5Y Capital's Kai Liu: The Hype Is Just Surface Noise — AI Entrepreneurship Is Only Just Beginning | 5Y View

Recently, 5Y Capital partner Kai Liu shared his perspectives on the current state of AI startups and capital markets at a roundtable organized by Waves ("暗涌") themed "Craziness Is Only the Surface." Liu noted that the current wave of hype is just the surface — the deeper shift is that AI is moving from centralization to openness, giving entrepreneurs far more agency. From large models to applications, from valuation logic to PMF assessment, he offered a ground-level view. In his eyes, today's AI entrepreneurs, like the early participants in the 2010 mobile internet wave, are only just reaching the starting line.

Selected excerpts from the event:

Q: From an investor's perspective, how have companies like DeepSeek and Manus changed the investment landscape?

Liu: I think DeepSeek, Manus, and many model and application companies in the US represent a trend — possibly a split between the "Protestant" and "Catholic" branches of the AI era. The old AI world was like the "Catholic" era: all knowledge and capability concentrated in the "Church," where ordinary people had to rely on authority (OpenAI, Meta, and other giants) to "access truth." But now, we're entering a "Protestant" era: technology is more open, tools are more accessible, and entrepreneurs have the opportunity to build their own models and run their own applications. We've invested in both large models and applications. So I think up until this year, everyone worshipped those few companies — OpenAI, Meta — and worshipped the big tech giants. But after round after round of information bombardment, people are realizing this doesn't need to be so centralized. There are plenty of entrepreneurial opportunities. DeepSeek started it, fired the first shot of the revolution. Qwen has done very well too. These things aren't that difficult.

On the application side, in the past, if you wanted to build an impactful AI product, conventional wisdom said you had to "go global to the US" or rely on big-company or famous-researcher credentials. Now it's different. As the two of you just mentioned, we're seeing more and more young people — including many students in universities — who can build excellent applications on their own, and consume massive amounts of tokens in the process. I think the barrier is dropping rapidly. So my own summary is: all this AI hype is just the surface. What's more important is that we're opening a new era.

Q: When evaluating early-stage AI projects today, are there any quantifiable standards? Is PMF still the core metric for screening AI projects?

Liu: I've been doing tech investing for over ten years, starting with software. From what I've seen of AI's development, there are mainly two paradigms: problem-driven and technology-driven. The PMF you mentioned mostly applies to problem-driven scenarios — you start with a product that solves a real problem, with clear feedback and visible results. But this thinking actually extends from our early SaaS investments. Back then, enterprise customers used B2B products, paid annually by credit card subscription, with very low churn. ARR was a meaningful reference. But now with AI tools, especially consumer-facing products, the situation is completely different. For example, I've subscribed to over thirty AI tools myself, and almost all of them are "pay before you use." But after two days, if it doesn't fit, I cancel. Many users' subscription lifecycles last only a month, or even days. ARR in today's AI space, especially in C-end products, has lost its reference value. In Silicon Valley, this is almost consensus — ARR no longer means you've found PMF.

The other type leans more technology-driven, like the large models or underlying infrastructure we've invested in. You could say it's hard to measure with any single metric, but technological progress is visible to the naked eye. So for this category, we don't really look at PMF. We focus more on the team, the people, and some investor intuition and judgment. But for application-type projects, the ones tied to specific scenarios and concrete problems — marketing agents, conversational products, entertainment content — we definitely look at key metrics like retention, activity, conversion. But we don't obsess over ARR. Because in our view, ARR is a metric from the last era. Now we're more often referencing the entire data framework from the mobile internet or SaaS eras to re-evaluate PMF. This is actually a dynamic game: founders keep trying to prove "I've found PMF," while investors keep validating "is the PMF you've found sustainable and robust enough?" This offense-defense battle keeps escalating. Sometimes a year later, a founder comes back with a new set of metrics and says, "Now I have a better way to observe PMF," and we find that pretty interesting. So to summarize: for technology projects, we rely more on judgment; for product-type applications, the old metric system is collapsing, and a lot of the time we're still "going with our gut."

Q: What would a founder have to tell you for you to feel they've achieved PMF?

Liu: I think one thing is especially important. When I meet entrepreneurs now, if you're doing a product-type, problem-solving business, I don't care about your deck. Just sit in front of me, open your product, and walk me through your flow — every product design detail, why you designed it this way, what thinking is behind it. After going through that once, I can basically tell whether they've found PMF. Because a truly refined product has thought behind every detail. Solutions copied from elsewhere can't withstand scrutiny. Of course, I think a lot of Silicon Valley startup products are getting a bit mystical these days too. I've honestly wasted a lot of money buying products just to experience them. If you're still talking big and empty, investors aren't buying it anymore.

Q: Compared to last year, startup valuations are clearly much hotter now. How do you view current valuation levels? Do they really match these AI companies' stage of development?

Liu: I've been in this industry for over ten years. I've seen markets go crazy. VCs "go collectively crazy" about every two years. But I always believe in one logic: as long as you've picked the right track, and that track is genuinely growing, and the team is growing too, then the project isn't expensive in the long run. Let me share a case: HSG invested in OpenAI at $15 billion in 2021. Everyone thought that was insanely expensive. At the time, it was still a pure non-profit, but had already built GPT-2, which was already quite good at conversational text, and had proven the scalability of transformers. Compared to peers, it was definitely very expensive — no AI company was worth $15 billion, none in the entire industry. But if you look four or five years later, it seems cheap. Now OpenAI is a $300 billion company, up 20x. So my observation is: expensive and cheap are relative. For 5Y Capital, our thinking has always been — if your track keeps growing and your team's capabilities keep improving, we don't think there's ever a time when a company is "too expensive."

Q: Despite many feeling the AI application track is already crowded, plenty of new founders still want to enter this industry and make something of it. If someone is preparing to jump into AI entrepreneurship right now, what advice would you give them? Where do you think the real gaps and opportunities still are?

Liu: I don't think AI entrepreneurship in China is overheated at all. If you compare it to US standards, or to the mobile internet wave we lived through, we're only just getting started. I experienced the full 2010 mobile internet wave — from UC Browser, Amap, to input methods and early social products, like the early Kuaishou. One "national-level app" after another emerged, with extremely high user penetration. But today, apart from large models themselves, we haven't seen this level of product landing in AI. So I even think: the current stage of AI entrepreneurship is earlier than 2010.

So when people say it's a bit late to start a company now, I say — it hasn't even started, how can it be late? Information and news are far more abundant today than before. Don't overthink things before you start. The most important thing is to build your product first, find users, iterate. This is the common thread for all companies that survive — just be pragmatic, nothing else. As investors, we're forced to absorb massive amounts of information every year. Personally, I'm very resistant to information overload. A lot of it is just noise and echo chambers that make us anxious, that trap you in FOMO. Just don't look at it.

5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing support from the spiritual to the operational. We believe that if the "crazy" you that others see begins to be believed in, the world will become a different place.

BEIJING · SHANGHAI · SHENZHEN · HONG KONG

WWW.5YCAP.COM