Representative End User Clients
Representative Automation Clients
Representative Software Clients
When Yin Qi re-emerged at the forefront of China’s AI industry, he already held two key pillars of the intelligent era: StepFun, focused on large models, and Thousand Miles Technology, dedicated to AI-powered smart vehicles. On January 26, 2026, this technical architect—who has witnessed the full transition from AI 1.0 to AI 2.0—formally assumed the role of Chairman at StepFun following the company’s over 5 billion RMB Series B+ funding round. This move completed the final piece of his vision for a Chinese version of an “xAI + Tesla” ecosystem.
This funding round not only set a 12-month record for single-round investment in China’s large model sector but also marked a clear industry shift from pure technological competition toward a new phase of synergistic development between the “brain” (large models) and the “body” (terminal devices).
Industry Background: Head Players Breaking Through Amid Reshuffling
China’s large model sector is entering a period of accelerated consolidation. As capital enthusiasm cools, the industry logic is shifting from technology demonstrations to a tangible race for value realization. In 2025, AI model-layer companies recorded only 22 disclosed investment deals, with total funding of 9.416 billion RMB, a sharp decline compared with 2024. Only three companies secured single-round investments exceeding 1 billion RMB, reflecting growing investor caution toward model-first projects lacking clear commercialization pathways.
At the same time, industry concentration is intensifying. In early January 2026, Zhipu AI and MiniMax both listed on the Hong Kong Stock Exchange, triggering market-wide attention. MiniMax’s share price surged 109.09 percent on its first trading day, pushing its market capitalization beyond 106.7 billion HKD, while Zhipu AI’s stock rose for three consecutive days to reach a valuation of 91.7 billion HKD. These developments underscore capital’s preference for leading players capable of forming closed-loop commercialization models.
StepFun’s 5 billion RMB funding round is a strong endorsement of its “technology + application” integration strategy. Its investor base is notably diversified, including state-backed capital such as Shanghai Guotou Xiandao Fund and China Life Private Equity, industrial capital such as Huaqin Technology, and continued backing from existing shareholders including Tencent, Qiming Venture Partners, and 5Y Capital. This diversity reflects broad-based confidence in StepFun’s approach.
Team Composition: A Golden Squad with No Weak Links
StepFun’s “1+3 Golden Squad” is widely regarded as top-tier within China’s large model startup ecosystem: Chairman Yin Qi, CEO Jiang Daxin, Chief Scientist Zhang Xiangyu, and CTO Zhu Yibo. Their complementary expertise spans strategy, foundational algorithms, large-scale engineering, and industry collaboration.
Yin Qi’s earlier founding of Megvii, one of the former “Four AI Dragons,” provided StepFun with deep industry resources and strategic foresight. CEO Jiang Daxin brings 16 years at Microsoft, where he served as Corporate Vice President leading Bing Search and AI R&D. A recognized authority in data mining and a 2025 IEEE Fellow, he is the only executive in China’s large model startup ecosystem to hold this distinction. Chief Scientist Zhang Xiangyu is a core contributor to the ResNet residual network, a foundational architecture in computer vision. CTO Zhu Yibo’s experience managing GPU clusters with tens of thousands of nodes underpins StepFun’s ability to deploy models at scale. Together, this team forms the backbone of StepFun’s technical depth and commercialization capability.
Technical Path: Differentiating with Native Multimodality
Rather than pursuing parameter scale alone, StepFun adopted a differentiated strategy centered on “native multimodality + device–cloud synergy,” building a “1+2” foundational model architecture aligned with AI + terminals. To date, the company has launched more than 30 models across diverse application scenarios.
Its flagship Step 3 model is a full-scale native multimodal reasoning system optimized for efficiency, delivering up to three times the inference speed of DeepSeek-R1 on domestic chips and more than 70 percent higher throughput on NVIDIA Hopper architecture GPUs. The model is scheduled for global open-sourcing in July 2025. StepFun places equal emphasis on infrastructure and algorithms, having invested in Shanghai Intelligent Computing Technology to secure long-term computing capacity.
On the device side, the Step GUI 4B model represents the industry’s first high-performance GUI model capable of large-scale smartphone deployment. The Step-Audio 2 end-to-end speech model reduces first-sound latency to under 0.7 seconds, significantly improving interactive experiences on smart terminals.
Commercial Closed-Loop: Realizing Value Through AI + Terminals
StepFun’s “large model + terminal” ecosystem closely mirrors Elon Musk’s xAI + Tesla paradigm, achieving scale in two major consumer domains: smartphones and automobiles. In smartphones, StepFun partners with 60 percent of China’s top brands, with cumulative deployments exceeding 42 million devices and nearly 20 million daily active users. Multimodal API calls surged more than 45-fold in the second half of 2024.
In the automotive sector, performance has been even more pronounced. The intelligent cockpit AgentOS, co-developed by StepFun, Thousand Miles Technology, and Geely Auto, has entered mass production in the Geely Galaxy M9, reaching nearly 40,000 units sold within four months. At CES 2026, its human-like voice interaction—powered by Step-Audio 2—attracted global attention. The system supports emotion recognition, personalized memory, and stable end-to-end interaction latency below 0.7 seconds, redefining intelligent cockpit benchmarks.
Investor Structure: Diverse Capital, Shared Conviction
The structure of this funding round highlights strong consensus around StepFun’s trajectory. Unlike many large model startups reliant on a single capital source, StepFun has attracted state capital, industrial capital, insurance funds, and market-oriented venture capital, forming a balanced and resilient investor base.
China Life Private Equity’s participation marks its first investment in large models, reflecting confidence in StepFun’s long-term prospects. Xiamen ITG Group represents its first entry into the large model space, underscoring regional governments’ strategic focus on AI. Huaqin Technology’s involvement strengthens alignment with StepFun’s AI + terminal roadmap, opening opportunities across the global supply chain.
The Physical AI Wave: The Industrial Starting Point for China’s AI 2.0
StepFun’s direction aligns closely with global AI trends. At CES 2026, Jensen Huang, CEO of NVIDIA, delivered a 90-minute keynote highlighting the imminent rise of Physical AI, drawing parallels to ChatGPT’s breakout moment. He emphasized that future AI systems must understand physical laws to interact meaningfully with the real world, announcing the open-source autonomous driving model Alpamayo and the Vera Rubin chip platform.
This perspective echoes Yin Qi’s view that AI 1.0 suffered from a disconnect between technology and real-world application—a gap Physical AI is designed to bridge. Yin Qi has outlined three objectives for 2026: surpassing 1 million installations of the Thousand Miles autonomous driving system, restoring StepFun’s foundational models to the global top tier, and launching AI-native hardware within 12 to 15 months. According to IDC, China’s smart terminal shipments are expected to exceed 900 million units in 2026, offering vast deployment scenarios for Physical AI.
From founding Megvii to leading both StepFun and Thousand Miles Technology, Yin Qi’s trajectory has consistently aligned with inflection points in China’s AI industry. The deep integration of large models and terminal devices is increasingly signaling the industrial dawn of China’s AI 2.0 era.