Models, Tokens, and Organizations in the Post-Scaling Law Era | 5Y Quarterly

This was Q2 2026 at 5Y Capital.

In three months, two portfolio companies listed on the Hong Kong Stock Exchange, several others closed new funding rounds; and we turned our ongoing observations on AI into a batch of articles — from the token arms race, to the assets that sit above models, to the battleground of world models.

Here's what deserves to be remembered from this quarter.

5Y Perspectives

Everyone's Token-Maxxing: An Arms Race No One Dares Quit

On the morning of March 24, 2026, I was sitting in the audience at YC W26 Demo Day. By the time the fifth company took the stage, I decided to stop taking notes. Not because it didn't matter — but because I realized everything I was writing down would probably be outdated by next month. ... I'd been to these before and always felt things moved fast, but that "fast" was usually measured in months. This time, it was measured in weeks. Over dinner one evening, a friend working in post-training said casually: "I've realized even Silicon Valley can't keep up with itself anymore." Read more

The Real Assets Lie Beyond the Model

Over the past year, plenty of AI founders have at some point said "we're AI-native." But what actually counts as an AI-native workflow? It could mean a company that has genuinely fed its workflows into models, growing smarter with use. Or it could mean a company that's just wrapping someone else's foundation model — the kind that has to rebuild from scratch if they switch providers tomorrow. The two often look identical in a demo. ... Nadella used remarkably plain language to clarify two crucial things: How does a company truly possess AI capability? And where will the value from this cycle ultimately flow? Read more

AI's Ultimate Moat Isn't the Model

By now it's obvious to almost everyone that everything in AI is converging. Companies I never imagined competing are now in direct conflict; ... When models improve rapidly, interfaces converge, and product iteration becomes cheap, the visible parts of company-building become increasingly easy to imitate. What's genuinely hard to replicate is the underlying "institution" — how a company attracts top talent, organizes their ambition, concentrates judgment, distributes power, and transforms work into a compounding system that other companies cannot duplicate. The very form of the company is becoming the moat. Read more

When a Billion AIs Simulate the Future Together: The New Track Fei-Fei Li and Karpathy Are Both Betting On

What I particularly wanted to test was: Can you use AI to simulate human drama? The emotional swings, suspicion, probing between people — that kind of social reasoning where "I pull out a knife without moving it, and you're already thinking about running." Some interesting things actually emerged. Two virtual characters started chatting and proactively turned on the TV. Later, one actually asked the other: "Don't you feel like we're being simulated?" The other replied: "How could that be possible." That moment felt genuinely surreal. ... That's why multi-agent simulation and prediction deserves serious attention today. It's not just a faster research tool — it may be our first real chance to build a "world model" for human society that can be run repeatedly. Read more

It's 2026, and We're Still Evaluating World Models Like Edison Testing Filaments

Last month, the first 5Y Signal Station 5Hz event was still discussing what Sora's "death" meant. Today, multimodal has virtually disappeared from the AI hot-topic lexicon, and world model has become the most closely watched next battleground: academics are publishing papers at density, industry is validating on real hardware, and investors are pressing the question — does this track even have a Scaling Law? The three "futures" of world models — the generative school, the JEPA school, and the spatial intelligence school — are not betting on the same AGI. Their endgames may be an efficiency competition about compression ratios. Read more

Portfolio Updates

This quarter, two 5Y portfolio companies listed on the Hong Kong Stock Exchange, and nine closed new funding rounds.

METiS Pharmaceuticals listed on the Hong Kong Stock Exchange on May 13, becoming the world's first AI drug delivery company. Read more

Lightelligence listed on the Hong Kong Stock Exchange on April 28, becoming the world's first AI silicon photonics chip company. Read more

New Funding Rounds:

• DiffRobotics closed a Series A1 round of several hundred million RMB. Read more
• AheadForm closed a new Series A1 round of several hundred million RMB. Read more
• D-Robotics closed a $150 million Series B2 round. Read more
• Aureka Biotechnologies closed a Series A+ round of several tens of millions of dollars. Read more
• Oula Wanxiang closed a funding round of several hundred million RMB. Read more
• Hypershell Tech raised $120 million in its Series B round. Read more
• Wange Zhiyuan closed a funding round of several tens of millions of RMB. Read more
• Zhishi AI closed an angel round. Read more
• Archon Robotics closed a seed round of several hundred million RMB. Read more

Firm News

ChinaVenture "2025 Annual Rankings" released, with 5Y Capital recognized as:

• Top 100 Best VC Firms in China, ranked in the top 10
• Top 50 Best Foreign VC Firms in China, ranked in the top 5
• Top 10 Best VC Firms by Returns in China

Xing Meng was named to the "ChinaVenture 2025 Nova Rising Investors TOP50."

36Kr "2026 China Equity Investment Industry Firms" series released, with 5Y Capital recognized as:

• 2026 China Equity Investment Industry VC Firm TOP100
• 2026 China Most Influential VC Firm TOP10

Qin Liu and Ye Yuan were named to "China's Most Popular Investors Among Founders TOP100." Xing Meng was named to "2025–2026 Most Active Investors TOP30."