Q1 AI funding hits 110 billion RMB in China

🔑 Enhanced Key Takeaways

•Foreign-currency investments in China's AI sector surged by 495% year-over-year in Q1 2026, with deal count more than doubling to 210 agreements, indicating international investors are increasingly tolerating geopolitical risks for Chinese AI exposure.
•China's AI investment is strategically shifting beyond large language models to include embodied AI and "AI + Terminals," aiming to integrate AI directly into consumer electronics like smartphones and vehicles.
•The Chinese government's "AI Plus" plan, a ten-year strategy launched in January 2026, aims for comprehensive AI integration across the economy and society by 2035, with a short-term goal of 70% AI penetration in "intelligent terminals" by 2027.
•US export controls have spurred China's focus on domestic GPU development and large-scale systems, with companies like Huawei developing Ascend series chips and utilizing enhanced inter-GPU connectivity to mitigate hardware limitations, though a performance gap with Nvidia persists.
•Chinese open-source large language models, such as DeepSeek's V4 and Moonshot AI's Kimi, are undergoing rapid iteration, with Chinese-developed models accounting for about 41% of LLM downloads on Hugging Face over the past year, and some models demonstrating cost efficiency advantages over US counterparts.

📊 Competitor Analysis▸ Show

Feature/Company	Moonshot AI	StepFun	Z.ai (Zhipu AI)	MiniMax
Primary Focus	Long-context LLMs, AGI, consumer software (Kimi chatbot)	Multimodal LLMs, "AI + Terminals" (integration into hardware like smartphones, vehicles)	Enterprise clients, generative AI models	Consumer software, generative AI models
Key Models	Kimi family (Kimi, Kimi K1, K2, K2 Thinking, K2.5)	Step series (Step-1V, Step-2, Step 3, Step-Video-T2V, Step-Audio)	GLM models (e.g., GLM-4)	Generative AI models
Technical Highlights	Kimi K2 Thinking: 1-trillion-parameter MoE, 32B active parameters, 256K token context, INT4 quantization, MoonViT vision encoder. Mooncake serving platform.	Step-1V (>100B parameters, multimodal), Step-2 (>1T parameters). Focus on inference efficiency and edge-cloud collaboration.	GLM-4 models show balanced performance in English and Chinese.	Notable breakthroughs in code generation and program comprehension.
Global Comparison	Kimi K2 Thinking claimed to outperform GPT-5 and Claude Sonnet 4.5 on some benchmarks.	Competes with global leaders like OpenAI and Anthropic.	Often compared to OpenAI and Anthropic.	Competes with global leaders like OpenAI and Anthropic.

🛠️ Technical Deep Dive

Moonshot AI's Kimi K2 Thinking model features a 1-trillion-parameter Mixture-of-Experts (MoE) architecture with 32 billion active parameters.
It supports up to 256,000-token contexts and utilizes native INT4 quantization for efficiency.
The Kimi K2.5 multimodal upgrade includes a 400-million-parameter vision encoder called MoonViT, enabling processing of images and video for agentic tasks.
Moonshot AI's custom serving platform, Mooncake, uses a KVCache-Centric Disaggregated Architecture, achieving up to a 525% increase in throughput in simulated scenarios.
StepFun's Step-1V is a multimodal large language model with over 100 billion parameters.
StepFun is testing its Step-2 model, which is reported to exceed 1 trillion parameters.
The company has released state-of-the-art open-source models like Step-Video-T2V (text-to-video) and Step-Audio (voice interaction).
StepFun's strategy emphasizes "AI + Terminals," integrating AI directly into hardware like smartphones and electric vehicles, focusing on edge-cloud collaborative models.

🔮 Future ImplicationsAI analysis grounded in cited sources

China will significantly reduce its reliance on foreign advanced AI chips by 2028.

Ongoing US export controls and substantial state-backed investments are accelerating domestic GPU development and the adoption of large-scale systems to compensate for hardware limitations.

Chinese AI companies will increasingly dominate the "AI + Terminals" market globally by 2030.

Companies like StepFun are strategically embedding AI into consumer devices and vehicles, leveraging China's advanced manufacturing base and large domestic market for rapid adoption and iteration.

The cost-per-token gap between Chinese and US frontier AI models will widen in favor of China, attracting more international capital.

Chinese labs are incentivized to optimize for efficiency due to compute scarcity, leading to models that run at a fraction of the cost of comparable American systems, making them attractive to investors.

⏳ Timeline

2017

China's government defined its long-term strategy to become a global AI leader.

2023-03

Moonshot AI was founded by Yang Zhilin, Zhou Xinyu, and Wu Yuxin.

2023-04

StepFun was founded by Jiang Daxin, a former chief scientist of Microsoft Asia Research Institute.

2024-02

Alibaba Group led a $1 billion funding round for Moonshot AI, valuing it at $2.5 billion.

2026-01

Moonshot AI raised $500 million in a Series C round, boosting its valuation to $4.3 billion, and released Kimi K2.5, a multimodal upgrade.

2026-01

StepFun secured over $717 million in a B+ financing round, the largest single investment in China's large model space in the past year.

Q1 AI funding hits 110 billion RMB in China

⚡ 30-Second TL;DR

Key Points

🧠 Deep Insight

🔑 Enhanced Key Takeaways

🛠️ Technical Deep Dive

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline

📎 Sources (19)

👉Related Updates