The Logic of Small Bets for Big Returns Has Failed. We're Now Betting on Jurassic Rodents
Dollar VC has had both its fundraising and exit ends cut off, and the leverage story — two or three hits out of ten checks — no longer holds. CMC Capital's answer: either invest in the present, where you can cash out in a year or two, or invest in the future, seven to ten years out. The middle ground is the hardest.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Dividend strategies are really debt in equity's clothing, and hard to do
One camp, represented by Zhu Xiaohu, has figured it out: the old VC logic of small bets for big returns no longer holds, so they pivot to cash flow — investing in companies with revenue, profit and the ability to pay dividends, where the exit doesn't even have to be an IPO. But Mengqiu points out this becomes debt in equity's clothing, and such deals are hard to do — you have to run into a company that already has good revenue and cash flow and still wants to raise. Why would a company that genuinely makes money dilute its equity to raise? So you can only count on serendipity. She thinks it's a clear enough strategy, neither right nor wrong, but fundamentally a phase — because the VC financial product was never built to do this from day one.
— Meng QiuRobot bodies are too expensive, so everyone can only spread their bets
Mengqiu's personal judgment is that robotics is the direction where AI meets the physical world. Whether or not this generation of technology can truly understand the physical world, this generation of AI's real value must be expressed in combination with the physical world to unlock productivity. But robot body companies are extremely expensive from the start — a single deal takes too much money, so no one can invest alone; each person's money only stretches to two or three companies, and in the end maybe two or three survive, or even just one — a one-half or one-third chance. She endorses Wang Haodong's strategy of investing in all three, because Jinqiao Robotics invested in Li Auto and holds a clear deployment scenario in hand; the logic of looking at upstream and downstream materials, batteries and components along the new-energy vehicle chain can be replicated in robotics.
— Meng QiuThe earlier a shareholder is, the harder it is to exit, and the less bargaining power they have
From her own experience exiting projects, Mengqiu draws a counterintuitive conclusion: the earlier you get in, the harder it is to exit. If you want to negotiate terms like a buyback, you have to negotiate it in every subsequent round, and since later investors put in more money, early shareholders actually have no bargaining power. So when CMC Capital first started looking at this generation of AI technology a year and a half ago, it drafted its own research framework and quickly concluded: don't look at large models, and don't favor anyone other than the big platforms building large models. Her plain theory: if OpenAI hadn't had Microsoft, it might have died on the road long ago; even now OpenAI's own ability to generate profit is far behind its ability to spend, and that's with Microsoft already handing it a massive deployment scenario.
— Meng QiuThe fool's window the big platforms used to leave open is gone
Mengqiu argues that many startup teams' opportunities came from a fool's window — the big platforms weren't awake yet. But with this generation of generative AI, the big platforms were awake from day one. They are rich, their thinking is not behind at all, and they have their own application scenarios and data feedback loops. So third-party startup teams building foundation models don't have much of a chance. She originally thought multimodal large models had a time window, that the big platforms were too busy to do it and startups could at least sell to everyone. But after Sora came out this spring, she felt the concept of an independent multimodal large model no longer holds, because all foundation models are fusing more modalities, and an independent multimodal model essentially still needs to call on the comprehension ability of a language model.
— Meng QiuOnly pushing something to the extreme can succeed; the middle ground is hardest
Mengqiu sums up her own investment philosophy: push something to the extreme — clear enough, crazy enough, unswervingly in one direction — and it can succeed; the middle ground is very hard, and no one has even succeeded in the middle ground. This explains why she bets on both ends — either invest in the present, where returns are immediate and money can be made next year or the year after, or invest in the future, seven to ten years out; she won't invest in the transitional middle. She applies the same framework to the AI application layer: voice is an underrated link, because everyone naturally assumes the voice market is small, pays it little attention, and is still tinkering with text, images and video. But when AIGC reaches real scenarios in the future, voice will matter — it needs to reach the point where, like in the film Her, you immediately feel a real person is speaking.
— Meng QiuYou now make money on consensus, not on non-consensus
Mengqiu points to a key shift: VC used to make money on non-consensus, now it has to make money on consensus. Because there is limited money in the market, everyone has to believe in one direction before everyone is willing to keep putting money in. When money is abundant, chaos is allowed; when money is scarce, it isn't. So she redefines VC as No Venture — we are not Venture, we want not to take risks, to avoid risk as much as possible. On the future end, you invest in what everyone believes, in consensus, which is relatively lower risk; on the other end, you invest in what can be monetized immediately. Two extremes, no middle ground.
— Meng QiuBe a Jurassic-era rodent and survive first
Mengqiu uses a metaphor: when the asteroid hit Earth, the dinosaurs all died, but mammals, which had been very small compared with reptiles and were the weak group, survived the catastrophe. She says she is not saying today's giants will die, but overall the giants will be hit harder. So CMC Capital wants to be a Jurassic-era rodent — a mouse, living in the gutter, with a smaller range and less food, but not lying flat. She turned down an offer from a local government guidance fund, because such funds come with many requirements — how to define the reciprocal investment, and the investment direction has to be set in advance. In the current situation, she feels it's better to have more control over the money in her own hands.
— Meng QiuThe new VC logic: PE equals 10, and rounds no longer matter
Mengqiu says primary-market valuations are now very clear: PE equals 10, because the secondary market is only giving 15; when the secondary market gives 30, PE equals 15. Rounds have also become chaotic, because the VC logic no longer holds — it's no longer a round-by-round game of passing the parcel to push valuations up. Projects that can genuinely make money within a year or two don't lack money themselves; they may have raised only one round in their history, or never raised at all, and just want to bring in some funds for endorsement. Such projects are very sparse, but CMC Capital invests in few projects anyway — it may look at a hundred and pick one. The criteria for the present are more diversified; the criteria for the future are still long-term value, seven to ten years.
— Meng QiuIn their own words · checked verbatim
In the past, our VC logic of small bets for big returns completely no longer holds, so everyone chases cash flow — and the question is, investors also chase cash flow, so you invest in companies with cash flow, with revenue, with profit.
过去我们VCP以小博大的逻辑 是完全不成立了 所以大家追求现金流嘛 那一问是 投资人也追求现金流 就投有现金流 有收入 有利润的公司
Meng Qiu4:37
Many startup teams' opportunities came from a fool's window — the big platforms weren't awake yet. But with this generation, the big platforms were awake from day one.
很多创业团队的机会来自于有傻瓜窗口 就是大厂没睡醒 但是这一代深沉是原来从第一天起大厂就行了
Meng Qiu17:06
Push something to the extreme and you can succeed — clear enough, crazy enough, unswervingly in one direction — you can succeed. I think the middle ground is very hard.
一件事情 push到极致 你都有可能成功 就足够明确 足够crazy 足够的朝一个方向 坚定不移的走 也都有可能成功 我觉得中间地带 是非常难的
Meng Qiu28:15
It's not non-consensus — you have to win money through consensus, because there is limited money in the market; everyone has to believe in one direction before everyone is willing to keep putting money in.
不是非共识 要通过共识争到钱 因为市场上的钱有限了 你必须是大家都去相信一个方向 大家才都愿意持续去投钱
Meng Qiu32:57
You're doing the Jurassic thing, being a mouse — you're already a mouse, how much worse can you get? You won't get worse.
你就是做Nature的东西 做老鼠 你都已经做老鼠了 你还能怎么更差呢 你不会更差
Meng Qiu44:20
Now primary-market valuations are very clear: PE equals 10. Yes, PE equals 10, because the secondary market is only giving 15.
现在现在一级市场估值 非常清楚 PE等于10 对 PE等于10 因为二级市场现在才给到15
Meng Qiu1:02:05
Figures
| Number of projects CMC Capital invested in last year | 8 | 38:18 |
| Most projects CMC Capital invested in a single year, historically | 16 | 38:18 |
| Fewest projects CMC Capital invested in a single year before last year | 15 | 38:18 |
| CMC Capital's USD-to-RMB ratio | 1:1 | 34:16 |
| Shrinkage of VC industry headcount | at least one third | 38:18 |
| Primary-market PE valuation | PE equals 10 | 1:02:05 |
| Secondary-market PE valuation | 15 | 1:02:05 |
| Number of AI projects CMC Capital has looked at | at least a hundred-plus | 22:10 |
Glossary
- debt in equity's clothing
- Nominal equity investment that in fact carries a fixed return or buyback — essentially debt.
- check size
- The amount of a single check a VC writes; early-stage funds usually write smaller checks.
- power law
- VC returns are highly concentrated in a very small number of projects, while most projects go to zero.
- foundation model
- A general-purpose model pretrained on large-scale data that can be adapted to many downstream tasks.
- AI for science
- Using AI and high-performance computing to simulate the microscopic world and accelerate the discovery of new drugs and materials.
- AI for coding
- Using AI to assist or automatically generate code; in China this often requires private deployment.
How to listen
Investors tracking the real survival conditions of China's primary market, hard-tech and AI founders thinking about the pacing of their next raise, and practitioners who want to understand how the rules of the VC game are changing.
The chat after 1:13:50 about the human comedy of a downturn — low information density.