Alibaba

Updated June 2026

Alibaba is the Chinese technology group behind the Qwen models and Alibaba Cloud. Its Tongyi Laboratory has shipped 400+ open-weight Qwen models that together top 1 billion downloads — the world's most-downloaded open model family — while keeping its flagship Qwen3.7-Max proprietary and API-only. In FY2026 its cloud business grew 38% as AI revenue posted an eleventh straight triple-digit quarter.

Website →API →Pricing →

Alibaba is the Chinese technology group behind Qwen, the most-downloaded family of open-weight AI models in the world. Its AI work runs through Alibaba Cloud’s Tongyi Laboratory, which has released more than 400 open Qwen models since 2023 — together passing 1 billion downloads and 200,000 derivative models on Hugging Face, overtaking Meta’s Llama as the default open foundation. Alibaba pairs that open ecosystem with a deliberately closed frontier: its flagship Qwen3.7-Max is proprietary and API-only on Alibaba Cloud. The strategy is two-tier — commoditise the middle with Apache 2.0 weights, monetise the top through the cloud.

AI is now Alibaba’s growth story. In the quarter ended 31 March 2026, Alibaba Cloud revenue rose 38% year on year to about $6.0 billion, with AI-related product revenue posting its eleventh consecutive quarter of triple-digit growth. The company is spending to keep up: a three-year, 380-billion-yuan (about $53 billion) AI-and-cloud infrastructure plan that management now expects to overshoot. This guide covers Alibaba as an AI provider — the Qwen models, the open-versus-closed strategy, the money, the leadership, and the risks.

Quick facts

History and founding

Alibaba Group was founded in 1999 in Hangzhou by Jack Ma and 17 co-founders, beginning as a business-to-business marketplace and growing into one of the world’s largest e-commerce, cloud and logistics companies. It listed on the New York Stock Exchange in 2014 in what was then the largest IPO in history, and is dual-listed in Hong Kong.

The AI lineage runs through Alibaba Cloud, founded in 2009 and built on the in-house Apsara distributed-computing platform. Alibaba’s large-language-model effort, Tongyi Qianwen, launched in beta in April 2023, and the English-facing Qwen brand followed. By 2025 Qwen had become the centre of Alibaba’s strategy: a full model family spanning language, vision, audio, image, video and coding, released largely as open weights. Today that research sits in Tongyi Laboratory, formalised in 2026 as a dedicated large-model business unit led by Alibaba Cloud CTO Zhou Jingren.

Business and financials

Alibaba is a profitable public company, not a venture-funded startup, and its AI economics are visible in its filings. For fiscal year 2026 (ended 31 March 2026), group revenue was RMB 1,023,670 million (about $148.4 billion), up 3% year on year (Alibaba).

The growth, however, is concentrated in cloud and AI:

To sustain this, Alibaba committed in February 2025 to spend RMB 380 billion (about $53 billion) over three years on AI and cloud infrastructure — and on its FY2026 results said it now expects to overshoot that figure, citing a roughly tenfold increase in data-centre build-out. The cost is real: group free cash flow swung to an outflow in the quarter, driven by cloud and quick-commerce capital spending. Alibaba’s market capitalisation was around $260 billion in June 2026, with the stock having rallied through 2026 on the strength of the AI-cloud story.

The Qwen ecosystem: open Apache and closed Max

Alibaba’s defining AI move is running the world’s largest open-weight model ecosystem while keeping its frontier flagship closed.

On the open side, Alibaba has released more than 400 Qwen models since 2023 under the permissive Apache 2.0 licence, covering sizes from sub-1B to hundreds of billions of parameters. These crossed 1 billion cumulative downloads and 200,000 derivative models on Hugging Face, overtaking Meta’s Llama as the most-downloaded open family in October 2025; by early 2026 roughly 69% of all new model derivatives on Hugging Face were Qwen-based. In a single month, Qwen downloads exceeded the combined total of the next eight model families. This is the strategic moat: developer mindshare, a vast fine-tuning ecosystem, and pull-through to Alibaba Cloud.

On the closed side sits the Max tier. The current flagship, Qwen3.7-Max (announced 20 May 2026), is proprietary and API-only through Alibaba Cloud Model Studio — the first time Qwen’s flagship shipped closed-weights only. It scores 80.4% on SWE-bench Verified on Alibaba’s own testing and ranks in the global top ten on Artificial Analysis’s independent intelligence index, putting it within striking distance of US frontier models at a fraction of the price.

The tension is deliberate and contested. Alibaba commoditises the mid-tier to win developers and sells the frontier through the cloud — but analysts warn that as revenue pressure mounts, “open” may become a secondary priority, with future flagships locked behind paid APIs. The closed Max tier is the first concrete sign of that shift.

The Qwen model lineup

Beyond the language models, the family includes Qwen-Image (text-to-image with accurate in-image text), Wan (text-to-video and image-to-video), Qwen3-VL (vision) and Qwen3-Omni (real-time voice and audio across 119 languages). Full benchmarks and pricing live on the Qwen model page.

Products and ecosystem

• Qwen app (Qwen Studio) — the free consumer assistant, renamed from Qwen Chat in April 2026, with ~203 million monthly active users by February 2026 and a place among the top three AI apps worldwide.
• Alibaba Cloud Model Studio — the developer platform (formerly DashScope) serving Qwen models via API, billed per token, with endpoints in Singapore, the US, Germany and China.
• Quark — Alibaba’s AI assistant and browser unit (70M+ monthly users), and the brand behind the Quark AI Glasses (from 1,899 yuan / about $277, launched November 2025), powered by Qwen.
• DingTalk — Alibaba’s workplace platform, where Qwen reaches enterprise users; the company reports 90,000+ enterprises using the Qwen family.
• Consumer hardware — in December 2025 Alibaba formed a new consumer business group around Qwen, overseeing the app, Quark and AI devices (glasses, earbuds, rings) with a global rollout planned through 2026.

Leadership

Alibaba reshaped its AI leadership in 2026 to put model development directly under the CEO.

• Eddie Wu (Wu Yongming) — Group CEO; took direct control of AI strategy in 2026 and leads a foundation-model task force.
• Joseph Tsai — Chairman and co-founder.
• Jack Ma — founder; stepped back from operational roles but remains the company’s defining figure.
• Zhou Jingren — Alibaba Cloud CTO; leads Tongyi Laboratory and the Qwen large-model business unit.
• Wu Zeming — Group CTO; co-leads the AI task force.

The reshuffle followed turbulence in the Qwen team: in March 2026, Lin Junyang, head of the Qwen division, departed alongside other key figures shortly after an open-source release, prompting Alibaba to form the task force and centralise control under Wu.

Competition and market position

Alibaba is the clear leader of China’s open-weight AI push and one of the most important AI providers outside the United States. Its peers split into two groups. Against the other Chinese open-weight labs — DeepSeek, Moonshot’s Kimi, MiniMax and Zhipu — Alibaba competes on the breadth and downloads of its open family, where Qwen is comfortably the most-used. Against the US frontier labs — OpenAI, Anthropic and Google — Alibaba trades the last increment of capability on the hardest tasks for openness, multilingual breadth and an order-of-magnitude price advantage.

Its structural edge is being full-stack: Alibaba owns the models, the cloud that serves them, the consumer apps that distribute them, and a fast-growing hardware line — a vertical integration no Western AI lab matches. Its weakness is the mirror image: dependence on constrained chip supply, the trust and regulatory baggage of a China-based provider, and a frontier that still trails the very top of the US labs on the most demanding work.

Controversies

• National Intelligence Law and data access. Like all Chinese AI providers, Alibaba is subject to China’s 2017 National Intelligence Law, which can compel companies to assist state intelligence work. Open weights let users self-host and sidestep this, but the hosted Qwen app and any China-routed API calls carry the concern.
• US national-security scrutiny. Chinese open models led by Qwen and DeepSeek now account for roughly 30% of global AI-model downloads, surpassing US models — a trend a US congressional commission has framed as a threat to American AI leadership. US lawmakers opened inquiries into companies including Anysphere (maker of the Cursor coding tool) and Airbnb after they disclosed building on Chinese open models such as Qwen and Kimi.
• Open-versus-revenue tension. Moving the flagship Qwen3.x-Max tier to closed, API-only access has drawn criticism that Alibaba’s open-source commitment may erode as monetisation pressure grows.
• Qwen team departures. The March 2026 exit of Qwen division head Lin Junyang and other researchers raised questions about continuity, even as Alibaba moved quickly to restructure.
• Content controls. Hosted Qwen, like other China-based models, aligns with Chinese content rules and avoids politically sensitive topics; outputs can reflect training-data and policy constraints, a consideration for neutral or politically sensitive use.
• Chip-supply risk. US export controls on advanced AI chips, tightened and codified in January 2026, constrain Alibaba’s access to the highest-end accelerators and push it toward domestic alternatives — a recurring overhang on its AI roadmap.

Recent developments (2026)

• Qwen3.7-Max (20 May 2026) became the flagship — closed-weight, API-only, 1M-token context — alongside the open Qwen 3.5 and 3.6 releases.
• AI revenue posted its eleventh straight triple-digit-growth quarter as cloud revenue grew 38%, and Alibaba said it would overshoot its 380-billion-yuan AI capex plan.
• Leadership reshuffle put AI model development under CEO Eddie Wu after the Qwen division head departed in March 2026.
• Qwen open downloads passed 1 billion with 200,000+ derivatives, cementing Qwen as the world’s most-used open model family.
• Consumer push continued with the Qwen app above 200 million MAU and Quark AI Glasses expanding internationally from MWC 2026.

Where Alibaba excels

• Open-weight ecosystem. Qwen is the most-downloaded open model family in the world, with the largest fine-tuning and derivative ecosystem — a developer moat no rival matches.
• Price-to-performance. Qwen models deliver near-frontier quality at a fraction of US flagship pricing, on the API and free in the app.
• Multilingual breadth. Coverage of 119+ languages, with particular strength in Chinese, makes Qwen a default for non-English and bilingual work.
• Full-stack integration. Owning the models, cloud, consumer apps and hardware lets Alibaba distribute and monetise AI end to end.
• Cloud momentum. AI is driving Alibaba Cloud’s fastest growth in years, with double-digit acceleration and 11 straight triple-digit AI-revenue quarters.

Where Alibaba falls short

• Frontier gap. Qwen3.7-Max is excellent but still trails the very top US models — Claude Opus 4.8 and GPT-5.5 — on the hardest reasoning and long-horizon agentic tasks.
• Trust and regulation. A China-headquartered provider carries data-access, content-control and geopolitical concerns that limit adoption in Western enterprises and governments.
• Chip dependence. Export controls cap access to top-end accelerators, a structural constraint on training the next frontier.
• Open-source credibility risk. Closing the Max tier and the loss of key Qwen researchers raise questions about how open Alibaba’s strategy will stay.
• Thin consumer privacy posture. The free Qwen app’s consumer terms are less transparent than Western rivals’, pushing privacy-sensitive users toward self-hosting.

Developer resources

Alibaba serves Qwen through Alibaba Cloud Model Studio (docs), an OpenAI- and Anthropic-API-compatible platform with per-token pricing and endpoints in Singapore, the US, Germany and China. Open-weight Qwen models are published on Hugging Face and GitHub under Apache 2.0 and can be run locally with tools such as Ollama, llama.cpp and LM Studio, or deployed on third-party clouds. Consumer access is via the free Qwen app at chat.qwen.ai. Model details and benchmarks live on the Qwen model page.

Frequently asked questions

Who owns Qwen?

Qwen is developed and owned by Alibaba, specifically the Tongyi Laboratory within Alibaba Cloud, part of Alibaba Group Holding Limited. Alibaba is a public company listed on the New York Stock Exchange (BABA) and the Hong Kong Stock Exchange (9988), headquartered in Hangzhou, China.

Is Qwen open source?

Partly. Alibaba has released 400+ Qwen models under the Apache 2.0 licence, including Qwen 3.5 and Qwen 3.6, which can be downloaded and self-hosted — making Qwen the world’s most-downloaded open model family. The flagship Qwen3.7-Max is closed-weight and API-only, so the top tier is proprietary while the mid-tier is open.

How big is Alibaba’s AI business?

In the quarter ended 31 March 2026, Alibaba Cloud revenue grew 38% year on year to about $6.0 billion, with AI-related product revenue at roughly 30% of that and posting its eleventh consecutive quarter of triple-digit growth. Alibaba has committed over $53 billion to AI and cloud infrastructure across three years and expects to spend more.

What is Alibaba’s flagship AI model?

The current flagship is Qwen3.7-Max, announced on 20 May 2026 — a closed-weight, API-only model with a 1M-token context window that scores about 80.4% on SWE-bench Verified (Alibaba-reported) and ranks in the global top ten on the independent Artificial Analysis intelligence index.

Is Qwen safe to use for sensitive data?

Use caution with the hosted app and China-routed API. Alibaba is subject to China’s National Intelligence Law, and hosted Qwen aligns with Chinese content rules. For sensitive or regulated work, the safer route is to self-host Qwen’s Apache 2.0 open weights so nothing leaves your infrastructure, or use a Western-hosted provider such as Anthropic or Mistral.

How does Alibaba compare to DeepSeek?

Both are Chinese open-weight leaders. Alibaba’s Qwen has the broader model family and by far the most downloads and derivatives, with full-stack cloud and consumer distribution. DeepSeek is smaller and more focused, with an MIT licence and aggressive pricing. Qwen wins on breadth and ecosystem; DeepSeek competes on licence permissiveness and cost.

Who runs Alibaba’s AI now?

AI model development sits under Group CEO Eddie Wu, who took direct control in 2026 and leads a foundation-model task force with Group CTO Wu Zeming and Alibaba Cloud CTO Zhou Jingren, who heads Tongyi Laboratory. The reshuffle followed the departure of Qwen division head Lin Junyang in March 2026.

Models

Apps

• Qwen