepSeek and exposes the bottlenecks in the current AI computing power infrastructure. Additionally, cyberattacks, such as DDoS attacks initiated by hackers, have exacerbated server pressure, leading to delays in responding to normal user requests.

In response to the computing power challenges posed by large models like DeepSeek, Silicon Flow has emerged. As a one-stop large model cloud service platform, Silicon Flow integrates high-performance computing resources through deep cooperation with Huawei Ascend, aiming to build a more powerful and stable AI service system.

Silicon Flow has recently made significant breakthroughs in technology applications. According to the company's official website, SiliconCloud has achieved the deployment of the full-version DeepSeek-R1 & V3 models on Huawei Cloud Ascend for the first time, pioneering the path for deploying DeepSeek models on domestic chips, significantly improving the inference speed and actual computing power utilization rate of large models based on domestic chips, greatly reducing model inference time and costs, and achieving results comparable to global high-end GPU deployment models

SiliconCloud has achieved explosive growth within less than a year of its launch. According to data from the company's official website, the total number of users on the platform exceeds three million, with an average daily call of over one hundred billion Tokens. The platform offers hundreds of mainstream models, including DeepSeek-R1 & V3, Qwen, FLUX.1, CosyVoice, and supports full-link services for these models.

In terms of industrial ecology, SiliconCloud has established partnerships with top cloud vendors, chip manufacturers, large model companies, and application companies such as Huawei, Amazon Web Services, and NVIDIA. For enterprise clients, the company has launched four service models, including exclusive instances, computing power management, private version MaaS, and an integrated machine based on Huawei's Ascend 910 series NPU.

Yuan Jinhui, the founder of SiliconCloud, stated in an interview with the "Daily Economic News" that this year, especially in the domestic market, the implementation of large models may be more reflected in B2B services. For AI infrastructure companies, the focus is on better meeting the needs of large model companies and industry clients, further lowering the threshold and cost of large model applications. As the cost of deploying large model inference continues to decrease, there will be more attempts at B2C applications, with greater opportunities for the emergence of super applications.

SiliconCloud: The "Builder" of Domestic AI Infrastructure, Accelerating the Popularization of AGI Technology

With the continuous evolution of AI technology, the popularization of AGI has become one of SiliconCloud's future visions. "In the future, SiliconCloud will continue to focus on the research and optimization of AI infrastructure technology, striving to occupy an absolute leading position in the global AI infrastructure market, enabling developers to achieve Token freedom through technological and product innovation, and promoting the popularization and application of AGI technology." SiliconCloud stated.

Although the total amount of investment and financing in the global AI field decreased by 31.2% year-on-year in the first quarter of 2024, the AI infrastructure market is still generally optimistic. CICC data predicts that various sub-sectors of AI infrastructure are expected to maintain a high growth rate of over 30% in the next 3 to 5 years.

Jiang Chun, a partner at PwC Capital, stated:

"The SiliconCloud team possesses both top-notch technical genes and continuous entrepreneurial experience, continuously promoting the inclusive development of large models with efficient AI infrastructure. Recently, they achieved a breakthrough in the domestic deployment of large models in collaboration with DeepSeek and Huawei's Ascend Cloud. We are optimistic about this team, which combines idealism with business acumen, as they become an irreplaceable builder in the large model ecosystem through breakthroughs in the 'deep water zone' of technology."

Wen Tingcan, Vice President of Investment at Yaotu Capital and a director at SiliconCloud, added:

"With outstanding technical engineering capabilities and precise grasp of market demands, the founding team has established a unique closed loop for technology commercialization—from technology research and development to product landing, from user accumulation to word-of-mouth fission, relying on the exponential word-of-mouth spread within the developer community, forming a dual reinforcement of technological barriers and ecological moats." The multiplier effect of 'excellent team × technological depth × productization capability × ecological momentum' is the core logic that we continue to invest in