📊Freshcollected in 38m

Alibaba Bets Nearly $10 Billion on AI

PostLinkedIn
📊Read original on Bloomberg Technology

💡Alibaba’s $10 billion AI bet reveals the costs and competitive stakes of scaling AI infrastructure.

⚡ 30-Second TL;DR

What Changed

Alibaba increased quarterly AI capital expenditure to almost $10 billion.

Why It Matters

Alibaba’s spending surge signals an aggressive push to build AI capacity, while its profit decline highlights the financial pressure of the infrastructure race. Meta’s large-scale use of Microsoft AI services also illustrates how major platforms may increasingly depend on external compute and models.

What To Do Next

Review your cloud AI budget and compare Microsoft’s AI services with Alibaba’s offerings before committing to a large-scale training or inference deployment.

Who should care:Founders & Product Leaders

Key Points

  • Alibaba increased quarterly AI capital expenditure to almost $10 billion.
  • The spending increase contributed to a sharp decline in Alibaba’s profit.
  • Meta has quietly become one of Microsoft’s largest AI customers.
  • Defense startup Castelion plans to increase production of hypersonic missile systems.

🧠 Deep Insight

Background and context from public sources — not the original article. 25 sources cited.

🔑 Enhanced Key Takeaways

  • Alibaba's net profit plummeted by 75% to $1.6 billion in the latest quarter, a direct consequence of its substantial increase in AI capital expenditure.
  • Despite the profit decline, Alibaba's AI Cloud and Compute Services revenue experienced a 45% year-on-year surge, reaching $7.14 billion, marking its fastest growth in 22 quarters.
  • Meta Platforms is reportedly investing hundreds of millions of dollars annually to access AI models from Microsoft through its Azure cloud platform, consuming trillions of tokens each week.
  • Defense startup Castelion recently secured $1 billion in Series C funding, elevating its valuation to $13 billion, to accelerate the scaled production of its Blackbeard hypersonic missile and broaden its product offerings.
  • Alibaba's chip design subsidiary, T-Head, introduced the Zhenwu M890, its most advanced AI processor, which offers three times the performance of its predecessor, the Zhenwu 810E, and is specifically designed for AI agents.
📊 Competitor Analysis▸ Show
Feature/ProviderAlibaba Cloud AIMicrosoft Azure AIAmazon Web Services (AWS) AIGoogle Cloud AI
Market Share (China AI Cloud)38.1% (largest)---------
Core AI ModelsQwen (Tongyi Qianwen) family (LLMs, multimodal, open-source & proprietary)Llama 2 (via partnership with Meta), OpenAI models (via Foundry), proprietary modelsAmazon Bedrock (various FMs), Amazon SageMaker (ML platform), proprietary modelsGemini, PaLM, Imagen, Vertex AI (ML platform), various FMs
AI Chip DevelopmentZhenwu M890 (inferencing-focused, for internal use)Custom AI chips (e.g., Maia 100)AWS Inferentia, AWS TrainiumTensor Processing Units (TPUs)
Pricing ModelOpen-source models (free for local deployment), API services via DashScope (tiered pricing)Azure Foundry (models from various vendors), pay-as-you-go for servicesPay-as-you-go for services and model usagePay-as-you-go for services and model usage
Strategic Partnerships---Meta (Llama 2 on Azure/Windows), OpenAI (major revenue contributor)------
Global Infrastructure105 availability zones across 32 regionsExtensive global networkExtensive global networkExtensive global network

🛠️ Technical Deep Dive

  • Alibaba's Qwen (Tongyi Qianwen) Models:
    • A family of large language models (LLMs) and multimodal models developed by Alibaba's Qwen Team.
    • Includes text-only, vision-language (Qwen-VL), audio (Qwen2-Audio), coding-focused (Qwen-Coder/Qwen3-Coder), and 'omni' (real-time multimodal) variants.
    • Many instruct models feature a context window of approximately 32,768 tokens, with specialized configurations supporting up to 1 million tokens.
    • Text models typically employ a decoder-only Transformer architecture, incorporating modern components such as Rotary Position Embeddings, SwiGLU, and RMSNorm.
    • The Qwen3.5-397B MoE (Mixture-of-Experts) model boasts a full parameter count of 397 billion, with 17 billion activated parameters, and is noted as the first MoE model released with multi-chip compatibility.
    • The cloud version of Qwen3.8-Max, released in August 2026, utilizes a sparse mixture-of-experts architecture with around 95 billion active parameters per forward pass and supports a context window of up to one million tokens.
  • Alibaba's Zhenwu AI Chips:
    • Alibaba's chip design subsidiary, T-Head, developed the Zhenwu M890, its most powerful AI processor to date.
    • The Zhenwu M890 delivers three times the performance of its predecessor, the Zhenwu 810E, and is equipped with 144 GB of on-chip memory and 800 GB per second of bandwidth.
    • Alibaba is also developing a new AI chip specifically for inferencing tasks, rather than training, to complement Nvidia chips, driven by the push for self-reliance amidst U.S.-China tensions.

🔮 Future ImplicationsAI analysis grounded in cited sources

Alibaba's aggressive AI investment will solidify its position as a dominant AI infrastructure provider in China and potentially globally.
The substantial capital expenditure, rapid revenue growth in AI cloud services, and development of proprietary chips and models indicate a strong commitment to building a comprehensive AI ecosystem.
The increasing reliance of major tech companies like Meta on third-party AI cloud services will drive further competition and innovation in the AI-as-a-Service market.
Even companies with significant internal AI development are opting to rent external AI tools for availability and economic reasons, pushing cloud providers to enhance their offerings and marketplace diversity.
Castelion's focus on low-cost, mass-producible hypersonic weapons could significantly alter global defense strategies and procurement.
By making hypersonic technology more affordable and scalable, Castelion aims to enable fielding in meaningful quantities, potentially shifting the balance of conventional deterrence.

Timeline

2021-XX
Microsoft's Azure cloud platform began hosting Meta AI.
2023-04
Alibaba launched a beta version of its Qwen (Tongyi Qianwen) large language model.
2023-07-18
Meta launched Llama 2, making it available on Microsoft's Azure platform and expanding their AI partnership.
2023-12-01
Alibaba Cloud unveiled its open-source 72 billion-parameter version of Tongyi Qianwen (Qwen-72B).
2025-08-29
Alibaba announced a three-year investment of over 380 billion yuan (approximately $53 billion) in AI and cloud computing infrastructure.
2026-05-20
Alibaba unveiled its Zhenwu M890 AI processor and a multi-year silicon roadmap.
📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

AI-curated news aggregator. All content rights belong to original publishers.
Original source: Bloomberg Technology

This is a summary, not the original. Read the source, or get the weekly briefing.

Weekly AI briefing

One email a week. Unsubscribe anytime.