Alibaba is going all-in on AI infrastructure. In its Q4 FY2026 earnings call on May 13, CEO Eddie Wu confirmed that the company will invest far more than 380 billion yuan (~$52.7 billion) in AI infrastructure over the next five years—a staggering sum that underscores just how critical the company believes this moment to be.
The announcement came as Alibaba reported that AI-related revenue reached 8.97 billion yuan in the quarter, marking the 11th consecutive quarter of triple-digit year-over-year growth. AI now accounts for over 30% of external cloud revenue for the first time.
The Token Economy
Behind the spending is a simple, explosive trend: China’s AI token consumption is growing exponentially. According to the National Data Bureau:
- Early 2024: 100 billion daily tokens
- End of 2025: 10 trillion daily tokens
- March 2026: 14 trillion daily tokens
That’s a 1,000x increase in roughly two years. OpenRouter data shows that for the week of March 16-22, 2026, China alone consumed 7.359 trillion tokens—36% of global AI model usage and the third consecutive week that China surpassed the United States in weekly token volume.
ByteDance’s Doubao large model alone processes over 12 trillion tokens daily, a 1,000x increase since its May 2024 launch. The company has reportedly been forced to reduce free-tier quotas and even cancel some services entirely because compute capacity cannot keep pace with demand.
“Not a Single Card Is Idle”
Wu was blunt about the supply crunch: “Currently, not a single card in Alibaba’s servers is idle. Customer demand cannot be fully met, and there is still a long queue. Model prices are expected to continue rising.”
ByteDance faces the same pressure. The South China Morning Post reported on May 9 that ByteDance’s 2026 AI infrastructure spending will exceed 200 billion yuan, at least 25% above its year-end plan.
Alibaba’s Three-Pillar Strategy
The 380 billion yuan investment targets three strategic priorities:
1. Close the compute gap
Building data centers, AI chips, high-performance networking, and intelligent computing clusters to expand supply and support accelerating AI cloud revenue growth.
2. Hardware self-sufficiency
Alibaba’s T-Head (平头哥) semiconductor unit has already shipped 470,000 self-developed GPU chips as of February 2026. While deployment within Alibaba Cloud remains limited due to production constraints, the long-term goal is full-stack self-reliance spanning GPU, CPU, storage, and networking silicon.
3. Model competitiveness
Adequate compute is the prerequisite for training next-generation models. Alibaba’s Qwen family—recently expanded with Qwen3.6-Plus, the open-world model HappyOyster, and the multimodal HappyHorse—requires massive training clusters. Wu confirmed that Alibaba is investing across world models, speech models, video models, and coding models.
The Competitive Challenge
Despite the spending, Alibaba faces a formidable rival in ByteDance. The TikTok parent has built the largest AI user base in China, with Doubao’s daily token consumption putting it in the same league as OpenAI and Google globally. ByteDance’s advantage in consumer distribution and application innovation is not easily replicated with capital alone.
Alibaba’s strength lies in the enterprise market. Its Bailian platform has seen customer numbers grow 8x year-over-year, and API call volume is increasingly driven by AI coding applications—complex tasks that command premium pricing. Wu predicted that Model-as-a-Service (MaaS) revenue will significantly improve Alibaba Cloud’s gross margins.
The Long Game
The 380 billion yuan figure is not just an investment plan—it’s a strategic declaration. Alibaba is effectively betting that AI infrastructure will become the defining layer of the next decade’s technology stack, and that owning that stack is worth any near-term cost.
Whether that bet pays off depends on whether Alibaba can convert compute capacity into model quality, model quality into user adoption, and user adoption into sustainable revenue. The infrastructure is merely the table stakes. The real game is everything that runs on top of it.
Sources: 电子发烧友网, Alibaba FY2026 Q4 earnings, South China Morning Post, National Data Bureau