What comes after DeepSeek? Predicting China’s next big AI moves.

Part one of three

The ripple effects of DeepSeek’s launch of its V3 and R1 models in late January is still being felt.

Compared with the expensively developed LLMs from OpenAI, Meta, and Google, DeepSeek is cost efficient, high performance, and open source.

Other tech giants and AI startups in China have also rolled out additional LLM and reasoning models that are outperforming DeepSeek in various benchmarks.

This includes Qwen2 from Alibaba, Doubao 1.5-pro by ByteDance, GLM from Zhipu, Kimi and 01-series from MiniMax, and Hunyun-Large from Tencent (which GitHub and Hugging Face call the largest open-source Transformer-based MoE model in the industry).

Last week, Wuhan-based startup Butterfly Effect launched the world’s first general AI agent, Manus. It has gone viral on and made its way into the global conversation, with influential voices in tech, including Twitter cofounder Jack Dorsey and Hugging Face product lead Victor Mustar, praising its performance. Some have even dubbed it “the second DeepSeek”.

Manus leverages multiple AI models and various independently operating agents to act autonomously on a wide range of tasks such as travel planning, code writing, and apartment searches. According to a tech reviewer for MIT Technology Review, this beta version of Manus is already performing at 80% of what she expects an AGI agent would. Numerous other technical reviews praising Manus have been published online since Monday when invitation-only tests started.

MIT Technology Review is actively tracking a Chinese AI ecosystem with more than a dozen high potential Chinese AI firms from Stepfun, Zhipu, Minimax, Moonshot, 01.AI, ModelBest, and Baichuan.

I have been following tech developments in China for many years, especially in AI and military tech. The scale and speed of innovations today is unprecedented.

I think it would be interesting to make some predictions about Chinese AI development in the next 2 or 3 years. I’m organizing this discussion in three parts, one on each the main trend -

- Embodied AI, i.e. humanoid, will be the next major Chinese tech disruption.

- Chinese AI will move into application stage across multiple industries, specially manufacturing. Vertical application-level AI will be the source of long-term growth and profitability compared with horizontal foundational AI, represented by LLMs today.

- AI will go mass market as costs are pushed down through smart engineering and algorithmic optimization. The moat being built by Silicon Valley tech heavies in the form of raw compute, proprietary software, and massive Capex will be eroded as smart AI systems emerge that will be less compute-intensive. Rather than protected by paywalls, Chinese AI will be affordable, open source, and widely adopted, democratizing this latest technological frontier.

Embodied AI

China’s robotics and humanoid industry is already world leading. A full ecosystem and supply chain is in place to enable product innovations and iterations in embodied AI. And over half of the world's industrial robots are used in China. Local demand is driving embodied AI innovations.

Companies such as DJI and Unitree are global leaders in drones, quadruped and humanoid robots. DJI has a 80% global market share for consumer drones. And Unitree accounts for 70% global market share for quadruped robots. Unitree humanoid can dance, lift heavy loads, flip forward, and make kung fu moves.

Beijing Humanoid Robot Innovation Center unveiled the world’s first full-size purely electric running robot, Tiangong, last November, capable of maintaining a steady speed of 6 kilometers per hour.

Humanoid robot from Shanghai-based AgiBot can achieve nearly human-like mobility such as riding a bicycle and balancing on a hoverboard. The robot can also quickly read and understand medical description. The robot can achieve millisecond-level interaction responses, assess human’s emotional states through their facial expressions and vocal tones, and provide corresponding responses. The humanoid is being deployed to nursing homes to assist the elderly and schools as security guards.

Much like drone swarms, these AI-enabled humanoids, based on multimodal large language models, can collaborate with other robots in performing complex tasks.

EV makers such as Xpeng and Xiaomi are building their own industrial robots and deploying them in their car assemble lines.

Xpeng plans to invest $14 billion in its current humanoid robot with Level 3 capabilities, Iron, to enter mass production by end of the year. Xpeng also unveiled its flying car, the Aeroht, at CES 2025 in January. The model features a modular architecture, which includes a detachable air module housed within the ground vehicle.

BYD, the largest EV maker in the world, is building a roof-mounted drone hanger for some of its EVs that can house a DJI drone to be launched at the push of a button. This is to enable better aerial photography for your road trip.

Ewaybot Technology develops intelligent service robots designed for healthcare, hospitality, and public services. The company's AI-enabled robots perform tasks such as disinfection, food delivery, and customer assistance.

Siasun Robot provides a wide range of industrial robots, service robots, and automated guided vehicles (AGVs), focused on factory automation.

Megvii is the world leader in computer vision and facial recognition technology. Its AI-powered systems are used in autonomous navigation and human-robot interaction. Megvii's technology is pivotal in enhancing the capabilities of robots to perceive and interact with their environment effectively.

RoboSense is the leader in LiDAR technology for autonomous vehicles and robots. LiDAR, which uses laser pulses to map the environment in 3D, is instrumental in creating accurate and reliable perception systems for robots. RoboSense's LiDAR technology is pivotal in advancing the capabilities of robots in navigation, obstacle avoidance, and mapping.

CloudMinds specializes in cloud-based AI and robotics solutions. It integrates cloud-based AI with on-ground robots, enabling real-time data analysis and enhanced decision-making capabilities.

Because China already does so much manufacturing, it has “the luxury” of trying out automation and robotic solutions without fearing disruptions if things do not run smoothly at first.

The country’s sprawling manufacturing sector also provides newer robotics firms with access to an industrial and critical mineral supply chain that they can tap into for parts, components and raw materials.

The Chinese embodied AI industry is predicted to be on the cusp of achieving mass production and commercialization in 2025. Over the next decade, embodied AI is projected to have revenue potential in the hundreds of billions of dollars.

Morgan Stanley recently published a report on the robotics industry. Of the 100 publicly-traded companies worldwide that Morgan Stanley tracks in developing humanoid robots, 56 are based in China.

China is also home to 45 per cent of the world’s integrators, which are firms that customise robots to match end-user needs, according to the report.

Most humanoid robots in use today operate in industrial settings, especially in the logistics and manufacturing sectors. This gives China a significant edge as the largest industrial country.

Unlike in the US where deep-pocketed industry giants such as Tesla, Nvidia, and Boston Dynamics dominate the robot landscape, China is characterized by a proliferation of small to medium-sized enterprises striving to navigate the competitive terrain of humanoid robots, making China a hotbed for breakthrough innovation.