What’s Driving the Latest AI Industry Shake‑ups? From Alibaba’s Reorg to Global Token Wars

1. Alibaba AI Organization Overhaul

Alibaba Group CEO Wu Yongming announced a major restructuring centered on AI, establishing a Group Technology Committee chaired by himself to coordinate AI strategy. New appointments include Zhou Jingren as Chief AI Architect (also head of the Tongyi Large Model Division) and Fei‑Fei Li as Alibaba Cloud CTO, while Group CTO Wu Zeming focuses on AI inference platforms. Business adjustments: the Tongyi Lab is upgraded to the Tongyi Large Model Division under Zhou, and the Taobao Flash Sale CEO role shifts to Lei Yanqiu. The Qwen 3.6 Plus model recently topped OpenRouter’s global large‑model call volume chart, demonstrating strong competitiveness.

2. US AI Giants Form Rare Alliance

OpenAI, Anthropic, and Google, traditionally competitors, have begun sharing information through a “Frontier Model Forum” to curb Chinese AI rivals. Their cooperation includes sharing adversarial distillation behavior data, identifying model‑distillation actions that violate service terms, and coordinating technical embargoes to prevent Chinese models from learning via API calls. The move follows the Cursor‑Kimi incident and three weeks of Chinese model usage surpassing US models.

3. DeepSeek V4 Accelerates Product Iteration

DeepSeek’s web interface now offers two new modes: “Fast Mode” (instant response with image/file OCR) and “Expert Mode” (deep reasoning for programming, inference, research). The full V4 version is scheduled for release on April 14‑15, expanding the context window to 3 million tokens and directly targeting Tencent’s Hunyuan 3.0.

4. Zhipu GLM‑5.1 Launch on Huawei Cloud

Zhipu released the flagship GLM‑5.1 model, which went live on Huawei Cloud the same day and integrated with multiple Huawei services. Key capabilities include 8‑hour continuous task execution for engineering‑grade results, Ascend‑chip‑level MOE layer balancing with optimized Ascend Attention operators, and a 30% system‑level throughput increase. It suits complex programming projects, long‑duration data analysis, and automated operations.

5. Google CEO Sundar Pichai’s 2027 AI Forecast

Pichai told the “Cheeky Pint” podcast that 2027 will be a pivotal turning point for AI reshaping production. He noted that despite $175 billion in cash, Google still cannot secure enough storage chips, highlighting ongoing compute bottlenecks. Google is exploring a “Space‑level” data‑center concept akin to SpaceX, positioning AI as a foundational production infrastructure.

6. Anthropic’s Project Glasswing and Claude Mythos Preview

Anthropic partnered with Apple, Nvidia, Microsoft, AWS, Google, the Linux Foundation, and other tech giants to launch Project Glasswing. The initiative released the Claude Mythos Preview model aimed at network defense, capable of discovering the latest smartphone kernel vulnerabilities, detecting unauthenticated remote code execution, and earning the nickname “bug‑catching master.”

7. China Issues AI Ethics Review Measures

The Ministry of Industry and Information Technology and nine other departments issued the “AI Technology Ethics Review and Service Measures (Trial).” The document establishes an AI ethics review mechanism, lists high‑risk AI activities, and defines ethical requirements for the entire AI development, application, and promotion lifecycle, including the formation of an ethics review committee.

8. Guangdong’s First “Zhenwu” 10k‑Card AI Compute Cluster

China Telecom Guangdong and Alibaba Cloud launched the region’s first AI compute cluster based on the “Zhenwu” chip, featuring up to 10,000 cards (future expansion to 100,000), 4 µs network latency, >95% network utilization, and the capacity to train and infer models with hundreds of billions of parameters. The service is offered via the “Guangdong Telecom Compute Marketplace” with per‑card and per‑hour billing, lowering entry barriers for SMEs.

9. Open‑Source SentiAvatar Framework for Interactive 3D Digital Humans

SentiPulse, together with the Peking University‑Gaoling AI Institute, open‑sourced China’s first interactive 3D digital‑human framework, SentiAvatar. It includes the SUSU 3D character model and the SuSuInterActs high‑quality Chinese multimodal dialogue‑action dataset. The framework addresses pain points such as motion‑semantic mismatch, stiff expressions (the “uncanny valley”), and voice‑rhythm misalignment, enabling applications in virtual anchoring, intelligent customer service, and educational companions.

10. Market Trend: Large‑Model Stocks Surge

Following Alibaba’s reorganization and anticipation of DeepSeek V4, large‑model concept stocks rose sharply in the morning session. Notable gains: 德适‑B (+14.96% to HK$269), 智谱 (+14.63% to HK$893), MiniMax (+7.64% to HK$1,022). The analysis notes growing market confidence in the commercial prospects of Chinese AI large models and a continued rise in token‑economy concepts.

11. Token Economy Becomes the New Moat

In March 2026, China’s daily large‑model token calls exceeded 1.4 trillion, burning over 16 billion tokens per second. The competition logic shifts: while the mobile‑internet era measured success by user traffic (WeChat 10 billion MAU, Douyin 8 billion MAU), the AI era measures value by token usage and compute efficiency. Nvidia CEO Jensen Huang asserts that tokens will become the core, most valuable commodity in the digital world.

12. China Releases First Panoramic Carbon‑Accounting Model

China unveiled the world’s first full‑industry‑chain, full‑lifecycle carbon‑accounting system, the “Panshi·Yuheng Carbon Accounting Model,” in Shanghai. It enables comprehensive carbon accounting across production, consumption, and natural sources, supports multi‑level carbon‑footprint tracking for enterprises, products, and projects, and provides precise data for carbon tariffs and trading, marking a major breakthrough in global carbon‑emission accounting.