Alibaba Cloud Named Container Management Leader for Fourth Consecutive Year

In the companion Critical Capabilities for Container Management report, Alibaba Cloud ranked in the top two in four of six scenarios, the report said. Gartner said the container management market's focus is shifting from adopting Kubernetes to scaling production. As enterprises prioritize AI and complex application architectures, they are focusing on how to use a unified control plane to manage diverse environments ranging from on-premises legacy systems to cloud-native and AI-driven workloads. Container platforms are expected to provide governance, security, observability and developer self-service as well as orchestration, reducing complexity in increasingly heterogeneous IT environments. AI is also changing the task boundaries of container platforms: model training requires efficient scheduling of heterogeneous computing power, inference requires a dynamic balance among latency, throughput and cost, and agents require execution environments with secure isolation, elastic scaling and state retention. The value of container platforms is no longer just running applications but enabling AI businesses to operate efficiently and stably at scale in complex environments.

Alibaba Cloud has upgraded its Container Service for Kubernetes (ACK) and Agent Sandbox service around model training, model inference and agent applications, moving AI workloads from development and testing to production. ACK can support up to 100,000 nodes and 200,000 online sandboxes in a single cluster, while Container Registry (ACR) supports concurrent downloads from 5,000 nodes and concurrent uploads from 1,000 clients. The service fully supports Alibaba Cloud's Lingjun Zhenwu M890 super node and uses multi-level topology-aware scheduling, automatic fault detection and self-healing, and AI Profiling diagnostics to improve computing utilization. In real business scenarios, topology-aware optimization increased inference throughput by more than 40%, the report said.

For model inference, ACK's AI Serving Stack coordinates request routing, load management, distributed inference, autoscaling, and model and data caching. In different trial scenarios, it can save 50% of computing resources and increase inference throughput by 200%, according to the article. Through direct weight transfer, operator compilation reuse and parallel engine initialization, switching time for the Qwen3 235B model was shortened from 13 minutes to 3 seconds.

Alibaba Cloud's recently released Agent Sandbox provides an out-of-the-box, elastically scalable and securely isolated execution environment. Through coordinated optimization of CIPU and virtualization technologies, each agent can obtain MicroVM-level isolation. A single region can host one million sandboxes and create 100,000 sandboxes per minute, with creation and cold-start P99 latency below 180 milliseconds after template preheating.

The services have been used in production. Xiaohongshu used mixed deployment of online and offline businesses to reuse computing power and raised CPU utilization by 45%. Moonshot AI used ACK and Lingjun clusters to support large-model training and inference, keeping GPU idle rate below 0.5%. The official website of the Milan Cortina 2026 Winter Olympics used ACK Pro to build a multi-region active-active architecture and provided millisecond-level responses to global audiences during the nearly month-long games, supporting real-time delivery of schedules, dynamic information and Olympic stories.

At the agent runtime layer, MiniMax launched 5,000 sandboxes within 10 seconds and reduced total cost of ownership by 25%, according to the article. 51job handled burst traffic of tens of thousands of requests per minute, shortened startup time by 60%, and met rigid delivery requirements while scaling elastically. An automotive customer migrated in one day through an E2B Adapter, with sandbox startup taking about 200 milliseconds. The cases cover mixed-deployment cost optimization, extreme GPU resource utilization and global high-availability architecture, Alibaba Cloud said.