Analyst Serenity said in a post on X that China’s primary-market AI funding is shifting sharply away from early-stage base-model startups and toward embodied AI, physical AI, and world models. In his view, fundraising for pure foundation-model companies is now close to shut, while the U.S. could follow a similar pattern, with capital concentrating around leaders such as Anthropic and OpenAI.
Using Serenity’s own tally, AI fundraising in China’s primary market is currently spread across five broad categories: $23.56 billion for large language models, $15.74 billion for AI infrastructure and technical layers, $13.36 billion for embodied AI and physical AI, $8.79 billion for AIGC applications, and $3.82 billion for autonomous driving and other top-20 clusters. Serenity noted that the last category is not directly comparable with the others because it uses a different statistical scope.
Capital is leaving early base-model bets
Even though LLMs remain the largest category by total funding, Serenity’s point was that most of that money is going to established players rather than new foundation-model entrants. That leaves less room for fresh startups built around pure base models. At the same time, embodied AI, physical AI, and world models are attracting more institutional attention as investors look for the next major theme.
Serenity wrote that “world models have become the biggest consensus in early-stage investing.” Yet the market has not produced a clean public-market target for that thesis. He said there is still no clear pure-play name, and that investors may need to wait for a later wave of IPOs before the field becomes easier to access.
Strong narrative, few pure-play names
The source material says Serenity had already flagged 4D AI and world models as areas worth watching months ago, and mentioned AEVA as a possible way to gain exposure. He also made clear that AEVA is not an obvious pure world-model proxy. The report cites external verification showing that AEVA launched its Omni sensor at CES 2026, supplied Daimler Truck, and posted record revenue in the first quarter of 2026.
Interest in the theme has also grown outside of public-market discussions. According to the report, the 2026 BAAI conference shifted its focus toward world models, and Alibaba led a $290 million funding round in April 2026 for Shengshu Technology’s world-model product Vidu. Those developments support Serenity’s view that capital is searching for an entry point into the sector, even if direct investment targets remain limited.
Serenity expects a similar concentration trend in the U.S.
Serenity did not describe the move as a simple one-way rotation. He said AI infrastructure and semiconductor supply chains are still drawing funds, while physical AI, embodied AI, humanoid robots, and world models are becoming key destinations for rotating capital. For frontier models, he expects money to keep clustering around a small number of leaders rather than spreading across a new class of startups.
The report also references verified figures showing that China’s embodied AI sector raised RMB 73.5 billion in 2025, with more than RMB 20 billion raised in the first two months of 2026. China accounted for more than 43% of global robot venture investment in that period. Serenity also said AIGC applications remain one of the most mature commercial areas in AI, though he does not see a clearly defined winner yet.