For AI Application Companies, Surviving Beats Innovating: Liblib Turns Cash-Flow Positive
An AI application company dismissed as ‘只会抄袭+投流’ — nothing but copying plus buying traffic — has been cash-flow positive since May, with paid advertising accounting for only 4% of revenue. Founder Chen Mian (陈冕): the era of applications has not arrived yet, and the most important goal is to stay alive; between being the first to find PMF and finally winning the market lies an enormous distance.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Chasing high gross margin at the app layer suppresses demand
Outsiders say the company has blown up, but in May and June there was more and more cash in its account and cash flow was positive; outsiders say it lives on buying traffic, but it spends less than 1 million RMB a month on paid acquisition, and paid advertising accounts for only 4% of revenue, because the mainstream video platforms cannot open accounts for AI tools — even wanting to spend the money, there is nowhere to spend it. Chen Mian puts the heavy outside attention down to AI being too hot and the app layer being a talking point, and says he "doesn't deserve" this many comments. On gross margin, his short-term target is explicit: no more than 30%, and positive. The app layer does not control the models, so its greatest asset is users, and chasing high gross margin suppresses demand instead.
— Chen MianAI subscription pricing bets that users never use up their credits
When LibTV launched, its membership discount of 3.9 折 (39% of list price) was read outside as a 3.9 折 discount on the Chat API. Chen Mian says that is a misunderstanding: it was 3.9 折 off the monthly fee when bought as an annual package. The essence of the pricing is a balance between two assumptions, renewal rate and consumption rate — a user buys 1 million credits and uses only 200,000, and the remaining 800,000 is profit. The bet is that "the overwhelming majority of users will not use up their credits in a year," the same logic as a gym betting that annual members will not show up often. He admits that if users burned all their tokens at the original pricing, the company would certainly lose money; but that was already the norm for tool software in the previous era, and this era merely adds the token as a unit of measure.
— Chen MianReal panic came at 6 million left, not at 4,000 yuan
The story of having only 4,000 yuan left in the account in the early days is true, but Chen Mian says the panic really started when there was only 6 million left. He had just come out of a vicious fight with a better-funded rival: because he had thrown all his energy into the fight, he turned down every round of fundraising, would not even add investors on WeChat, and burned 3 to 4 million US dollars in a few months. The first mistake in his own postmortem is that "founders don't look at the books, they assume that once the business works the money will be there." Only after the fight ended did he start meeting investors frantically, and he found that nobody wanted to fund a company on the verge of collapse. That experience set the "trade time for space" undertone of his later strategy, and explains why he is so extremely sensitive to cash flow.
— Chen MianBeing first to PMF and winning the market are worlds apart
On the originality controversy, Chen Mian concedes that LibTV was not the first product to find PMF, but insists that "between being the first to find PMF and finally winning this market lies an enormous distance." He argues that interfaces rarely allow 0-to-1 original innovation: within a vertical, the optimal interface converges, and an engineering innovation holds for only a few days to a week. He is judged harshly, he says, because "we weren't the first to PMF, yet we got to relatively the largest at extremely high speed." The real barrier lies not in the interface but in marketing rhythm, a single point that earns word of mouth, and the ability to amplify fast. He also answered the acquisition rumour: at first he wanted to put a little money in, but the other side ignored him; later he judged that this market was bigger than his main business, so he stepped in himself.
— Chen MianManagement matters least now, and neglecting it is already costing us
Chen Mian quotes the line "management doesn't matter, but organisation and culture matter a lot," then adds one of his own: management matters relatively least in this era, but the cost of neglecting management is now showing. He breaks out two reward models. The first is business-driven — whoever fits steps up, whoever does not steps down — and he has done that well; it is why the company grew. The second he has done badly: what happens to the people who get pulled off? The company did not give them enough respect, or enough feedback to find a new place. He admits he has sometimes been careless with people: early on he casually introduced 10 co-founders to outsiders because "I didn't take titles seriously," while the other side took them as a promise. Only one CTO actually left, because the company was close to collapse, but four technical leads reporting directly to him have gone.
— Chen MianThe near-death moment wasn't running out of money, it was Sora's launch
The moment that came closest to death was not the 4,000 yuan left in the account but seeing Sora launch in January. The team had spent a year exploring the direction of "workflow control," and then a new model simply internalised the workflow — "it went straight through to the next stage." Chen Mian was in Shanghai; the instant he opened his laptop and saw the page he was in tears, and he phoned a shareholder who was then raising the round to say "we're done for." On the flight back from Shanghai to Beijing he worked out the direction — from producing assets to supplying context, a new product composed of agents and workflows. Off the plane he went to his head of marketing, elated, and said "let's build it now." It was also the first time he had cried in front of a shareholder, though he later said that cry was closer to a breakdown than the one when a startup of his failed.
— Chen MianWhile the giants fight over the Central Plains, unify Jiangdong first
Chen Mian uses a Three Kingdoms metaphor to explain the path to surviving independently: when the giants contend for the Central Plains (中原) they have an order of priorities, and Jiangdong (江东) is not their top priority — that gap is the startup's window of time. "When he charges into the Central Plains, your unifying Jiangdong has nothing to do with his charging into the Central Plains." Trade time for space, then space for resources, and finally pour the resources into a barrier — a real barrier has to be watered by time, and "any barrier built at a stroke is not a real barrier." Geographic advantage is dynamic as well: you may start out in Bashu (巴蜀) and only become Jiangdong once you have carved out the natural defence of the Yangtze. Today's Bashu is the "hard road to Shu" that lies between multimodal and text — a general-purpose product does not easily become a professional creation tool. But he also concedes that this road is "not all that thick either."
— Chen MianOnly companies that survive get to talk about innovation and barriers
At the end, Chen Mian sums up his three years of building the company as "driving a car that's running incredibly fast, repairing the car while driving it, with the people in the car coming and going and the scenery outside the window coming and going too" — with no time to stop and say hello or say goodbye. He admits his biggest regret is having no time for farewells; all he can do is grip the wheel and look ahead. Asked about the premise that "if the model companies are the train, you may just be a little car bolted onto the outside of it," he says he is afraid of being thrown off too, but he would rather use the ride to finish building the car. This closes the loop with the opening: the survive he keeps repeating sits above every specific tactic, and is finally a philosophy about time — stay alive first, and only then do you earn the right to talk about anything else.
— Chen MianIn their own words · checked verbatim
If I'm not strongly aggressive, how am I supposed to be any good? And if I'm not any good, how do I survive?
不侵略性强，我怎么厉害呢？我不厉害，我怎么活下来呢？
Chen Mian20:03
Come on, innovation has a cost, it has a price. If you innovate something and then can't hold on to it, can't do it well, what's the point of innovating it?
拜托创新是有成本的，是有代价的。你如果创新了一个东西，你都把它守不住，把它做不好，你创新它的意义是啥。
Chen Mian58:25
When he charges into the Central Plains, your unifying Jiangdong has nothing to do with his charging into the Central Plains.
当他逐入中原的时候，你一统江东跟他逐入中原没有关系。
Chen Mian1:31:40
So fundamentally it comes down to whether you believe that road exists or don't believe that road exists. I believe that road exists.
所以本质上是你相信那条路存在，还是不相信那条路存在。我相信那条路存在
Chen Mian1:38:03
I think a startup is like driving a car that's running incredibly fast, and you're repairing the car while you drive it, and the people in the car come and go, and the scenery outside the window comes and goes, but you have no time to stop and say hello and say goodbye to these people, and no time to really look at the scenery outside. All you can do is grip the wheel.
我觉得创业就是开了一台。超级快在跑的车，然后你边开车边修车，然后车上的人来来去去，窗外的风景也来来去去，但你没有时间去停下来跟这些人say hello和say goodbye，然后也没有时间好好看窗外的风景。你能做的事情就是握紧方向盘。
Chen Mian2:06:34
Figures
| Company valuation and amount raised | USD 2 billion valuation, USD 300 million raised | 1:00 |
| Cash flow in May-June | Positive, with more and more cash in the account | 5:14 |
| Share of revenue from paid advertising | About 4% | 5:14 |
| Monthly paid-acquisition spend | Under 1 million RMB | 5:14 |
| Short-term gross margin target | Above 0% and no more than 30% | 8:28 |
| Lowest cash in the account | Under 4,000 RMB | 22:13 |
| Technical leads who left | 4 left; only 1 CTO actually resigned | 48:35 |
| lib's contribution to revenue | More than half | 47:22 |
| LibTV peak daily revenue | USD 1 million | 1:26:23 |
| Ceiling on global annual video-creation revenue | About USD 100 billion | 1:27:26 |
Glossary
- PMF / product-market fit
- The state in which a product exactly meets a demand that is both real and scalable.
- Reward model
- The episode borrows the reinforcement-learning term as a metaphor for the incentive and feedback mechanisms inside a management system.
- LTV / lifetime value
- The total revenue one user contributes across their entire life as a user.
- GTM / go-to-market
- The playbook and timing for taking a new product or a new market from definition to execution.
- AI native
- Talent naturally suited to how work is done in the AI era, with a high density of capability.
- ARR / annual recurring revenue
- The recurring revenue of a subscription business stated on an annualised basis; the core metric in valuation narratives.
How to listen
AI application founders, investors watching app-layer valuations, and product leads running growth and management in the shadow of the model vendors.
1:46-1:54, on his internet-addicted teenage years and his time in theatre, can be skipped; if you only care about the strategic path to survival, go straight to the Jiangdong metaphor at 1:28.