Chips, open-source models, and 50 trillion: Cai Chongxin revisits Alibaba again

Author: Su Yang, Tencent Technology

At this year's VivaTech conference, Alibaba Chairman Daniel Zhang systematically elaborated on Alibaba's long-term AI vision during a "fireside chat," marking his second public review of Alibaba after the Yale University summit at the end of May.

"From a macro perspective, we are fully committed to AI, and the logic is very simple."

Daniel Zhang stated that the global GDP exceeds $100 trillion, with at least half coming from human intelligence and productivity contributions. "This $50 trillion is the total AI market, much larger than any company's IT budget and far bigger than the software market."

Everyone is talking about All in AI, and Daniel Zhang is no exception. He summarized Alibaba's strategic layout as covering all fields except the energy sector, including chips, cloud infrastructure, models, and applications.

"We mainly focus on four layers, but we do not touch the energy layer at the bottom because China's energy efficiency is high and costs are relatively low."

In Daniel Zhang's view, this near full-stack layout stems from the uncertainty of the future, as no one can precisely define whether the ultimate value will be in chips, cloud infrastructure, or the model layer. "We choose to participate comprehensively; regardless of where the value ultimately settles, we are present."

Compared to Alibaba, more aggressive are U.S. cloud giants, who are almost fully invested in infrastructure, with total capital expenditure reaching $800 billion by 2027. This has also been criticized by shorts as a "bubble." Daniel Zhang not only disagrees with the bubble theory but emphasizes that Chinese companies need to increase investments in infrastructure.

"Investments in digital infrastructure are indeed astonishing," Daniel Zhang said. "We need to look back at that $50 trillion total market, which is the reason for optimism."

Regarding open source, Daniel Zhang first mentioned the recent halt of Anthropic's most powerful model by the Trump administration, directly calling it the consequence of "putting all eggs in one basket." In his view, models from Google, OpenAI, and Anthropic are now all closed-source, while open source is the path Chinese companies are taking.

"You really can't build trust on the premise that third-party governments will never do anything harmful to you."

Below is a condensed version of Daniel Zhang's interview:

The $50 Trillion "Market"

Q: Alibaba has changed a lot over the years, such as achievements in open-source large models. But many still think of you as just a B2B or B2C platform. Can you tell us about the development process of the entire group?

Daniel Zhang: When Alibaba started in 1999, it was indeed a B2B platform. The idea was simple: to bring China's small manufacturers and trading companies online, allowing them to wholesale goods to the world. Later, we entered the B2C space with Taobao, which is now China's largest consumer e-commerce platform.

Q: How many consumers does this service serve?

Daniel Zhang: 820 million Chinese consumers, and this platform helps European companies and brands sell about 30 billion euros worth of goods to Chinese consumers each year. But the story doesn't end there; we also heavily invest in AI and cloud.

We started investing in cloud technology 17 years ago, but it was driven by necessity. At that time, e-commerce generated massive amounts of data daily. Relying on others' databases and storage technology would mean all the profits go to the tech providers. So we decided to develop our own proprietary technology to manage this data, which marked the beginning of our cloud business.

From a macro perspective, we are now fully committed to AI, and the logic is very simple.

If you ask me how big the AI market is, I would say it exceeds any company's IT budget and is much larger than the software market. Because AI is fundamentally about producing human intelligence and productivity. Today, global GDP exceeds $100 trillion, with at least half ($50 trillion) related to human productivity and intelligence—that's the total AI market. So we must invest wholeheartedly.

Q: Do you really believe AI can boost productivity? Many have invested heavily but haven't seen results yet.

Daniel Zhang: Many CEOs will tell you that engineers are consuming a huge number of tokens, and costs are rising. But I want to say, we are on the verge of a real productivity explosion.

Take our company as an example: some engineers are super users of AI. They not only use programming tools to do their jobs but also explore new applications with them. Give engineers a toy, and they will find more ways to use it, often without realizing that the company is paying for these consumptions—that's the current reality.

But I am firmly convinced that this is more like a belief—that AI-generated intelligent units can add value to human intelligence. It’s almost a kind of faith. I don’t want to persuade you that this will definitely happen, but we believe in it ourselves.

All in AI Logic

Q: Returning to Alibaba's layout, which layer do you invest the most in— infrastructure, models, or cloud services?

Daniel Zhang: We mainly focus on four layers, but we do not touch the energy layer at the bottom because China's energy efficiency is high and costs are low.

Our real entry point was at the chip layer, which is the first layer; the second is the infrastructure layer, corresponding to our cloud business; the third is the model layer, like Qwen, which is now one of the most popular open-source models globally; the fourth is the application layer, where we have a complete digital lifestyle ecosystem—e-commerce, food delivery, local services, travel, maps, etc. These scenarios can directly embed AI capabilities to serve users.

The benefit of this approach is that we do not bet on a single track.

Today, everyone sees that pure model companies have high valuations, as if all value is in the model layer. But in the next five or ten years, no one can say whether the value will settle in chips, cloud infrastructure, models, or applications. We choose to participate comprehensively; regardless of where the value ultimately resides, we are present.

Q: Speaking of AI infrastructure, when you see such huge investments, do you think there is a bubble? Do we really need so much computing power? After all, some models are more efficient and require fewer resources.

Daniel Zhang: I don’t think it’s a bubble. The digital investments are indeed astonishing. Looking at the top cloud providers in the U.S., four or five leading companies will have capital expenditures exceeding $800 billion next year, possibly over a trillion in the following year. Such scale of investment naturally raises concerns about overcapacity.

But we need to look back at that $50 trillion total market, which is the reason for optimism.

Moreover, in China, our investments in AI infrastructure and supply chains are still insufficient. In theory, all Chinese companies should increase their investments. Of course, we won’t reach the scale of U.S. giants, but our investment efforts are already very substantial.

Q: Why not reach the level of the U.S.?

Daniel Zhang: Sometimes it’s limited by capital, depending on how much free cash flow you can generate. Fortunately, Alibaba is one of the few companies with core e-commerce operations, which generate about $25 billion in free cash flow annually, supporting our AI investments. So, our position is relatively good.

Q: Currently, e-commerce still accounts for over 80% to 85% of Alibaba’s total revenue?

Daniel Zhang: Yes, e-commerce revenue still accounts for over 80%. This segment produces stable cash flow, enabling us to invest in the future, mainly in AI and cloud.

Open Source and the Second Basket

Q: Qwen is an open-source model. Who are your main clients, and how do you help them?

Daniel Zhang: In recent weeks, I’ve been communicating with executives and scientists across Europe. The most frequently mentioned word here is “sovereignty.”

But what is sovereignty?

Ask ten Europeans, and you might get twelve answers. For me, it boils down to two core points.

First, technological independence. Everyone worries about the risk of a "one-click shutdown," fearing over-reliance on a certain country's technology, which could be turned off at any time. We’ve seen a vivid example of this in recent days.

Second is data privacy. People want to use AI technology but hope that their data fully belongs to them, used within their own environment, protected by a firewall.

I believe open source can address both issues. It is essentially free software—you can download it into your own data center or even onto a laptop. At this point, it has little to do with the original manufacturer; we can’t even think about charging. This achieves independence.

More importantly, with open-source models, you can use your own data for further training, fine-tuning, or post-training, with all processes and data kept completely confidential behind your firewall. This is crucial for European companies.

But I want to emphasize that open source is not a cure-all or the only path, but it is a practical way to achieve a certain level of sovereignty.

Interestingly, today’s open-source movement is actually driven by Chinese companies. Most major U.S. players have closed-source models. They want you to access models via API calls, and you have no idea where your data goes. When you chat with a bot, your most private questions and thoughts go into their data pools, used to further train the models. The data flow is completely opaque to you.

Q: Honestly, sovereignty is a huge concern for Europe now. We’ve just realized over-reliance on U.S. technology. I agree open source has benefits, but I still worry about future disconnection from models, which is a significant risk for Europe.

Daniel Zhang: You’re right; such concerns cannot be entirely eliminated. Simply put, you cannot build trust on the assumption that third-party governments will never do anything harmful. But the problem is, right now, all your eggs are in one basket.

Why not choose a second basket and split the eggs? Even if Europe develops its own basket in the long run, at least for now, you have two baskets.

Factories and AI

Q: That makes sense. How do you collaborate with German companies and what do you help them with?

Daniel Zhang: These German manufacturing companies are very interesting. In the Chinese market, they are all Alibaba Cloud clients. We cooperate with them in manufacturing, covering design, testing, quality control, and other stages.

I believe this will be a very important field in the future. Currently, most AI applications are either consumer products like ChatGPT or tools aimed at programmers and knowledge workers like Copilot. But in the future, these manufacturing companies will be highly valuable because they accumulate unique high-quality data during production, which can be used to train dedicated models and improve manufacturing processes.

We have collaborations with BMW, Siemens, and Bosch. Last week, I attended Bosch’s Connected World Conference, where they are developing assisted and autonomous driving technologies that require massive computing power.

There are many interesting developments happening in manufacturing.

Q: Can I understand that U.S. export controls on high-end chips have actually created opportunities for you?

Daniel Zhang: That’s a fair understanding. There are two paths:

One is that they directly adopt our open-source models and deploy them on their own infrastructure, such as data centers. Our infrastructure is tightly integrated with the models, highly efficient, and helps clients train models. If they use our open-source models, they can also purchase computing power from us. The models and infrastructure are mutually supportive—this is one path.

The other is that emerging inference platform companies now offer multiple model options. You don’t necessarily have to use Qwen; as long as the model provider and platform have an agreement to open weights in a private environment, customers can use these models on the platform.

AI, Agents, and Humanity

Q: A more philosophical question. How do you view the balance between AI, large language models, and humans, even the future of human nature? What will humanity be like in the next ten years?

Daniel Zhang: Today, I was chatting with colleagues in our Paris office. We just moved into a new building, on a beautiful floor. I looked out the window at a café, the weather was nice, people sitting outside drinking coffee, enjoying life.

I told my colleagues that this scene represents the future of AI.

You might think they’re just drinking coffee and having fun, seemingly not working, but in fact, they’ve deployed intelligent agents that are working for them. While you sleep, these agents are working for you. Think about this productivity boost—having “people” working for you 24/7.

Q: This is similar to some ideas in Silicon Valley, where many believe people won’t need to work because intelligent agents and robots will do everything for them.

Daniel Zhang: I believe this will definitely free up people's time to enjoy life, spend more time with family, and participate in more entertainment. That’s one reason I value live entertainment so much. When people spend less time in offices, where will they want to go? They won’t just stay at home—they’ll want to go to concerts, football, basketball games.

Q: Chinese are known for their diligence. Even with intelligent agents and AI, Chinese engineers still work long hours.

Daniel Zhang: There will always be people working harder than others, but I believe most people hope to enjoy more of life and spend more time with their families.