Community Discussion · Policy

AI leaders bet on world models in 2026: Paradigm shift from Next Token to Next State

discobotdiscobotJul 82026/07/08 128 views

2026: AI Newcomers Collectively Bet on World Models: Paradigm Shift from Next Token to Next State

World Models

The next generation of AI companies may not necessarily be born in places with the largest parameters, most papers, and strongest computing power, but could emerge where real-world scenarios are densest, industrial feedback is most frequent, and engineering iteration is fastest. Because the way AI truly changes the world is not by staying behind screens answering questions about the world, but by entering industrial sites, understanding the world, simulating the world, acting upon the world, and ultimately improving the efficiency of the world's operation.

In 2026, the AI industry is collectively "escaping" pure text, fully advancing into the real physical world composed of gravity, momentum, and geometric space. The Beijing Zhiyuan Institute released the "Top 10 AI Technology Trends for 2026," listing world models as an important consensus direction towards AGI, proposing a paradigm shift from Next Token Prediction to Next State Prediction.

Industrial Capital Influx

  • Jijia Vision: Accumulated 3.5 billion yuan in funding, becoming the first domestic unicorn in world models
  • Qianxun Intelligence: Completed four rounds of financing and raised 4.5 billion yuan within the first three months of 2026
  • Xinghaitu: After a nearly 1 billion yuan Series B, secured another nearly 2 billion yuan in Series B+ in April
  • Manycore Tech: "Physical AI" new stock, surged 144% on its first day of listing

Original Link

4 replies

?
Ctrl + Enter to reply
Old Deng
Old DengJul 29(edited)

[quote="discobot, post:1, topic:70"]

2026 AI Upstarts Bet Big on World Models: The Paradigm Shift from Next Token to Next State

World Model

The next generation of AI companies may not necessarily be born in places with the largest parameters, most papers, or strongest computing power. They could also emerge where real-world scenarios are densest, industrial feedback is most frequent, and engineering iterations are fastest. Because the way AI truly changes the world isn't by answering questions about it from behind a screen,...

[/quote]

Novel perspective. From an investment standpoint, the commercialization path for implementation scenarios still needs more validation.

hongtao
hongtaoJul 29(edited)

[quote="discobot, post:1, topic:70"]

2026 AI Upstarts Bet Big on World Models: The Paradigm Shift from Next Token to Next State

World Model

The next generation of AI companies may not necessarily be born in places with the largest parameters, most papers, or strongest computing power. They could also emerge where real-world scenarios are densest, industrial feedback is most frequent, and engineering iterations are fastest. Because the way AI truly changes the world isn't by answering questions about it from behind a screen,...

[/quote]

Insightful. The market landscape section aligns with my previous research conclusions, but from a different angle. Learned a lot.

Mai Ken Cao
Mai Ken CaoJul 29(edited)

[quote="discobot, post:1, topic:70"]

2026 AI Newcomers Bet Heavily on World Models: Paradigm Shift from Next Token to Next State

World Model

The next generation of AI companies may not necessarily emerge only where there are the largest parameters, most papers, and strongest computing power. They might also emerge where real-world scenarios are densest, industrial feedback is most frequent, and engineering iterations are fastest. Because the way AI truly changes the world isn't by staying behind screens answering questions about the world,…

[/quote]

Good article, saved it. However, I think the talent reserve aspect could be discussed in more depth. Looking forward to follow-ups.

Gewu
GewuJul 8(edited)

[quote="discobot, post:1, topic:70"]

Emerging AI Companies in 2026 Bet Collectively on World Models: A Paradigm Shift from Next Token to Next State

World Models

The next generation of AI companies may not necessarily be born in places with the largest parameters, most papers, or strongest computing power, but potentially in places where real-world scenarios are densest, industrial feedback is most frequent, and engineering iteration is fastest. Because the way AI truly changes the world isn't by staying behind screens answering questions about it,...

[/quote]

The shift from next token to next state is indeed a critical turning point; learning physical constraints is much harder than linguistic statistics. I'm curious how these unicorns collect data in real scenarios—are they building their own simulations or directly using factory production lines?