Liang Wenfeng's goal is very singular.

All AI companies are increasingly becoming similar, while Liang Wenfeng is growing more and more distinct. 

Others raise financing to craft a grander business narrative. Right after securing a new round of funding, Liang Wenfeng spent nearly four hours telling his investors: We are not here to make money.

Others are scrambling to build super apps, seize traffic entrances, and compete for higher DAU. He says: We will not aim to become the next ByteDance or Tencent.

Others are researching how to achieve commercialization. He says: The potential benefits of AI are so enormous that we don't need to consider commercialization right now.

Others are competing fiercely in video generation, AI agents, robotics, and AI smartphones. He says: Those are not the main path toward AGI.

Others are sparing no effort to recruit the most expensive talents. He says: We are just a group of ordinary people.

After learning about Liang Wenfeng, my first reaction was not shock, but pure confusion. 

Nearly every single word Liang Wenfeng says goes against the mainstream consensus of today's AI industry.

The oddities do not end there. All these "counterintuitive moves", when traced back to their roots, stem from one single reason. 

I. The Biggest Illusion in the AI Industry Is the Ever-Growing Number of Objectives

Let's put DeepSeek aside for a moment and look at what the entire industry is busy with. 

Financing, IPOs, valuations, model rankings, AI agents, world models, robotics, super traffic entrances... 

Every company is chasing multiple goals simultaneously. 

This has led to a very interesting misalignment. 

All companies claim they are working on AGI, but what they actually discuss in daily meetings are valuations, financing, commercialization, next-round growth, and capital market performance. 

Fewer and fewer people are genuinely putting AGI on the table for in-depth discussions.

OpenAI, Anthropic, Kimi, Zhipu AI, MiniMax, Meta... each one is pushing the competition further, dragging the entire industry into a capital-driven cycle. When there are too many objectives, the path becomes fragmented, actions slow down, and every choice has to be made through trade-offs. 

There is only one exception. 

DeepSeek has never changed its core objective from start to finish. 

II. The 4-Hour Session Actually Revolves Around One Single Core Point

After going through the full transcript of the 4-hour investor exchange meeting. 

After stripping away redundant remarks, polite formalities, and discussions on technical details, the one single point Liang Wenfeng repeatedly emphasizes is: 

How to maximize the probability of successfully achieving AGI.

No talk of profit maximization. No talk of growth maximization. No talk of user base maximization. No talk of valuation maximization. 

Everything they do is focused on increasing the success probability of AGI. 

All the moves that seemed incomprehensible to the outside world now suddenly make perfect sense. 

Why not build consumer-facing apps? Because it does not increase the probability of achieving AGI.

Why not work on video generation, robotics, or AI smartphones? Because it does not increase the probability of achieving AGI.

Why insist on open-sourcing? Because AGI can only be achieved through collective efforts from more people.

Why exercise restraint on commercialization? Because if you take too much market share, you will eventually be defeated by competitors who take less.

Why raise financing but not rush for an IPO? Because going public requires telling endless growth stories, which will distort your judgment.

Why price APIs based on a 10-month payback period for hardware costs? Because if you charge more, you will squeeze out peers, and the entire AGI ecosystem will fail to thrive.

Liang Wenfeng put it very bluntly himself: 

The AI industry is large enough that it could eventually account for 10% of humanity's total GDP. Even a tiny slice of that pie is already more than enough for us.

To put it in an even more straightforward way. 

Anything that does not help increase the probability of achieving AGI is irrelevant. 

Revenue does not matter. Traffic and rankings do not matter. Scale and superficial prestige do not matter either. 

This is DeepSeek's true operational philosophy. 

III. Everyone Else Is Managing a Company, While He Is Managing a Vision

There is a passage in the transcript that defies common business sense. 

Liang Wenfeng says DeepSeek has no KPIs, no performance reviews, no rigid rules, and no formally written vision statement. 

Running a company does not depend on rules and regulations, but on a shared vision.

Many people's first reaction is that this is nothing more than a nice-sounding marketing line. But upon closer inspection, you will realize he is not just making empty talk, let alone being hypocritical. 

What he truly manages has never been his employees, but the core objective itself.

Half of the engineers are given full freedom to explore on their own. They can work on whatever they want. No pre-set requirements, no mandatory progress reports, no OKRs, no performance interviews. 

The other half of their work goes to "formal" corporate matters. Liang Wenfeng drew a clear line for himself: Official corporate affairs should never take up more than half of employees' working time.

You can barely find a second AI company operating this way in today's market. 

Large tech giants are obsessed with OKRs, organizational restructuring, bi-monthly reviews, quarterly performance reports, and performance calibration. Startups are also copying these practices. 

Everyone else is using management frameworks to chase speed. DeepSeek does the opposite, gaining speed by minimizing unnecessary management.

This scenario might feel familiar to you. 

More than a decade ago, Google incorporated the 20% Time policy into its engineer culture. Today, Google itself has most likely forgotten about this practice. 

Everyone who has ever built a truly great product will eventually arrive at the same conclusion: 

Processes can keep a company running, but they can never support the making of something truly great.

Truly great achievements have always relied on only two things: an extremely singular core objective, and people who genuinely believe in that objective.

DeepSeek has simply reclaimed this underlying rule that the entire industry has forgotten. 

IV. He Is Not an Idealist, but an Extreme Realist

At this point, people would easily jump to conclusions, putting Liang Wenfeng on a pedestal and labeling him as a sentiment-driven idealist. 

It is time to pour a bucket of cold water on that idea. He is not naive at all. 

APIs can generate revenue. Serving enterprise clients can generate revenue. They can also go public in the future. He has never rejected any of these paths. He simply says it is not the right time to do them now. 

He set the API price at a specific level: Priced to recover hardware costs within 10 months, which corresponds to roughly six times profit margin.

This number is very thought-provoking. 

From a pure business perspective, a six-times profit margin is relatively low. If they truly wanted to maximize profits, they could raise the price to 20 times, 30 times, or even 100 times the cost. 

From a strategic perspective, a six-times profit margin hits the sweet spot: it is enough to keep the company running, pay the team well, and continuously fund research and development; at the same time, it is low enough to make third-party deployment unprofitable, naturally drawing the entire open-source community to gather around them. 

He did the math himself, and explained it very clearly. 

If the entire AI market eventually accounts for 20% of humanity's total GDP, it is theoretically possible for one company to capture 5% of that share. But that company will definitely be defeated by another competitor who is only targeting 1% of the market. And that competitor will then be defeated by a third party who is only targeting 0.1% of the market. 

Those who take more will be defeated by those who take less. Those who take less will then be defeated by those who take even less.

Even before you actually make any money, if your vision is to grab as much as possible, you have already lost the game before it even starts. 

Restraint is long-termism, and restraint also filters out competitors.

Once you set a very low upper limit on your profit target, most other players will automatically abandon this track. On their balance sheets, this business simply does not make financial sense. 

As a result, DeepSeek has obtained something no one else can get: Tranquility.

Tranquility to train models, tranquility to conduct research, tranquility to push AGI forward step by step. 

To outsiders, it looks like a laid-back "Buddha-style" approach, but internally it is all carefully calculated precision. 

V. He Believes in AGI, But Holds No Illusions

During the 4-hour conversation, the phrase Liang Wenfeng repeated the most is not AGI, but: We have no special advantages, we are just ordinary people.

His remarks are packed with extreme, almost self-deprecating honesty: 

We have limited resources. We do not have enough GPUs. Our total computing power is one-tenth of what our counterparts in the US have. We are roughly one year behind the US. Our team is just a group of randomly gathered ordinary people. I myself am just a college graduate, not even from a top-tier university. 

Compare this with the buzzwords flooding today's AI industry: disruption, revolution, redefining the world, the next traffic entrance, superintelligence, super apps, super AI agents. 

The contrast could not be more stark. 

Everyone else is busy building myths around AI. Only DeepSeek keeps tearing these myths down.

Others claim AI will change everything overnight. He says the next generation of models will first solve the problem of continuous learning.

Others say they want to build the next universal traffic entrance. He says we should first get coding capabilities right.

Others talk about super apps. He says that is unnecessary.

Others claim they want to monopolize the entire market. He says OpenAI thought from day one that it could monopolize the whole world, but it will definitely eventually face challenges from players who are willing to take far less share. 

He even laid out the AGI roadmap in a tone so calm that it almost feels anticlimactic: 

First solve the problem of continuous learning. Then reach the singularity where models can self-iterate. Then move on to embodied intelligence. This sequence is the most labor-efficient path. If you try to do it in reverse, you will end up exhausting yourself for no good reason.

Someone who genuinely believes in AGI has chosen the path that requires the least unnecessary effort. 

This is the most honest form of faith. 

People who truly believe in something will never force their way through obstacles blindly. They will calculate every step carefully, to maximize the chance that their goal will eventually come true. 

Everyone else is creating illusions around AI. Liang Wenfeng keeps stepping back, all the way back to the first principles. 

What truly blows my mind has nothing to do with their latest financing, nothing to do with their $54.3 billion valuation, and nothing to do with the fact that Tencent and CATL are co-investors. 

It is the realization that in today's entire AI industry, Liang Wenfeng is likely the only person left who is still seriously discussing AGI itself. 

VI. The Person He Most Resembles

After reading through all of Liang Wenfeng's remarks, the first person that popped into my mind was not any founder of a well-known AI giant, but Zhang Xiaolong from more than a decade ago. 

Not because both of them practice restraint, which is just a superficial trait.

The true connection between the two lies in the exact same capability they share: At the noisiest moment of the industry, they keep their eyes locked on the one and only core problem.

Back then, Zhang Xiaolong's core question was: How to make connections between people simpler. 

Today, Liang Wenfeng's core question is: How to maximize the probability of successfully achieving AGI. 

Once your core question is singular enough, you will never hesitate over any choices. 

Do we need to build a super app? No. Do we need to work on video generation? No. Do we need to grab consumer-side traffic? No. Do we need to chase enterprise orders? No. Do we need to raise prices? No. Do we need to do vertical integration? No. Do we need to develop our own chips? We will avoid that as long as we can... 

To outsiders, this looks like a whole series of counterintuitive, contrarian moves. 

To them, they are just continuously doing the exact same one thing. 

Words Beyond the Layout:

While writing this article, I kept thinking: If this article can only leave one single answer for readers, what should it be? 

Most likely, it is the answer to why DeepSeek looks so different from every other AI company out there? 

The answer has nothing to do with open-sourcing. Nothing to do with their refusal to rush for profits. Nothing to do with how well they tell their financing story.

The real reason can be summed up in one single sentence: While all other players are greedily chasing multiple objectives, Liang Wenfeng has always kept his eyes fixed on that one single goal. 

A goal that is distant enough, grand enough, and singular enough. So grand that you cannot see the end of it, so singular that you can easily abandon all other distracting thoughts at any time.

That is why every move this company makes looks like it is moving against the tide. 

But it is not really moving against the tide at all. 

It is just that on this long path, he is the only one left who keeps walking forward. 

This article is from WeChat Official Account "Beyond the Layout", author: Huahua, published with authorization from 36Kr.