Forbes China, in partnership with First Momentum, has released the “2026 Forbes China AI TECH Enterprises TOP 50,” providing the most comprehensive annual snapshot of the nation’s artificial intelligence industry. The 2026 edition, officially announced on May 17, reveals a decisive shift in the character of China’s AI sector: the industry is moving from application-driven growth toward technology-driven depth, with full-stack capabilities and domestic hardware independence emerging as the defining competitive advantages.
Technology Firms Dominate the List
Technology-oriented enterprises account for 61.2 percent of the selected companies, while application-oriented firms represent 34.1 percent. This ratio reflects a maturing industry in which foundational capabilities, large language models, AI chips, embodied intelligence, and enterprise AI infrastructure, are increasingly valued over consumer-facing applications. The competitive landscape displays a three-tiered structure: at the head tier, embodied intelligence and robotics claim 13 enterprises (15.3 percent) and large language models account for 11 (12.9 percent), together representing more than 28 percent of the list. In the mid-tier, AI marketing and content generation (10.6 percent), AI healthcare (9.4 percent), and enterprise AI and intelligent agents (9.4 percent) collectively account for nearly 30 percent.
Geographically, Beijing leads with 21 enterprises, followed by Chengdu with 11, and Shanghai and Shenzhen with 12 each. Hangzhou, home to Alibaba and a dense cluster of AI startups, contributes 9 enterprises and leads all cities in the number of large language model companies selected. The geographic spread, with significant representation from Wuhan, Guangzhou, Changsha, and Chongqing, confirms that AI development is no longer concentrated exclusively in China’s three major tech hubs.
The AI Stack Decouples From Western Hardware
The most strategically significant finding in the 2026 report is the accelerating decoupling of China’s AI stack from Western hardware. DeepSeek’s V4 model, released in May 2026, has fully migrated its inference workload from CUDA — Nvidia’s proprietary computing platform, to domestic compute infrastructure. The model also pushes context windows past one million tokens and cuts long-context processing costs by over 50 percent through sparse attention and fine-grained expert parallelism.
On the chip side, Cambricon’s Siyuan 590 has achieved full-scenario mass shipment, while the newer Siyuan 690 delivers more than 700 TFLOPS of FP16 performance with Day-0 support for DeepSeek-V3.2 and V4. Biren Technology’s Bili BR166 series is in full-form mass production, having already delivered multiple multi-thousand-card intelligent computing clusters to enterprise customers, with the next-generation BR20X planned for later in 2026. These developments confirm the trajectory that China’s domestic chipmakers have been on since seizing 41 percent of the local AI market as Nvidia’s footprint in China has contracted.
Model Innovation Across the Ecosystem
Beyond DeepSeek, the report highlights a diverse ecosystem of model innovation. Kimi’s K2.6 has pivoted from consumer chatbot to enterprise productivity tool, pairing a 256K-token context window with an Agent Swarm that scales to 300 simultaneous agents and scoring 80.2 percent on the SWE-Bench software engineering benchmark. StepFun has shipped 42 models in total, with Step 3.5 Flash dominating OpenRouter call volume and StepAudio 2.5 TTS topping the Artificial Analysis audio benchmark. Volcano Engine’s Doubao 2.0 processes 120 trillion daily tokens across a full-modal matrix spanning text, voice, image, video, and 3D content.
Alibaba’s Qwen3.6 series maintains Day-0 DeepSeek support and is anchoring China’s open-source LLM push into global markets, with the model family now widely used by developers outside China. The 2026 edition of the Forbes China AI Top 50 also expanded its evaluation framework from four to seven sub-award categories and added new dimensions, including global expansion, sustainability, and open-source contribution — reflecting the industry’s maturation beyond purely domestic metrics. Taken together, the list presents a picture of an industry that has moved well past the imitation phase and is now generating original capabilities at the frontier.
From Domestic Champion to Global Contender
The addition of global expansion as an evaluation dimension in the 2026 Forbes China AI Top 50 reflects a significant shift in the industry’s self-conception. For most of the past decade, Chinese AI companies were primarily focused on the domestic market, which was large enough to sustain growth without requiring international expansion. That calculus is changing. As the domestic market matures and competition intensifies, the companies on the Forbes list are increasingly looking outward, to Southeast Asia, the Middle East, Europe, and eventually the United States.
The open-source strategy is central to this global push. Models like DeepSeek V4, Qwen3.6, and MiMo-V2.5-Pro are being downloaded and deployed by developers worldwide, building brand recognition and technical credibility that commercial sales teams can then leverage. The Forbes China AI Top 50 list, by adding global expansion as a criterion, is effectively signaling to the industry that international reach is no longer a bonus, it is a requirement for elite status. For a sector that was once dismissed as a fast follower of Western innovation, that ambition represents a remarkable transformation.