An article containing an internal speech transcript from DeepSeek’s CEO Liang Wenfeng to investors has been going viral online all day, and in the afternoon it was gradually taken down on domestic public platforms.

I saw it in a group friend’s document, but I still can’t stop reading it with interest!

In the 42-page text transcript, he talked about AGI, open source, commercialization, compute power, and domestic chips. He also discussed organizational management and talent. There’s a lot of information, but I think what he kept emphasizing over and over is only one thing: how to increase the probability of ultimately getting AGI built.

For example, what impressed me most is that DeepSeek didn’t choose to attack everywhere, but instead kept making “subtractions.”
Not rushing to seize users, nor building the next super App;
Not taking everything from both the consumer (C) and business (B) sides just to chase revenue;
Even though video generation, 3D, and world models are all hot, they don’t jump on them;
The model could be closed-source to earn more money, but it still chooses open source;
Even though the price could be set higher, it only earns the “reasonable profit” it thinks is appropriate.

This isn’t because he doesn’t want to make money.
On the contrary—because he knows which money can be earned, and which money earned now will instead slow himself down.

1️⃣ Restraint is a strategy!
There’s a line that really stuck with me: Restraint is a strategy.
Many people understand business competition as grabbing users, grabbing entry points, and grabbing revenue—whatever can be obtained, grab it first.

But Liang Wenfeng’s understanding is completely the opposite.
The AI thing is big enough that no company can truly monopolize it. If the future watermelon is big enough, there’s no need to stop for the sake of the sesame seeds in front of you.
You can do things for the C side; you can also earn B-side revenue—but all of these are only side products on the road to AGI. They shouldn’t be turned around into the company’s goal.

This is also the most philosophical part of the entire exchange:
The more you want, the harder it may become to get in the end; the more willing you are to take a little less on your own initiative, the more easily you may reach the end.

Open source follows the same logic.
Many people think open source means giving up a moat, but Liang Wenfeng believes that as long as DeepSeek maintains advantages in cost and efficiency, open sourcing won’t harm the business. Instead, it can attract talent, build an ecosystem, and reduce enemies.
He even judged that if a company wants to take too much of the利益 in the AI era, it will ultimately be defeated by another who is willing to take a little less.
Taking too little means the company can’t survive;
Taking too much, and you’ll be challenged by new competitors.
In the end, the one that can exist long-term is definitely the one earning “reasonable profit.”

So his restraint isn’t being detached or having no desire.
It’s a very strong sense of purpose:
You can take less profit, you can skip hot businesses, you don’t need to抢 short-term users—but the things you truly cannot lose are only two: the main line of AGI, and the stability of the core team.

That’s where restraint is truly sharp.
He’s not without desire—he just concentrates all desire into one thing.

2️⃣ Technical roadmap
Technically, Liang Wenfeng’s route is also very clear:
After language models comes the chain of thought; after the chain of thought comes Agents. After Agents, the real thing that needs to be solved is “continuous learning.”
Now AI capabilities are already quite strong, but it still requires humans to provide complete context. It can’t, like a new employee, enter a company, learn for two months, understand the people here, relationships, rules, and work habits, and then continue to grow.

Once AI has the ability to continuously learn, it can start helping humans research the next generation of AI, forming a loop of “AI accelerating AI.” After that, it’s self-iteration and embodied intelligence.

Of course, he also admits very honestly:
Continuous learning hasn’t truly been solved yet; the whole world is still exploring.

That kind of candor is also hard to come by.
On the one hand, believing AGI will definitely happen; on the other hand, admitting that the specific path is still full of unknowns. Not to secure funding, nor to force investors a fixed timetable.

3️⃣ The gap between China and the U.S.
His judgment on the AI gap between China and the U.S. is also direct:
The biggest gap isn’t talent—it’s compute power and resources.
As long as there remains an order-of-magnitude gap in compute, China can’t comprehensively surpass the U.S.; it can only catch up or even lead in some directions by having higher efficiency, lower costs, and clearer trade-offs.

He believes that in the future, the real gap between large models won’t be some mysterious technology that can never be replicated. It will be three things:
Cost, time, and user experience.
Who can build it earlier, who can provide it at lower cost, and whose product makes people feel more comfortable to use—whoever can do those better will be able to stay.

4️⃣ Long-termism
The biggest inspiration I got from this exchange isn’t that we should go buy any particular AI company. It’s that it made me重新理解 the word “long-termism” in those four characters.
True long-termism isn’t just saying you see farther with your mouth.
It’s whether you have the ability to refuse when short-term interests are truly laid in front of you.

Investment is the same.
Every hot topic makes you want to participate; every market move makes you want to profit; every time prices rise you fear missing out. In the end, your funds, attention, and judgment are all scattered.

A person’s real ability circle may not only be knowing what you’re good at—you also need to know:
Which money doesn’t originally belong to you.

But Liang Wenfeng’s whole model also can’t be simply romanticized.
DeepSeek dares to restrain itself because it has technical efficiency, a team, and funding, and it also believes the market ahead is big enough. Open-sourcing, not setting KPIs, and letting employees explore freely—these are not things that any company can copy and succeed with.

Liang Wenfeng himself also admits that as the company scales up, some departments still need more clearly defined organizational structures; DeepSeek also must rely on APIs and commercial revenue to stay alive.

So what’s truly worth learning isn’t the surface-level “not making money,” “not working overtime,” and “not setting KPIs.”
Instead, it’s first figuring out clearly:
What is absolutely impossible for you to lose, and what you can proactively give up.

For DeepSeek, what it can’t lose is AGI and the core team.
For ordinary people, it might be health, family, cash flow, judgment, and the direction that they genuinely want to invest in long-term.

Today, everyone is discussing how to have more. But Liang Wenfeng spent nearly four hours explaining why he can afford to want so much less.
I think this might be the most powerful part of the entire exchange:
It’s not that you’re厉害 because you seize every opportunity—it’s that even with countless opportunities, you still know where you truly want to go.

Many people think the hardest part of long-termism is persistence.
Actually, it may be even harder: giving up.