Only corrected typos, incorrect terminology, and logical inconsistencies from the original transcript.
Meeting Time: May 20
Meeting Duration: 4 hours
Meeting Purpose: DeepSeek Founder Liang Wenfeng's Investor Exchange Meeting
When we first started this company, our original intention was not about how much money we would make, going to the capital markets, or listing. That was not our goal.
The first few dozen people never thought that way. If they did, they wouldn't have come. Overall, we are doing this with a great deal of goodwill toward the world, believing it is useful for humanity—something beyond money.
Of course, later on, when the potential interests became very large, other temptations arose, but that is another matter. Our original intention, our vision, and the vision we maintain today is not based on maximizing commercial interests. I think this is key.
About twenty years ago, the manager I admired most was Jack Welch, the former CEO of GE. Looking back now, most of what he said might be wrong, but he was right about one thing: the most important thing for a company is its vision.
Managing a large company doesn't rely on regulations; it relies on vision. Vision isn't a slogan on the wall; it's how you act, not what you say. It's how you actually operate.
So, how do we manage and organize so many people? Actually, we don't have a formal organization; we are vision-driven. We are organized by a vision.
This has pros and cons. In the future, we will try to leverage the strengths and avoid the weaknesses, but this is our characteristic. We don't operate through KPIs or assessments; only vision.
This vision isn't even written down. It exists in our methods and our attitude toward the world. Everyone in the company might understand it differently, but the general direction is consistent.
I believe it's about having great goodwill toward the world and wanting to accomplish something. That is how we are organized.
I'll speak first, and then everyone can ask questions. I will frame everything around this vision. This vision is real, not fabricated. We truly think and act this way; otherwise, you couldn't explain many of our actions.
Why are we so persistent about open source? Because the vision itself demands it. Without this vision, you can't organize people.
For example, Zhipu also open-sources, but their open-sourcing feels forced, as if it's not their original intent. For us, it is our original intent.
We thought very clearly about open source from the beginning. First, the vision; second, we believe open source is beneficial for making AI commercially successful.
This sounds contradictory because, historically, open source and commercialization have been in conflict. But I think AI is different. Historically, a software company's market might be a few billion dollars; if you open-source, it disappears, leaving only millions.
But AI is large enough that it might eventually account for 10% of human GDP. You cannot monopolize such a thing. You must share it with others, or you won't survive.
If we tried to monopolize the benefits, we would be discarded by history. This is an objective law, a historical perspective.
It's not that if I don't open-source, I can monopolize the market. That doesn't fit objective facts. You will face resistance and other methods will stop you. In this case, traditional commercial thinking isn't necessary. You need a mechanism to ensure your own benefits are limited to succeed. It requires restraint.
If we want to succeed in AI, we must be restrained. You can't think that a percentage of global or Chinese GDP belongs to you. The more you think that way, the less likely you are to succeed.
So we felt the need for restraint from the start. The more you restrain yourself, the more likely you are to succeed. This is a macro commercial consideration.
At least, this is how I truly think. We don't have many other advantages. We aren't richer or better staffed than other companies. When we started two years ago, we had no money, few chips, and no fame. We were just a group of ordinary people.
[00:11:49]
We are truly ordinary people. If there's a narrative I like, it's ordinary people doing extraordinary things, not geniuses doing extraordinary things. This is tied to our restraint and our vision.
Does open source conflict with commercialization? In AI, if you aren't restrained, you won't succeed. Open source is part of that restraint. Our restraint is also shown in other areas. Generally, we don't even need to debate whether to open-source or be restrained.
Restraint makes success easier. At least so far, this has been proven. Otherwise, you can't explain why we succeeded with such a low starting point and few resources. I'm just a university graduate, and not from a top-tier school.
This restraint is part of our vision. The interests in AI are too large. If we are restrained and succeed, the remaining benefits will still be massive. You only need a small slice. We only seek reasonable profit, not profit maximization. Our API pricing reflects this: we aim to recover equipment costs in ten months. That is a reasonable profit.
Our V3.2 Flash and others follow this standard. It's not profit-maximizing. If it were, we would set prices higher because demand in this range is inelastic. Doubling the price wouldn't significantly drop consumption, but it would double revenue.
I'll tell you a story. With DeepSeek, we initially set the price high because we feared too much demand. The team wasn't happy. When I lowered it to one-fourth, everyone was happy.
This is our true thought. We want the tool to be useful, not just to make the most money. When we cut prices, people in the company chat were cheering. That is our consensus.
This is unique. For competitors, price cuts aren't good for their ARR. But for us, ten months to recover costs is satisfactory. It's a win-win for the company and society.
Someone commented that ten months is still too high. There is indeed room for further price cuts as models optimize. But we can achieve this cost while others like Alibaba or Tencent, without our optimizations, likely have costs several times higher.
We don't cut further right now because demand is inelastic; everyone can already afford it. Lowering it further doesn't add much social value. Restraint is a strategy. It's a long-term play. Open source is a form of giving back that builds internal cohesion and social goodwill, increasing our probability of achieving AGI.
I don't doubt AGI's commercial value. Therefore, I prioritize the probability of success over market share. This restraint also meant we didn't chase the 'Super App' dream or fight ByteDance for users last year. We focused on service, not monetization. We ignored the 'sesame seeds' to wait for the 'watermelon' of AGI.
Even if we don't have a clear business model for AGI yet, the opportunity is so large that a path will exist. Our current C-end and B-end revenues are just byproducts of the road to AGI. If our API revenue reaches hundreds of millions or a billion dollars, it will cover our R&D, but it's not our primary goal.
Regarding open source: we will continue to open-source our strongest models. I see no benefit in being closed-source. Even if you open-source everything, the barrier to entry and cost-optimization is very high. This is the 'sweet spot' for a company of our size.
Open source doesn't hurt my revenue if I only aim for a 6x profit (10-month payback). It only hurts if you want 100x profit. This is sustainable. We even want to help the community deploy our models correctly.
Our Goal Should Be AGI
AGI is our long-term vision. The roadmap is clear. Current AI excels if given perfect context and instructions, but it lacks 'continuous learning.' A human employee takes two months to learn the environment; AI can't do that yet without massive context injection.
AI development is a ladder. Last year was CoT (Chain of Thought). This year is Agents. Each step builds on the last. After Agents, the next bottleneck is continuous learning—making models learn over time like humans.
Once a model can learn continuously, we reach a singularity where it can iterate on itself and develop its own next version. This isn't a sudden jump but a gradual process. After that comes embodied AI (robotics).
This roadmap is the 'easiest' path. If we solve continuous learning and self-iteration first, embodied AI becomes easy because the AI helps build it. We don't want the 'hard mode' of doing embodied AI first.
While others fight for C-end traffic or B-end revenue, we focus on the AGI roadmap. This is a 'dimensional strike' (降维打击). By aiming for a higher technical goal, the lower-level applications become easy byproducts. We didn't even try to keep users last year, but they wouldn't leave because the technology was superior.
The Core Interest is Team Stability
Our only non-negotiable core interest is team stability. If the team stays, we will succeed in AGI. Money and resources are secondary. Recent financing has helped secure this through options. Our talent turnover is lower than peers because people want to be in an environment that can actually achieve AGI.
We don't want to be enemies with big tech; we want to empower them. We focus only on the 'main line' of AGI (GPT, CoT, Agents). We don't do 3D, video generation, or world models yet because they aren't on the critical path to intelligence. Video generation is a good business, but it's not the path to AGI.
Our gap with the US is purely resources (chips). We have about 20,000 H-equivalent GPUs, mostly arrived recently. We are aggressively expanding. If I could spend all our funding on chips in six months, I would. It's better than money in the bank. Talent-wise, there is no gap; the smartest Chinese engineers are both in China and abroad.
We aren't trying to compete with the US at the 800B parameter scale yet because we can't afford the research cycles. We are perfecting the 50B-150B scale first. The ultimate gap in AI will be cost, time, and user experience.
On Commercialization
We are always commercializing, but it's not the goal. Focusing on product lines now is a waste of time because the field moves too fast. We want others to take the commercial opportunities while we provide the engine. Our investors were chosen because their interests align with ours.
Our management is based on consensus, not top-down mandates. I seek consensus before pushing anything. We maintain a 'research culture' where formal work shouldn't exceed 50% of a researcher's time, allowing for free exploration. We don't encourage overtime because research needs a relaxed environment and we are highly focused on fewer tasks.
Q&A Highlights
On Domestic Chips: The CUDA moat is being dismantled by AI-assisted coding and new languages like our TileLang. In a year, the perception that domestic chips lack an ecosystem will change. We work closely with Huawei. Their 950 clusters can effectively replace Nvidia's GB200/300 in performance, even if the price is higher.
On Scaling: We believe in scaling. We haven't hit a wall; we just hit a resource limit. We will continue to push the limits as we get more chips.
On the Future: The next generation of models must have continuous learning. Our goal is for the AI to first be useful to us in developing the next AI. If it helps us reach AGI faster, it's a success.
