No Rush for Commercialization: Consensus Decision-Making and China's Role in the Global Division of Labor

I think it's not time yet. This is our judgment, and at least our past experience supports it. Over the past three years, if you came to me at any point and talked about commercialization paths or product lines, it would have been a waste of time because you cannot predict or foresee the future. You can foresee very little.

Checking the time, I'm not sure if everyone wants to take a break now? Grab something to eat? Or should we continue? If no one objects, I'll keep going.

Conducting world-leading AGI research and commercializing at the appropriate time. I think we have always been doing commercialization. We just haven't made it our goal. We aim for AGI, but we have always been doing commercialization, which is why we have consumer-side users and enterprise-side revenue.

From historical experience, this strategy has been successful. I think the point where we completely shift to commercialization is still far away. So the biggest focus is on extending technology and developing next-generation tech to better solve current problems. The return on these efforts, in my personal view, in the foreseeable future, makes it too early to focus on products at any point.

So this is part of our restraint. I hope these business opportunities can be taken up by others. We want these opportunities-how to use this AI-to be pursued by the whole society and all our partners, sharing the benefits together, not for us to hog them all. That's impossible. And we don't have that much energy or that many people in our organization to do this.

Regarding partners, our financing has been carefully selected. As for the specifics of how to cooperate, first, I think our interests are relatively aligned. Those who share the most aligned interests with us, have the least hostility toward us, or most want us to succeed. Not everyone wants us to succeed, because we have hurt the interests of many others.

The process and decision-making mechanism for major company strategy and technology business decisions. Our company is built on consensus. I don't decide everything on my own; I seek consensus. My authority and influence within the company are based on consensus.

For example, if I want to do something, I first look at what the consensus is-whether everyone wants to do it. Then I might provide some guidance or inclination, but the effect of that guidance is limited, very limited. It still needs to be based on consensus. This decision-making mechanism is essentially a consensus-seeking mechanism. It's not that I can push something through; it only works if there is consensus, and then I will push it forward.

Currently, most of my energy is on DeepSeek. Let's take a five-minute break. I'll write quickly. Everyone can unmute and give me some feedback. We're fine here. Let's take five minutes.

The final performance gaps between models should be a comprehensive matter. Comparing model performance must be done at the same cost to be meaningful. It's like comparing two cars-you compare them at the same price point. The difference between a good model and a bad one shouldn't be in a specific component; it should be overall.

Anthropic surpassing OpenAI now-is that long-term? I don't think so. It's definitely temporary. OpenAI and Google will likely alternate in leading. In fact, Anthropic's advantage in Code Agent isn't that huge; it's not like they are crushing OpenAI.

Within our company, about half the people normally think OpenAI is better. Actually, Anthropic has a first-mover advantage, but that advantage should quickly disappear-it's not something they can hold long-term. All three are very strong, and among them, the one with the highest efficiency spends the least cost, burns the least money.

When the global AI division of labor is dominated, Chinese companies are likely to play the role of having the largest output. Generally speaking, our production capacity is the largest, including chips-our chip production capacity is probably the largest, and we have the most electricity. So our AI might ultimately be among the three major players.

Chinese people will make the product the cheapest, and in terms of performance, after all, for many products, Chinese-made and American-made products don't differ much. In the future, AI might be the same, but AI made in China will probably be cheaper. This cheapness might be systematically low, similar to how Chinese services are cheaper in other industries.

When I approach tasks, my habitual thinking is: what gives the greatest return at this moment? If I think making products gives the greatest return, I'll do products; if I think achieving AGI first gives the greatest return, then I'll do AGI. Clearly, I think making products is not the greatest return now.

If you're talking about domestic GPU compatibility, there is actually a historic opportunity now for domestic chips. The previous difficulty with domestic chip compatibility was poor ecosystem. After buying the card, you couldn't effectively use it because it didn't have Nvidia's ecosystem. So Nvidia's moat is very strong.

But this is changing. Nvidia's CUDA moat is quickly eroding. The reasons for the rapid erosion may be threefold. One is that with AI now, building an ecosystem is much easier than before, because AI can write code. We can use AI to build this ecosystem, and then we can build an ecosystem identical to NVIDIA's. First, because of AI; second, because of some new technologies. For example, our team developed a technology called TileLang, which is a high-level language. Using this high-level language to write CUDA operators, we can quickly rewrite NVIDIA's entire ecosystem. Combined with AI, there don't seem to be any obstacles. But it's not finished yet, not completed. However, this technical route seems to have no obstacles.