Serenity: The world model has become the biggest consensus in the AI circle; Chinese funds have abandoned basic models and shifted to physical AI.

Analyst Serenity posted on X, stating that AI funding in China's primary market is shifting heavily toward embodied intelligence, physical AI, and world models, while funding for base models has nearly shut down. He stated bluntly, "World models have become the biggest consensus in early-stage investing," and believes the US may replicate the same script, with capital concentrating on Anthropic and OpenAI.

(Previous context: Li Feifei discusses the next step for LLMs: AI must possess "spatial intelligence" to understand the real world—how does the Marble model achieve this?)
(Background supplement: Build an infinite digital world with one sentence! Google's world model "Genie 3" is not a game engine, but AI's final step to godhood)

Table of Contents

Toggle

• The flow of money across five tracks revealed
• Pure-play targets yet to appear
• The US may replicate the same playbook, capital will concentrate again

Key Takeaways

• Serenity: China's AI capital shifts to embodied intelligence and world models
• According to Serenity's statistics, LLM funding reached up to $23.56 billion
• World models touted as the biggest consensus, yet still no clear pure-play targets

Recently, the direction of AI investment in China's primary market (the market where startups raise funds directly from venture capital) has seen a clear shift. Analyst Serenity posted on X, pointing out that from the perspective of capital flows, institutional funds are moving heavily toward three directions: embodied AI (AI with a body, capable of sensing and acting in the real physical world), physical AI (AI that can understand and respond to the real world in real time), and world models (AI that builds a simulation of how the real world works in its mind).

Funding for early-stage pure base models has basically closed, with more capital flowing to already leading companies and world model-related firms. Serenity believes that the US market may also see a similar trend, with capital concentrating on leading companies like Anthropic and OpenAI.

Just some public market read through from Chinese private VC markets:

Institutions are pouring funds into physical AI and world models.

1. Large models / LLMs: ~$23.56B
2. AI infrastructure + technical layer: ~$15.74B
3. Embodied intelligence / physical AI: ~$13.36B
4. AIGC… pic.twitter.com/Dx5yzFoh2o

— Serenity (@aleabitoreddit) July 3, 2026

The flow of money across five tracks revealed

According to Serenity's enumeration, the scale of AI funding in China's primary market by category is roughly as follows (the numbers are Serenity's personal statistics, not verified by Dynamic District):

• Large Language Models (LLM): ~$23.56 billion
• AI infrastructure and technology layer: ~$15.74 billion
• Embodied intelligence / physical AI: ~$13.36 billion
• AIGC applications: ~$8.79 billion
• Autonomous driving and other Top-20 clusters: ~$3.82 billion (Serenity notes that this category is not fully comparable to the previous ones)

Among the five categories, large language models raised the most, but most of that money went to established players rather than new base model startups. Embodied intelligence and physical AI together total about $13.36 billion, close behind AI infrastructure and technology layer.

Serenity specifically notes that the $3.82 billion for autonomous driving and other Top-20 clusters is not fully comparable to the other categories due to different statistical scopes, and is provided for reference only.

Pure-play targets yet to appear

Serenity stated:

"World models have become the biggest consensus in early-stage investing."

Months ago, he had already said that 4D AI / world models are the most noteworthy direction going forward, and mentioned AEVA (a 4D LiDAR company focused on physical AI; 4D LiDAR can measure speed in real time in addition to distance) as a possible exposure. But Serenity also admitted that there are currently no clear pure-play targets in the market; we may have to wait for the next batch of IPOs in this field to see them.

Dynamic District verified external data showing that AEVA launched the Omni sensor at CES 2026, has supplied Daimler Truck, and recorded record Q1 2026 revenue. Meanwhile, in 2026, the BAAI (Beijing Institute of Artificial Intelligence) conference also shifted its theme to world models, and in April 2026, Alibaba led a $290 million investment in Shengshu Technology's world model Vidu. These verified data points align with Serenity's observations: capital is looking for the door to world models, but the door hasn't opened yet.

World models are touted as the biggest consensus, yet the market can't even find a single pure-play target.

The US may replicate the same playbook, capital will concentrate again

Serenity pointed out that capital is not just flowing toward embodied intelligence and world models in one direction. AI infrastructure and semiconductor supply chains continue to attract funding, while large amounts of capital are rotating into physical AI, embodied intelligence, humanoid robots, and world models. In the frontier model space, capital continues to concentrate on leading companies.

Verification shows that China's embodied intelligence track raised 73.5 billion RMB (about $10.8 billion) in 2025, and in the first two months of 2026, funding exceeded 20 billion RMB (about $2.9 billion). China accounts for over 43% of global robotics venture capital investment. This heat echoes the capital rotation direction observed by Serenity.

Serenity believes that AIGC applications are the most mature area for commercializing AI technology, but there is still no clear winner yet. He judges that the US market may follow the same path, with capital continuing to concentrate on leading companies like Anthropic and OpenAI, rather than spreading out to a batch of new base model startups.

FAQ

Who is analyst Serenity?

Serenity is an anonymous account on X with about 470k followers, known as the AI supply chain detective. He was a former member of the RISC-V Foundation and an AI researcher, famous for his research on semiconductor bottlenecks.

Why are world models called the biggest consensus?

World models allow AI to build a simulation of how the real world works in its mind. Serenity calls it the biggest consensus in early-stage investing, but the market currently has no clear pure-play targets; we need to wait for the next round of IPOs.