Alibaba CEO Takes Direct Control of AI Unit as China Accelerates Domestic Technology Push
Eddie Wu consolidates AI operations under new Token Hub division amid chip sanctions and intensifying competition from ByteDance, Baidu.
Alibaba CEO Eddie Wu has assumed direct leadership of a newly formed AI business group, consolidating the company’s sprawling artificial intelligence operations under a single division as China’s tech giants race to commercialize generative AI under US export restrictions.
The move positions Alibaba to capture enterprise AI opportunities while signaling C-suite commitment to AI-as-core-strategy. Bloomberg reported the reorganization follows the early March departure of Qwen AI division head Lin Junyang, the third senior Qwen executive to exit this year.
Token Hub: AI Under One Roof
The Alibaba Token Hub (ATH) consolidates Tongyi Laboratory—developer of the Qwen model family—with the company’s Model-as-a-Service business, Qwen consumer app division, Wukong enterprise workflow unit, and AI Innovation team. The structure places Wu in operational command of AI development, product deployment, and commercialization, replacing a fragmented approach across multiple business units.
“I will lead ATH directly, with a mandate to drive strategic coordination across our AI businesses, embed AI deeply into how we work, and preserve the agility that lets us move fast,” Wu wrote in a memo to staff reviewed by Bloomberg.
The division’s name references tokens—the fundamental units of computation in large language models. ATH’s stated mission is to “create, deliver, and apply tokens,” with a focus on AI-native enterprise platforms. The naming convention underscores Alibaba’s intent to monetize AI workloads, which generate higher revenue per customer than traditional cloud services.
“ATH is built around a single organising mission: create tokens, deliver tokens and apply tokens.”
— Eddie Wu, Alibaba CEO
Export Controls Shape Domestic AI Strategy
Alibaba’s reorganization occurs against a backdrop of tightening US semiconductor export controls. Since January 2026, the Bureau of Industry and Security shifted licensing policy for advanced AI chips from “presumption of denial” to “case-by-case review” for exports to China, but volumes remain capped and subject to end-use certification requirements.
Chinese firms have responded by accelerating domestic chip development and architectural efficiency. Alibaba disclosed in Q1 FY2026 earnings that it deployed $5.4bn in AI and cloud infrastructure spending, part of a three-year $53bn commitment. CEO Wu noted that “AI chip supply chain restrictions could create fluctuations in capex spending,” though the company is working with multiple partners—implicitly domestic suppliers like Huawei’s Ascend line—to mitigate risk.
China now hosts more than 700 registered generative AI services, per industry data. Alibaba competes with Baidu’s Ernie, ByteDance’s Doubao, Tencent’s Hunyuan, and startups including DeepSeek and Zhipu AI. Token pricing has collapsed amid fierce competition: Chinese models cost 10 to 20 times less than US equivalents, squeezing margins and forcing revenue models toward enterprise cloud integration rather than consumer subscriptions.
| Company | Flagship Model | Key Advantage |
|---|---|---|
| Alibaba | Qwen 3 | Cloud/enterprise integration, 1bn+ downloads |
| ByteDance | Doubao | Super-app distribution (Douyin/TikTok) |
| Baidu | Ernie 4.5 | Search integration, 200m users |
| Tencent | Hunyuan | WeChat/QQ embedding, social data |
| DeepSeek | DeepSeek-R1 | Cost efficiency, open-source strategy |
Cloud Revenue Accelerates, E-Commerce Stalls
Alibaba’s Cloud Intelligence Group posted 34% year-over-year revenue growth in Q2 FY2026, reaching $5.6bn, according to Data Center Dynamics. AI-related products sustained triple-digit growth for the ninth consecutive quarter, now representing more than 20% of external cloud revenue. By contrast, overall group revenue grew just 2% to $34.98bn as core e-commerce faced deflationary pressure and competition from Pinduoduo and Douyin.
Wu told analysts the company is rationing GPU capacity, prioritizing customers who use the full Alibaba Cloud stack—storage, big data, and AI services—over those renting bare GPUs for simple inference tasks. “Demand is accelerating,” he said, adding that both legacy and new-generation GPUs are running at full capacity. Supply constraints span fabs, DRAM vendors, and CPU manufacturers, creating global undersupply across the AI server value chain.
The cloud unit’s momentum reflects enterprise AI adoption patterns: workloads once confined to internet platforms and financial services are now spreading to manufacturing, logistics, and agriculture. Alibaba has partnered with SAP to deploy Qwen models for enterprise AI transformation services globally, positioning the model family as infrastructure for business process automation.
Alibaba open-sourced more than 400 Qwen models since 2023, accumulating over 1 billion downloads. The open-source strategy mirrors Beijing’s broader AI policy: distribute foundational models widely to build ecosystem lock-in and data feedback loops, then monetize through cloud services and enterprise licensing rather than consumer subscriptions.
Leadership Churn and Competitive Pressure
The Token Hub formation follows internal turbulence in Alibaba’s AI division. Lin Junyang, head of the Qwen AI model division, resigned in early March 2026, joining post-training lead Yu Bowen and research scientist Hui Binyuan, who exited earlier in the year. The departures raised questions about internal alignment and retention as competition for AI talent intensifies across China’s tech sector.
Despite leadership churn, Qwen usage surged. The consumer AI app reached approximately 203 million monthly active users in February 2026, up from 31 million in January, citing data from tracker AICPB.com reported by Digital Commerce 360. Growth reflects aggressive “red packet” promotions during Lunar New Year, though the app still trails ByteDance’s Doubao in user engagement.
Alibaba is due to report Q3 FY2026 earnings on March 20. Analysts expect continued cloud acceleration but margin pressure from e-commerce subsidies deployed to counter Pinduoduo and Meituan. The company has executed $33bn in share buybacks over the past 18 months as management pivots from premature IPO plans for Cainiao and Cloud units toward capital return and operational efficiency.
- Revenue mix shift: AI/cloud now drives growth as e-commerce matures; cloud margin expansion depends on enterprise AI adoption rates offsetting infrastructure capex.
- Ecosystem competition: ByteDance’s distribution advantage (Douyin, TikTok) and Baidu’s search integration challenge Alibaba’s cloud-first monetization model.
- Chip constraints bind: Domestic accelerator scaling (Huawei Ascend, Cambricon) determines training capacity; efficiency gains (MoE architectures, GPU pooling) buy time but don’t eliminate gap.
- Beijing alignment critical: Government accounts for 39% of China’s AI investment; regulatory compliance and state contracts increasingly shape competitive dynamics.
What to Watch
March 20 earnings will clarify whether AI revenue growth translates to margin expansion or remains a capex-intensive bet. Track Qwen monthly active user retention post-promotion cycle—sustainable engagement separates infrastructure plays from user acquisition burns. Monitor Huawei Ascend chip deployment pace; if Alibaba can train next-generation models on domestic silicon at scale, export controls lose leverage. Finally, watch for ATH product launches: enterprise AI agents that execute tasks across Alibaba’s ecosystem (Taobao search-to-checkout, DingTalk workflow automation) would demonstrate commercial traction beyond model benchmarks. The gap between releasing competitive models and monetizing them profitably remains the defining challenge for China’s AI giants in 2026.