At Apsara 2025, Alibaba Cloud unveiled its global AI roadmap, announcing plans to increase investment beyond its previously disclosed three-year RMB 380 billion (US$53 billion) AI infrastructure program, alongside a global expansion strategy, its most powerful AI models to date, and upgrades across its cloud portfolio.

Eddie Wu, Chairman and CEO of Alibaba Cloud Intelligence, called the move a defining moment: “Alibaba Cloud is strategically positioned as a full-stack AI service provider, dedicated to delivering robust computing with maximized efficiency for training and deploying large AI models on the cloud. To underscore our long-term commitment to advancing AI, we will progress with our RMB 380 billion investment plan in AI and cloud infrastructure over the next three years.”

“We remain committed to open-sourcing Qwen and shaping it into the ‘operating system of the AI era,’ empowering developers around the world to build transformative AI applications.”

Here’s why these announcements matter — and how they could reshape the industry landscape.

Qwen3-Max: A trillion-parameter flagship model

During Alibaba Cloud’s annual flagship technology conference, Alibaba Cloud introduced a series of new large language models from Qwen3 family, headlined by Qwen3-Max and previewed its latest visual-generation models.

At the center of the announcements was Qwen3-Max, Alibaba’s largest LLM model and the most powerful LLMs to date, is a trillion-parameter model designed for businesses and developers. It achieves impressive performance across a wide range of benchmarks especially in code generation and agentic capabilities.

Complementing this flagship model is Qwen3-Omni, a multilingual omni-model capable of processing text, images, audio, and video inputs, while delivering real-time, streaming response in text and speech. By achieving real-time multimodal interaction, with ultra-low latency, it is an ideal solution for intuitive, hands-free interaction in intelligent cockpits, smart glasses and mobile phones.

Alibaba also broadened its portfolio with several new domain-specific models, including Qwen3-VL, a vision-language model optimal for robotics and navigation; an upgraded Qwen3-Coder for faster and more accurate code generation; Qwen3-Image-Edit, which now supports editing multiple images with improved visual consistency; and Wan2.5 Preview, the latest visual generation model designed to elevate multimedia content creation.

Enterprises can access and deploy these models through Model Studio, Alibaba’s AI development platform. The newly-upgraded Model Studio features a high-code agent development framework Model Studio-ADK (agent development kits) to help enterprises enhance agent development efficiency. It also introduces a range of enterprise-grade features, including seamless connectivity via Model Context Protocol (MCP), RAG multi-modal fusion, dynamic inference scheduling, and sandbox service, to address key enterprise challenges such as multi-source data processing, resource constraints, and cross-environment deployment.

Upgrading the cloud for AI at scale

To support trillion-parameter models at scale, Alibaba Cloud has undertaken a sweeping overhaul of its infrastructure, with a focus on speed, scalability, and security.

At the core of these upgrades is the new HPN8.0 network, which delivers up to 800 Gbps throughput—twice the capacity of the previous generation, ensuring faster, more reliable data transfer for AI workloads. Storage has also been reimagined through Vector Bucket technology, which integrates raw and vector data to enable real-time search and retrieval-augmented generation (RAG) pipelines.

Database performance is being pushed further with PolarDB database powered by Compute Express Link (CXL) technology, delivering 72% lower latency and sixteen times greater scalability; The upgraded Alibaba Cloud Container Compute Service (ACS) is capable of scaling up to 15,000 pods per minute, providing the elasticity required for enterprise-grade deployments.

Security has also been strengthened through enhancements to the Cloud Threat Detection and Response (CTDR) solution, which now leverages five Qwen-powered agents to raise boost investigation accuracy from 59% to 74% while and automating 70% of incident responses.

Looking beyond its core infrastructure, Alibaba Cloud is also expanding its global footprint. The company announced plans to open first data centers in Brazil, France, and the Netherlands, along with upcoming facilities in Mexico, Japan, South Korea, Malaysia and Dubai in the coming year.

Two newly added service centers in Germany and Indonesia will further extend its reach, supporting international customers with round-the-clock, multi-language services.

Growing base of real-world users

Alibaba Cloud isn’t starting from scratch. Earlier versions of Qwen and Wan are already in use across a diverse range of sectors, proving Alibaba Cloud ’s stack can handle real-world enterprise workloads.

To date, Alibaba has reported over 600 million model downloads of its Qwen and Wan series, approximately 170,000 derivative models, and more than 800,000 AI agents built on Model Studio, as well as over one million corporates and individuals using Qwen through Model Studio, forming a solid base for Qwen3’s rollout.

The technology is already making an impact in key sectors. In healthcare, AstraZeneca China uses a Qwen-based model to detect and summarize adverse drug events with 95% accuracy, delivering a threefold  efficiency gain.

In financial services, a Tokyo-based AI startup, FLUX has developed a 32billionparameter Japanese language model designed specifically for the industry, giving professionals more reliable, context-aware responses to industry-specific queries.

In digital marketing, GladCube has partnered with Alibaba Cloud to integrate Alibaba’s visual generation model, Wan, into its vertical video transformation platform “Dra Vis”, enabling Japanese customers to produce high-quality video, content quickly and at scale, enhancing both efficiency and creativity.

Laying the groundwork for the future

Alibaba Cloud is now turning its attention to Physical AI systems where AI not only processes data but also takes action in the physical world, from robots, and autonomous vehicles, to industrial machinery.

A key step in this direction is the integration of the full suite of the NVIDIA Physical AI software stack, into Alibaba Cloud’s Platform for AI (PAI), marking a milestone collaboration in the Physical AI domain. The initiative provides developers with a comprehensive, cloud-native platform to accelerate advancements in Physical AI solutions.

Looking ahead, Alibaba Cloud envisions a future where large AI models are deeply embedded across devices, functioning much like operating systems with persistent memory, seamless coordination between cloud and edge, and the capacity to continuously evolve.

With its foundation models, upgraded infrastructure, and expanding global network, Alibaba Cloud has assembled the full-stack building blocks that allow organizations around the world to build, deploy, and govern AI on their own terms.