Alibaba Cloud
Last reviewed
Sources
57 citations
Review status
Source-backed
Revision
v5 · 3,882 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
Sources
57 citations
Review status
Source-backed
Revision
v5 · 3,882 words
Add missing citations, update stale details, or suggest a clearer explanation.
| Alibaba Cloud | |
|---|---|
| 阿里云 (Aliyun) | |
| * | |
| File:Alibaba-Cloud-Z-Space.jpg | |
| Alibaba Cloud headquarters in Hangzhou, China | |
| Type | Subsidiary |
| Industry | Artificial intelligence; Cloud computing |
| Founded | September 2009 |
| Headquarters | Hangzhou, China (mainland headquarters); Singapore (international headquarters) |
| Key people | Eddie Wu (CEO, Alibaba Group); Jeff Zhang (CTO, Head of DAMO Academy); Zhou Jingren (CTO, Alibaba Cloud); Joe Tsai (Chairman, Alibaba Group) |
| Parent | Alibaba Group |
| Products | Qwen (Tongyi Qianwen) large language model family; Model Studio (GenAI platform); Platform for AI (PAI); ModelScope (open-source model hub); Elastic Algorithm Service (PAI-EAS); CDN, compute, storage, databases |
| Revenue | RMB 41.6 billion / US$6.0 billion (quarter ended March 2026)[7] |
| Employees | 10,001+ (2025) |
| Website | alibabacloud.com |
Alibaba Cloud (Chinese: 阿里云, also known as Aliyun) is the cloud computing and artificial intelligence arm of Alibaba Group, and the developer of the Qwen (Tongyi Qianwen) family of large language models. It is the largest cloud provider in China and the Asia-Pacific region by revenue and the fourth-largest worldwide, holding a 32.8% share of China's infrastructure-as-a-service (IaaS) market and 7.7% globally in 2025 according to Gartner.[5][53] Alibaba Cloud's full-stack AI offering pairs Qwen with the Machine Learning Platform for AI (PAI), the Model Studio foundation-model API service, and the ModelScope open-source model community.[1][2][3][4]
Alibaba Cloud provides public cloud services and AI platforms used within Alibaba's ecosystem and by external customers. Its AI stack spans foundation model development (the Qwen family), model hosting and inference (PAI-EAS), application-ready APIs (Model Studio), and an open community for model weights, datasets, and demos (ModelScope).[2][3][1] By January 2026, the open-weight Qwen models had surpassed 700 million cumulative downloads on Hugging Face and spawned more than 100,000 derivative models, overtaking Meta's Llama as the most-downloaded open-model family in the world.[54][55]
AI-related products have driven Alibaba Cloud's return to rapid growth. For the quarter ended March 31, 2026, Cloud Intelligence Group revenue reached RMB 41.6 billion (US$6.0 billion), a 38% year-over-year increase, while AI-related product revenue of RMB 8.97 billion delivered its eleventh consecutive quarter of triple-digit year-over-year growth and made up roughly 30% of external-customer revenue.[7][8]
In February 2025, Alibaba Group announced it would invest more than RMB 380 billion (about US$53 billion) over three years in cloud and AI hardware infrastructure, which the company called its largest computing investment ever and more than its total cloud and AI spending over the prior decade.[9] CEO Eddie Wu (Wu Yongming) said Alibaba "will spare no effort to accelerate the construction of cloud and AI hardware infrastructure to promote the development of the whole industry."[9] In September 2023, Alibaba Cloud gained Chinese regulatory approval to open its Tongyi Qianwen models to the public, marking a milestone in commercial generative AI availability in China.[10]
2009: Alibaba Cloud is founded in September to provide cloud infrastructure for Alibaba's platforms and external customers.[6]
2010: Introduces proprietary cloud computing engine Apsara, which becomes the foundation of Alibaba Cloud's services; supports Taobao's first Singles' Day event, managing 2.4 billion page views in 24 hours.[11]
2012: Becomes first Chinese cloud service provider to pass ISO 27001:2005 certification.[6]
2013: Merges with HiChina, acquiring the www.net.cn domain services business.[6]
2014: Hong Kong data center goes online; successfully defends against 14-hour DDoS attack peaking at 453.8 Gbit/s.[6]
2015: Alibaba Group invests US$1 billion in Alibaba Cloud; opens Singapore data center and designates it as overseas headquarters.[6]
2017: Becomes official cloud services partner of the Olympics; establishes DAMO Academy (Discovery, Adventure, Momentum and Outlook) with $15 billion commitment for AI and technology research; placed in Gartner's Visionaries quadrant for Cloud IaaS.[12][6]
2019: T-Head (Alibaba's semiconductor unit) unveils Hanguang 800, a data-center AI inference ASIC with 820 TOPS performance used to accelerate machine-learning workloads on Alibaba Cloud.[13][14]
2021: Opens Philippines data center; unveils in-house ARM-based Yitian 710 chip for data centers.[6]
2022: Launch of ModelScope, an open-source "Model-as-a-Service" (MaaS) platform hosting hundreds (later thousands) of models and datasets for developers and researchers; announces US$1 billion pledge to upgrade global partner ecosystem.[4][6]
2023: Tongyi Qianwen (Qwen) debuts in April and later receives regulatory clearance to go public in September; Alibaba Cloud begins releasing multiple model sizes (including open-weight variants).[10][1][15]
2024-2025: Expansion of the Qwen ecosystem (Qwen2/2.5 and QwQ reasoning models; later Qwen3 series), multimodal variants (Qwen-VL), and Model Studio APIs.[16][17]
2025: February announcement of the RMB 380 billion (US$53 billion) three-year investment; September release of Qwen3-Max with over 1 trillion parameters; at the Apsara Conference in September, Eddie Wu announced Alibaba would increase capital expenditure beyond the original RMB 380 billion commitment, sending the company's shares to a four-year high.[9][18][19]
The DAMO Academy (Academy for Discovery, Adventure, Momentum and Outlook) is Alibaba's global research initiative established in October 2017 with an initial commitment of $15 billion over three years.[12] The academy operates seven research labs globally in Beijing, Hangzhou, San Mateo, Bellevue, Moscow, Tel Aviv, and Singapore.[20]
Data intelligence
Internet of Things (IoT)
Financial technology (Fintech)
Quantum computing
Human-machine interaction
Network security
Visual computing
DAMO Academy has developed breakthrough AI-powered healthcare tools:
PANDA (Pancreatic Cancer Detection with Artificial Intelligence): Deep learning model detecting pancreatic lesions using non-contrast CT scans, achieving 92.9% sensitivity and 99.9% specificity[21]
GRAPE (Gastric Cancer Risk Assessment Procedure): Framework for analyzing 3D CT scans to detect gastric cancers
Multi-cancer screening supporting seven cancer types from a single body CT scan
In 2024, Alibaba became the only Chinese company ranked in Fortune's top 10 "Change the World" list due to its AI-assisted cancer detection technology.[22]
DAMO Academy leverages AI for renewable energy forecasting, improving the reliability of solar and wind power predictions for grid operators across China.[23]
PAI is Alibaba Cloud's comprehensive end-to-end machine-learning platform providing data processing, training, and deployment. It supports large-scale training (tens of billions of features / hundreds of billions of samples) and integrates Elastic Algorithm Service (EAS) for one-click online inference with elastic scaling and monitoring.[2][24][25]
Sub-products include:
PAI-iTAG: Multimodal data labeling for efficient data preparation
PAI-DSW (Data Science Workshop): Interactive programming tool with Jupyter Notebook support, optimized TensorFlow, and open-source frameworks for model building; supports CPU/GPU hybrid scheduling
PAI-Designer: Codeless development tool with drag-and-drop components for classic ML algorithms
PAI-DLC (Deep Learning Containers): Cloud-native deep learning platform for training, compatible with predefined and custom frameworks
PAI-EAS (Elastic Algorithm Service): Model deployment tool for inference, enabling one-click deployment, scaling, and online debugging of complex models
PAI-Blade: Universal optimization for inference performance
The platform features over 140 built-in optimization algorithms and supports popular frameworks including TensorFlow, PyTorch, Megatron, and DeepSpeed.[26] In 2025, PAI-EAS introduced distributed inference capabilities with multi-node architecture, achieving a 92% increase in concurrency and 91% increase in tokens per second with the Qwen2.5-72B model.[27] At the Apsara Conference in September 2025, Alibaba Cloud integrated the full suite of Nvidia's Physical AI software stack into PAI, giving developers a cloud-native environment for humanoid robotics, autonomous driving, and smart-space applications.[45]
Model Studio is a managed GenAI platform that exposes Alibaba's latest foundation models, such as Qwen (text), Qwen-VL (vision-language), and Wan (image/video generation), via REST APIs and SDKs, with key management, usage metering, and enterprise controls.[3][28]
Features include:
Access to Qwen series models including Qwen-Max, Qwen-Plus, and Qwen-Turbo
Supervised fine-tuning (SFT) and low-rank adaptation (LoRA)
Model compression and inference acceleration
One-click RAG with AnalyticDB (supporting 10 billion vectors)
OpenAI-compatible API interfaces
Secure deployment in Virtual Private Cloud (VPC) environments
Alibaba Cloud has also offered serverless options for PAI-EAS and promoted Model Studio for regulated industries via private-cloud deployment options.[29][30]
Launched at Apsara 2022, ModelScope is Alibaba Cloud's open-source hub where developers can browse and run models (including open-weight Qwen variants) and access datasets and demos. The platform provides:
Over 300 ready-to-deploy AI models
150+ state-of-the-art (SOTA) models
Support for computer vision, natural language processing, and audio tasks
Free online testing and deployment capabilities
Integration with Alibaba Cloud computing resources
In 2024 Alibaba expanded ModelScope's English-language access for international users.[4][31][32]
Qwen (Chinese: 通义千问, meaning "universal truth" or "general meaning") is Alibaba Cloud's flagship LLM family. Initial public availability followed Chinese regulatory approval in September 2023, with subsequent open-weight releases of several sizes. The family expanded through 2024-2025 (Qwen2/2.5, reasoning-optimized QwQ, multimodal Qwen-VL, and Qwen3 models with both dense and Mixture-of-Experts variants).[10][16]
The Qwen3 models were pre-trained on roughly 36 trillion tokens spanning 119 languages and dialects, up from 18 trillion tokens and 29 languages for Qwen2.5, and are released under the Apache 2.0 license.[56] By January 2026 the family had been downloaded more than 700 million times on Hugging Face, with cumulative downloads overtaking Meta's Llama by October 2025; weights are also distributed through GitHub and ModelScope.[54][55]
| Release | Model/Variant | Parameters | Context Window | Key Features | Notes |
|---|---|---|---|---|---|
| 2023-08 | Qwen-1.8B / Qwen-7B | 1.8B / 7B | 32,768 tokens | Lightweight (1.8B) suitable for edge; Mid-sized (7B) balancing performance | Initial open-source releases[33] |
| 2023-09 | Tongyi Qianwen (Qwen) | Various | - | Text generation | Opened to public after regulatory approval[10] |
| 2023-11 | Qwen-72B | 72B | 32,768 tokens | High-performance model competitive with global LLMs | First large-scale release[33] |
| 2024-02 | Qwen1.5 Series | 0.5B-72B | 32,768 tokens | Improved performance, multi-language support | Multiple size options[33] |
| 2024-06 | Qwen2 Series | Up to 72B | 131,072 tokens | Next-gen with dense and MoE options; significant improvements in coding, math, reasoning across 27 languages | Extended context window[34] |
| 2024-11 | QwQ (reasoning preview) | Based on Qwen2.5 | - | Reinforcement-learning-enhanced reasoning | Part of Qwen lineage[16] |
| 2025-03/04 | Qwen2.5-Omni / Qwen2.5-VL-32B | Various | - | Omni (audio/image/video I/O); improved VL | Multimodal expansion[35] |
| 2025-04 | Qwen3-235B | 235B (22B active) | - | Mixture of Experts (MoE) architecture | Efficiency focused[33] |
| 2025-09 | Qwen3-Max (flagship) | 1+ trillion | up to ~1M tokens | Largest model to date; emphasis on code and agentic capabilities; closed-source, API-first | Preview Sept 5, official release Sept 23, 2025[18][57] |
| 2025-09 | Qwen3-Next | 80B (3B active) | - | Optimized for efficiency with 10x throughput | Performance optimization[33] |
Qwen3-Max scored 69.6 on SWE-Bench Verified, a benchmark of real-world software-engineering tasks, placing it among the strongest coding models at release, and its Instruct version reached a top-three position on the LMArena Text leaderboard.[57]
| Model Series | Capabilities | Architectures/Features |
|---|---|---|
| Qwen-VL | Vision-language understanding | 2B-72B parameters; multimodal processing[36] |
| Qwen-Audio | Audio processing and understanding | Large language audio model |
| Qwen-Coder | Code generation and programming | Specialized for software development |
| Qwen-Math | Mathematical reasoning | Optimized for complex calculations |
| Wan Series | Text-to-video, image-to-video generation | Realistic visuals, complex movements, physical principles adherence[37] |
By mid-2024 Alibaba said Qwen models had more than 90,000 enterprise clients across sectors such as smartphones and gaming, reflecting rapid uptake in China's generative-AI market.[38] In May 2024 Chinese tech firms, including Alibaba, cut LLM API prices substantially; Alibaba reduced some Qwen prices by up to 97% as competition intensified.[39] By early 2026 the open Qwen lineup, with nearly 400 open-sourced models and more than 100,000 derivatives on Hugging Face, accounted for a large share of new open-model downloads worldwide.[54][55]
Alibaba Cloud's AI technologies power various aspects of the company's e-commerce ecosystem:
AI Search: Advanced search capabilities on Taobao and Tmall platforms
AI Advertising Platform: Intelligent ad targeting and optimization
Recommendation Systems: Personalized product recommendations using deep learning
| Industry | AI Applications | Key Features |
|---|---|---|
| Healthcare | Medical imaging, cancer screening | 7+ cancer types detection from single CT scan[21] |
| Finance | Risk management, fraud detection | Real-time transaction monitoring |
| Education | Smart Class, pronunciation evaluation | Intelligent teaching assistants, answer sheet analysis |
| Transportation | Traffic optimization, mobility services | Electronic toll collection, intelligent parking |
| Manufacturing | Quality control, predictive maintenance | Computer vision for defect detection |
| Retail | Inventory management, customer analytics | Demand forecasting, price optimization |
ET Industrial Brain: Optimizes manufacturing processes by analyzing industrial data to improve efficiency, detect defects, and perform predictive maintenance. Deployed in industries like cement, solar energy, and rubber.[40]
ET City Brain: Uses computer vision and big data analytics to optimize urban management, including traffic flow, emergency response, and public services. First deployed in Hangzhou.[40]
Tongyi Lingma: Intelligent coding assistant integrated into IDEs for code writing, debugging, and optimization
Tongyi Tingwu: AI-powered meeting assistant with real-time transcription, summaries, and speaker distinction
Tongyi Wanxiang (通义万相): Text-to-image AI model generating high-quality images in various styles (sketches, 3D cartoons, watercolor paintings) from textual prompts in Chinese and English[41]
DingTalk: Enterprise collaboration app uses Tongyi Qianwen for summarizing meeting notes, generating reports, and creating business proposals[42]
Tmall Genie: Smart speaker assistant powered by Alibaba's AI for voice commands and smart home control
Alibaba's in-house semiconductor division T-Head announced Hanguang 800 in 2019, a 12 nm AI inference chip with:
Peak performance of 820 TOPS
17 billion transistors
Strong ResNet-50 performance metrics
In 2021, Alibaba unveiled the ARM-based Yitian 710 chip for data centers, further strengthening its AI infrastructure capabilities.[6]
Alibaba has emphasized "AI-driven" growth for its Cloud Intelligence Group:
Q3 FY2025 (quarter ended December 2024): Revenue of US$4.35 billion, 13% year-over-year growth[43]
Q1 FY2026 (quarter ended June 2025): RMB 33.4 billion revenue, 26% year-over-year increase driven by AI demand[7]
Quarter ended March 31, 2026: Cloud Intelligence Group revenue of RMB 41.6 billion (US$6.0 billion), up 38% year-over-year, with external-customer revenue accelerating to 40%[7][8]
AI-related product revenue of RMB 8.97 billion (an annualized run rate of roughly US$5.2 billion), the eleventh consecutive quarter of triple-digit year-over-year growth, making up about 30% of external-customer revenue[8]
Group capital expenditure for fiscal year 2026 reached RMB 126.1 billion, up from RMB 84.3 billion the prior year, with the bulk going to data-center and cloud infrastructure[7]
Largest cloud computing company in China and Asia Pacific by revenue[5][53]
32.8% of China's IaaS market in 2025 (up from 30.1% in 2024) and 22.5% of the Asia-Pacific IaaS market, ranking first regionally, per Gartner[53]
Fourth-largest IaaS provider globally in 2025, with a 7.7% share (up from 7.2%)[53]
Placed in Visionaries quadrant of Gartner's Magic Quadrant for cloud infrastructure (2017)[6]
2016: SK Holdings C&C for Korean-Chinese cloud services[6]
2016: SoftBank joint venture for Japan[6]
2016: Vodafone Germany for European operations[6]
2017: Official cloud services provider for the Olympics[20]
2017: Elastic for Alibaba Cloud Elasticsearch[6]
2017: Malaysia's Fusionex for Southeast Asia cloud solutions[44]
2025: Nvidia for AI products and development tools for physical AI applications[45]
The Alibaba Cloud AI Alliance brings together global AI innovators including:[46]
AI Independent Software Vendors (ISVs)
Channel Resellers
Managed Service Providers (MSPs) and System Integrators (SIs)
Consultants
AstraZeneca: Adverse event summary system using Qwen LLM[47]
OxValue.AI (University of Oxford): AI-driven company valuation services
China Unicom: Deployment of Alibaba's Pingtouge AI accelerators
International Telecommunication Union (ITU): Partnership for AI in health, climate, and science[48]
Launched in 2025 to support global developers leveraging Qwen models, offering:[50]
Free cloud credits
Training workshops
Tech show invitations
Product co-marketing opportunities
Most Qwen models are released under the Apache 2.0 license, although the trillion-parameter flagship Qwen3-Max launched as a closed-source, API-first product.[56][57] Open-source support includes:
Apache 2.0 licensing for most Qwen models
Comprehensive SDKs and APIs
OpenAI-compatible API interfaces
Multi-platform availability: Hugging Face, GitHub, ModelScope
Key improvements introduced in 2025:[33]
Hybrid attention mechanism
Highly sparse mixture-of-experts (MoE) structure
Training-stability-friendly optimizations
Multi-token prediction mechanism for faster inference
Context Length Scaling and Total Parameter Scaling
Interleaved-MRoPE: Full-frequency allocation for video reasoning
DeepStack: Multi-level ViT feature fusion
Text-Timestamp Alignment: Precise event localization in videos
Prefill-decode disaggregation: Performance optimization for large models
Fortune "Change the World" list - 6 appearances (most for any Chinese company)
2024: Top 10 ranking for AI-assisted cancer detection[22]
Stanford AI Index Report 2024: Highlighted research in medical AI
Gartner Magic Quadrant recognition for cloud infrastructure
CEO Eddie Wu stated in 2025: "If AGI is achieved, the AI-relevant industry will very likely become the world's largest industry."[51] On the same earnings call, Wu likened AI to "the electricity of the future" and cloud computing to the "electricity grid," and framed AGI as Alibaba's primary long-term objective.[51] The company positions:
AI as "the electricity of the future"
Cloud computing as the "electricity grid"
Focus on becoming the backbone of the AI-driven economy
More than RMB 380 billion (US$53 billion) in cloud and AI infrastructure over three years (announced February 2025)[9]
Additional, undisclosed capital expenditure beyond the RMB 380 billion commitment, announced at the Apsara Conference in September 2025[19]
Development of "unified global cloud network" for consistent AI services
Expansion in Japan, South Korea, Southeast Asia, Middle East, Europe, and Americas
Planned first data centers in Brazil, France, and the Netherlands, with additional sites in Malaysia, Dubai, Mexico, Japan, and South Korea[52]
Tongyi Qianwen (Qwen)
Model-as-a-Service
Cloud computing
DAMO Academy
Cloud computing in China
Alibaba Group