Xiaosuantai builds a lightweight computing power platform and continuously connects with the ecological resources of the industry.
Startup "Xiaosuantai" Launches an Intelligent Computing Power Full-Industry Chain Platform to Solve the AI Implementation Problems of Small and Medium-sized Enterprises
Against the backdrop of the explosive growth of generative artificial intelligence technology, computing power resources have become a key bottleneck for the implementation of the AI industry. Facing the problems commonly faced by small and medium-sized enterprises, such as "high computing power threshold, expensive deployment costs, and difficult operation and maintenance management", the intelligent computing power platform "Xiaosuantai", created by Dasantou (Hangzhou) Artificial Intelligence Technology Co., Ltd., provides a one-stop solution through technological integration and model innovation, promoting the transformation of AI technology from the laboratory to industrial applications.
Project Background: The Dilemma of Computing Power for Small and Medium-sized Enterprises in the AI Era
The training and inference of large AI models require huge computing power support. However, the standardized services of IDC rooms and cloud computing providers often struggle to meet the flexible and low-cost needs of small and medium-sized enterprises. When a financial institution attempted to apply a large model to strategy generation, the project was shelved due to a single training cost exceeding 100,000 yuan. A university laboratory spent several weeks debugging due to the complex configuration of the mirror environment... Similar cases reflect the shortcomings of the traditional computing power supply model.
The founding team of "Xiaosuantai" discovered from their actual service experiences that there is a significant structural contradiction between the supply and demand sides of computing power: computing power holders (such as IDC rooms and blockchain mining farms) have idle equipment, while demanders (such as AI enterprises and scientific research institutions) are restricted by scarce resources and high costs. Based on this, the team spent a year developing a computing power management system and a resource sharing platform, achieving efficient circulation of computing power resources through the "SaaS + service" model.
Technical Highlights: Heterogeneous Resource Scheduling and Lightweight Innovation
The core competitiveness of "Xiaosuantai" lies in its self-developed computing power management system. This system supports the mixed management of heterogeneous GPUs (such as A100, 4090, 3090), integrating local devices, hosted computing power, and public cloud resources onto the same platform. Through a distributed scheduling algorithm, the platform can monitor device temperature, power consumption, and usage in real-time, and achieve multi-task priority allocation and fault self-healing.
The technical team compressed the deployment time from the industry average of 3 days to within 4 hours through a lightweight architecture design. For example, a quantitative trading company completed the integration of the ChatGLM model in just 2 hours through a private deployment solution. The data ran entirely within the internal network, and the training efficiency increased by 30%. In addition, the platform pre-sets "out-of-the-box" environments for mainstream models such as ChatGLM, Whisper, and Stable Diffusion, allowing developers to start tasks without complex configuration.
Market Outlook: A Trillion-Yuan Blue Ocean and Differentiated Competition
According to data from the China Academy of Information and Communications Technology, the global intelligent computing power scale reached 1571 EFlops in 2024, accounting for 70% of the total computing power scale. With the iteration of large models such as Sora and Gemini and the surge in computing power consumption of Web3.0 applications, the market demand for flexible and low-threshold computing power services continues to rise.
Compared with traditional cloud providers, "Xiaosuantai" has a clear differentiated positioning:
1. Resource integration ability: By managing scattered IDC room and blockchain mining farm equipment, it constructs a hybrid computing power pool, with prices 30% - 50% lower than those of leading cloud providers;
2. Scenario-based services: It launches customized mirrors for industries such as law, finance, and education. For example, the legal document automatic generation system has helped law firms reduce 40% of manual writing costs;
3. Security and controllability: It supports private deployment and permission isolation, meeting the data compliance requirements of the finance and government affairs sectors.
Currently, the platform has served over 20 paying customers, including AI startup teams, quantitative trading institutions, and university laboratories, with an average daily task processing volume of over 5,000 times.
Future Plan: Building an Ecosystem for Computing Power Technology Sharing
Next, "Xiaosuantai" plans to advance in three stages:
1. SaaS-based computing power leasing: Cover 100 small and medium-sized enterprises by the end of 2025 and launch pay-as-you-go packages;
2. Low-code technology platform: Launch a developer community in 2026 to attract technology contributors to share AI plugins and deployment tools;
3. Industry chain integration: Collaborate with hardware manufacturers and industry associations to explore models for computing power financial derivatives and asset securitization.
The team revealed that the first-round financing is under negotiation, and the funds will be used for technology R & D and market expansion. "We are not the creators of large models, but the pathfinders for AI implementation." This statement by the founder may be the most appropriate footnote for this startup.