On July 16, Great Wall Strategy Consulting unveiled the "GEI China Unicorn Enterprise Research Report 2026" in Shenzhen. This report, now in its tenth consecutive year, highlights a significant divergence within the AI industry between large models and embodied intelligence.

The report reveals that China was home to 376 unicorn enterprises in 2025, boasting a combined valuation exceeding $1.4 trillion. Among these, 19 operated in the field of embodied intelligence, including 12 newcomers. On average, one new embodied intelligence unicorn surpassed the $1 billion valuation mark each month.

Simultaneously, unicorns in the AI large model sector are collectively shifting their focus from technical parameter contests to commercial monetization strategies.

One sector is surging forward, while the other is changing gears. This dynamic reflects a structural divergence within the AI industry.

Embodied Intelligence: The Entire Industrial Chain Ascends Together

According to the report's breakdown of the embodied intelligence industrial chain, among the 19 unicorns, 11 specialize in the embodied ontology (physical embodiment) direction, encompassing general-purpose humanoid robots, quadruped robots, and collaborative robotic arms. Five focus on algorithms and embodied data, all being new entrants, with an emphasis on end-to-end embodied large models and simulation data loops. Additionally, three new entrants are involved in core components and specialized chips.

The 12 newly emerged embodied intelligence unicorns are evenly distributed across the upstream components, midstream bodies, and downstream models of the industrial chain, indicating a simultaneous rise across the entire spectrum.

Ten companies, including Galaxy General, Zhiyuan Robotics, Xinghaitu, and Lingxin Qiaoshou, achieved unicorn status within three years of their establishment, with some reaching a $1 billion valuation in just over a year. This rapid growth rivals that of the most aggressive companies during the 2023 large model startup surge.

Notably, 83.8% of unicorn enterprises received investments from state-owned capital. Government industrial funds and state-owned capital are strategically deployed, focusing on strategic sectors such as semiconductors, embodied intelligence, and clean energy. In 2025, RMB financing accounted for 82.5% of all financing events, totaling $15.89 billion, or 75% of the total financing volume.

This indicates that China's capital support for embodied intelligence primarily stems from domestic industrial capital and government-guided funds, rather than the traditional reliance on US dollar funds.

Across the Pacific: Building Bodies vs. Training Brains

Shifting our gaze to the other side of the Pacific, embodied intelligence in the US also accelerated in 2025, albeit along a vastly different trajectory.

Figure AI secured its Series C funding in September 2025, reaching a post-money valuation of $39 billion. NVIDIA, Intel Capital, and Qualcomm Ventures all participated. According to Figure AI's official announcement and Securities Times reports, the committed funding in this round exceeded $1 billion. The valuation of this single US company may surpass the combined total of China's 19 embodied intelligence unicorns.

However, the narratives in China and the US diverge significantly. Chinese unicorns concentrate on embodied bodies and core components, emphasizing hardware mass production and supply chain integration advantages. In contrast, US companies place greater emphasis on foundational models. Figure AI announced in 2025 that it would part ways with OpenAI and independently develop the Helix VLA model (a multimodal system integrating vision, language, and actions), with the core narrative being the "general-purpose robot model."

Tesla's Optimus follows a distinct path. Musk set a target of producing 1 million units annually within five years. However, according to a research report by Minsheng Securities, Optimus currently operates at only 20-30% of human efficiency, costs $60,000 per unit, and has core components like harmonic reducers with lifespans of less than a year, while its dexterous hands last only one to three months. To use a supercar analogy, it can run, but not far; it can function, but not stably over the long term.

Upon comparison, it becomes evident that on opposite sides of the Pacific, China is focusing on building robot bodies, while the US is training robot brains. This reflects industrial development trends: China's accumulated strengths in precision manufacturing and supply chain integration align well with the scaling (mass production) needs of embodied bodies. Meanwhile, the US maintains a significant first-mover advantage in foundational large model research.

Large Models: From Showmanship to Monetization

If embodied intelligence is surging ahead, the AI large model sector is undergoing a strategic realignment.

The report clearly states that unicorns in the AI large model sector are accelerating their transition from technical parameter competitions to scenario-based implementation and commercial monetization.

Take Yuezhi Anmian as an example. Leveraging its ultra-long context processing capabilities and a large C-end user base, it recently joined the ranks of super unicorns. Among the 376 unicorns, only 12 are super unicorns. Yuezhi Anmian's rise validates that C-end user scale is becoming the core metric for valuing large model companies, rather than model parameters themselves.

Additionally, other large model unicorns each have their unique strengths. Jieyue Xingchen is committed to a full-stack multimodal approach, releasing a complete model system covering text, images, video, and voice. MiniMax introduced a hybrid architecture inference model and a high-quality video generation large model, making significant commercial progress with overseas C-end products. Zhipu Huazhang, relying on its GLM series foundational models, has built a comprehensive product matrix ranging from general capabilities to industry-specific solutions.

Although these four companies follow different paths, their direction is highly consistent: transforming model capabilities into chargeable products. The focus is no longer on who has larger parameters but on who achieves higher conversion rates and faster ARR (Annual Recurring Revenue) growth.

From a financing structure perspective, there were 206 financing events involving unicorn enterprises in 2025, nearly doubling from 2024, totaling approximately $21.38 billion. However, US dollar financing events accounted for only 17.5%. The main financing arena for the large model sector has shifted from Silicon Valley back to China.

A Decade: From Traffic to Computing Power

Great Wall Strategy Consulting summarizes the decade-long evolution of Chinese unicorns into three distinct stages. From 2016 to 2018, during the internet integration development phase, e-commerce and internet finance dominated. From 2019 to 2021, during the rapid growth phase of cutting-edge technologies, integrated circuits and new energy rose to prominence. From 2022 to the present, during the high-quality advancement phase, AI chips, large models, embodied intelligence, and commercial aerospace have become the main fronts.

A decade ago, unicorns were primarily traffic-driven companies. Today, the average age of the 376 unicorns is 7 years, with an average time of 4.3 years from inception to unicorn status. Sixteen companies achieved unicorn status within one year of establishment, a speed nearly impossible during the internet era.

The profile of founders is also evolving. Over 40% of unicorn enterprise founders hold doctoral degrees. Tsinghua University has produced 46 alumni founders, with the total valuation of the companies they founded exceeding $400 billion. Yang Zhilin of Yuezhi Anmian, Zhang Peng of Zhipu AI, and Wang He of Galaxy General are all Tsinghua graduates.

By 2025, cutting-edge technology unicorn enterprises will reach 320, accounting for 85.1% of the total, up 14.9 percentage points from the previous year. The criteria for selecting unicorns have shifted from business model innovation to building technological barriers.