Cai Chongxin's Depth Review of the HKU Speech: The Four Cards of China's AI and the Misunderstood US AI Rules

Joe Tsai's HKU Speech: China's Four AI Assets and the Misread U.S. "AI Winner Rule"

Note from the author: On November 5th, the Hong Kong University of Science and Technology's School of Business and Management hosted the "Chen Kun Yao Distinguished Scholar Lecture," with the theme "Looking Forward to the Next Decade: The Sci-Tech Engine Driving China's Economic Growth." They invited Mr. Cai Chongxin, co-founder and chairman of Alibaba Group, to share insights on how innovation, technology, and artificial intelligence are reshaping the business landscape and driving long-term economic growth in China. The event took place at the Hong Kong Mainland You Hall, and according to the organizers, this lecture saw the highest registration rate in history, with the number of registrants exceeding 1200 within two hours of the email being sent out.

In a sense, this can also be seen as a historical echo. Seventeen years ago, Jack Ma stood on the same stage to give a speech... In this in-depth dialogue with Professor Deng Xiwei, the Vice President of the University of Hong Kong, Cai Chongxin set aside pleasantries and directly addressed the core of the Sino-American AI competition, dissected Alibaba's business evolution theory, and provided piercing advice for young people...


The following excerpt is from Zhiding Technology Gao Fei.

1. Redefining the AI Competition: China Holds "Four Trump Cards"

Cai Chongxin opened by throwing out an counterintuitive point: the AI competition rules defined by Americans may be wrong.

The current "American scoreboard" only looks at whose large language model (LLM) is stronger, today it is OpenAI, tomorrow it is Anthropic, and the day after tomorrow it is Gemini; however, in Cai Chongxin's view, this evaluation system itself has problems.

1. The true winner logic: penetration rate > model parameters

The winner is not about who has the best model... The winner is about who could use it the best in their own industries, in their own lives...

The true value of AI lies in its penetration rate. Compared to the infinite stacking of parameters, the Chinese government's AI plan appears more pragmatic: the goal is to achieve a penetration rate of 90% for AI agents and devices by 2030. No metaphysics, just universality.

2. Why can China popularize faster? Four system-level trump cards.

To support this penetration rate, Cai Chongxin listed four key advantages that China possesses:

Card One: Advantage of Power Cost (40% Lower). Training and inference are essentially energy-consuming battles. Thanks to the construction of ultra-high voltage transmission networks initiated 15 years ago (such as "West-East Electricity Transmission"), China's State Grid has annual capital expenditures reaching $90 billion, which is three times that of the United States. This enables China's power installed capacity to be 2.6 times that of the United States, with new capacity being 9 times that of the United States.

Card Two: Infrastructure Dividend (60% Lower). The cost of building data centers in China is 60% lower than in the United States. This is just the cost of infrastructure, not including hardware such as chips.

Card Three: Engineer Bonuses and Language Advantages. About half of the AI scientists and researchers globally have an educational background in China. Cai Chongxin shared an interesting phenomenon: within Meta's AI team, non-Chinese speaking employees often feel "lost" because everyone is speaking Chinese.

This is the first time Chinese language is an advantage... (This is the first time that Chinese has become a natural communication advantage in the field of technology.)

Card Four: The computing power is greatly restricted by the U.S. government, which instead forces system-level innovation. The U.S. has abundant GPU resources, while China does not. But this creates a "Advantage of Starvation."

When you don't have a lot of resources, you are forced to innovate at the systems level...

In order to train trillion-parameter models on limited hardware, the Chinese team must optimize system efficiency to the extreme. DeepSeek is a typical example, and in a recent AI competition for cryptocurrency and stock trading, Alibaba's Tongyi Qianwen (Qwen) ranked first, while DeepSeek came in second.

Cai Chongxin has a very high opinion of Tongcheng's DeepSeek: "They are doing incredible things."

II. Why Open Source Will Win: The Triple Logic of Cost, Sovereignty, and Privacy

Regarding the debate of "Open Source vs Closed Source", Cai Chongxin has made a clear judgment: open source models will ultimately defeat closed source.

This is not simply a matter of technological superiority, but rather because open source aligns more closely with the interests of the majority of users worldwide. He used "Saudi Arabia wants to develop AI, but also wants to maintain AI sovereignty" as an example to elaborate on the underlying business logic:

🔹 Closed-source path (e.g., OpenAI): Expensive to pay, and must input data into a black box, which poses data sovereignty risks.

🔹 Open source path (e.g., Alibaba Qwen): Free download, deploy on private cloud. Data is fully controllable, with very low cost.

As long as the government and enterprises conduct rational cost-benefit analysis, open source is the better solution.

So how does Alibaba make money?

Cai Chongxin said frankly: "We do not make money from AI."

Alibaba's business model: "We don't make money by selling models, we rely on cloud computing." Cai Chongxin admitted that open-source models serve as traffic entry points, while the resulting demands for storage, security, containerization, and other cloud infrastructure are the sources of profit. This is similar to the early internet: products are free to attract customers, and value-added services are monetized.

3. The Evolution of Alibaba: Technological Independence is "Forced" Out.

When asked how Alibaba evolved from an e-commerce company to a cloud computing giant, Joe Tsai's answer was very straightforward: "There is no secret, just follow customer demand."

🔹 B2B Era: To address the export needs of small and medium-sized enterprises after entering the WTO.

🔹 Taobao/Alipay: To address trust issues in C-end transactions.

🔹 Alibaba Cloud: To solve the cost issue of processing vast amounts of data. 16 years ago, if we continued to use traditional IT facilities from Dell, EMC, and Oracle, Alibaba's profits would have been drained.

We developed cloud computing really out of necessity... out of the need to become self-reliant in technology... (We developed cloud computing completely out of necessity, out of the desire for technological autonomy.)

So Alibaba Cloud's starting point is "eat our own dog food": using it internally first, and after doing well, opening it up to external customers.

Advice for Entrepreneurs: Prioritize organic growth rather than mergers and acquisitions. The capabilities developed by your own team have a purer DNA and a better cultural fit.

4. A Guide for Young People: Thinking is More Important than Skills

During the Q&A session, Cai Chongxin provided a high density of advice on personal growth.

1. Skills aspect: Learn to ask questions.

In the age of AI, obtaining answers has become easier. Therefore, asking the right questions is more important than finding the answers. At the same time, it is essential to establish an independent analytical framework rather than relying on rote memorization.

2. Programming Level: Focus on Logic

Even if natural language can command machines, programming still needs to be learned. Even mastering Excel formulas is excellent logical training.

The purpose is not to actually operate a machine. The purpose is going through that thinking process...

3. Professional Choices: Three Potential Directions

🔹 Data Science: With the explosion of data, professionals who understand how to manage and analyze data are always in short supply.

🔹 Psychology/Biology: The human brain is the most efficient machine, and understanding the human brain is the shortcut to understanding AI.

🔹 Materials Science: This is a world dominated by bits, but the speed of bits is limited by atoms. There will be numerous innovations and breakthroughs in semiconductors in the future, with materials at the core.

5. Risks and Bubbles: A Financial Perspective

1. Career choice: Asymmetric risk

Cai Chongxin reflected on his decision to give up a million-dollar salary to join Alibaba in 1999, describing it as "asymmetric risk-reward": limited downside risk (worst case, go back to being a lawyer), unlimited upside potential (like a call option).

He emphasized, "Opportunities come looking for you; what you need to do is be always prepared (Preparedness)."


2, AI Bubble: Distinguishing Finance from Technology

Is today's AI like the internet of 2000? Cai Chongxin suggests distinguishing between two types of bubbles:

🔹 Financial bubble: Valuations may be too high, and this is difficult to judge.

🔹 Technological bubble: The technology itself is real. Just as the stock market crash in 2000 did not erase the existence of the internet, all the investments made today in AI infrastructure and model development will not go to waste; they are the cornerstone for the future.

Six, three core insights.

Q1: What is the real advantage of Chinese AI?

It is not the model itself, but the entire ecosystem that allows AI to be widely used. The cost of electricity is 40% lower, the cost of building data centers is 60% lower, half of the AI talent globally has a Chinese degree, and resource scarcity drives system-level innovation. All of these factors together make it more likely for China to achieve large-scale AI adoption. And the adoption rate is the real scoreboard.

Q2: Why will the open-source model win?

For the majority of users worldwide, open source simultaneously addresses three issues: cost, data sovereignty, and privacy. Closed source models require payment, and data must be fed into a black box; open source models are free, and data can remain local. This is not a contest of technical superiority, but rather a result of the interests at play.

Q3: How should young people prepare for the AI era?

Learning programming is not just about writing code, but about training logical thinking; studying statistics (data science) is because data will explode; studying psychology is to understand the most efficient "machine" which is the human brain; studying materials science is because what makes bits run faster are atoms. More importantly, learning to ask the right questions is more valuable than finding the answers.