On November 6, the 26th China Patent Award Ceremony was held in Wuxi. The patent submitted by Alibaba Cloud, titled “A Processing Method and Device for User Requests,” won the China Patent Gold Award. This was also the only cloud provider that received a gold award in this session. This patented technology can dynamically and real-time detect changes in computing power demands, especially in response to the intense demand fluctuations of the Internet and AI era. It enables the allocation, scaling, and secure supply of ultra-large-scale intelligent computing power as needed—sustaining capacity during peak times and recovering when demand is low, along with multi-tenant security isolation.
The China Patent Award is jointly selected by the National Intellectual Property Administration and the World Intellectual Property Organization. It is the highest honor in the field of patents in China. Each national selection for invention patents does not exceed 30 awards, serving as a national-level authoritative certification for a company’s innovation capabilities and the value of patent industrialization. Currently, Alibaba Cloud has over ten thousand patents worldwide, covering more than ten countries and regions, focusing on core technical areas such as cloud computing, artificial intelligence, large models, and chips.
Breaking through the three major challenges in large-scale intelligent computing scheduling in the AI era
In the AI era, ultra-large-scale intelligent computing clusters face intense demand fluctuations: sudden traffic, training peaks, and reasoning surges overlap, continuously testing the limits of computing flexibility. Between peaks and troughs, computing demands can vary by a hundred times—if configured for peak levels, a large amount of resources remain idle under normal conditions; if configured for daily levels, insufficient capacity and unstable service risks arise when the peak arrives. Therefore, the flexibility of heterogeneous computing power, the instantaneous response to dynamic loads, and secure isolation in a multi-tenant environment become the three core challenges for ultra-large-scale computing in the AI era.
The core invention of this Alibaba Cloud patent is a completely new way of handling user requests. Traditional approaches rely on pre-allocated resources and passive waiting for requests, resulting in long-term misalignment between resource allocation and actual load. The method proposed by this patent can dynamically and real-time sense changes in demand, intelligently match requests with underlying heterogeneous computing power, and ensure that each request receives timely and appropriate computing power responses.
Specifically, the system can real-time detect fluctuations in business load between user requests and massive computing power, and dynamically allocate computing power to where it is most needed. During peak business times, it automatically expands capacity; during low periods, it automatically reduces it. This allows computing power to be supplied on demand, like “water and electricity,” and used as needed, ensuring that every resource is utilized effectively. The strategy of elastic preheating and improved scheduling efficiency also enable application startup in seconds, and millisecond-level inference responses for large models become a reality. At the same time, the data of various applications is processed and stored in independent operating environments, isolated from each other and without interference. While improving resource utilization efficiency, it also establishes an end-to-end protection barrier for data security.
Elastic computing power serves various industries in the AI era more efficiently
Based on technological accumulations such as elastic computing power scheduling, Alibaba Cloud has continuously developed a range of products covering general computing, model training and deployment, and intelligent agent operation. It has successively launched products like Lingjun Zhenwu super-node instances, Agent Sandbox for intelligent agents, AI Serving Stack for container services, and Fluid for distributed caching, which have been widely applied in various industries in the AI era.
For example, MiniMax uses Ali Cloud Agent Sandbox to launch five thousand sandboxes within 10 seconds, reducing the total cost of ownership by 25%; Kimi of Moon’s Dark Side enables elastic scaling of tens of thousands of sandboxes per minute, reducing startup time by 50%, and provides stable business responses through flexible supply; StepbyStarry combines Agent Sandbox and Ali Cloud ACK, enabling millions of sandboxes to operate smoothly and compressing the model evaluation cycle from weekly to daily. Shenqianwuyou AI Recruitment Assistant uses Agent Sandbox and hierarchical sleep to handle tens of thousands of pulse requests, supporting session data restoration; Kingsoft Office relies on Agent Sandbox and Alibaba Cloud cloud-native infrastructure, achieving millisecond-level access to the sandbox through a preheating pool, combined with tidal scheduling and multi-cluster disaster recovery, to balance experience, cost, and service continuity.
Yi Li, the patent inventor and head of Alibaba Cloud’s container and sandbox research, said, “Cloud computing has become a key productivity factor as important as electricity and water resources. From the mobile internet era to the AI era, Alibaba Cloud has always focused on real-time perception of load changes, unified scheduling of heterogeneous computing power, and secure isolation in multi-tenant environments. This enables computing power to flow as needed and match precisely, reducing the barriers to large model training and inference, as well as the operation of intelligent agents, and promoting the large-scale implementation of intelligent applications.”