Masayoshi Son stated that superintelligent AI will be 10,000 times smarter than the human brain, requiring 400GW of power, 200 million chips, and $9 trillion in funding—perhaps even an underestimate.
On Wednesday, October 30, renowned entrepreneur, investor, and founder of SoftBank Group, Masayoshi Son, was interviewed by CNN reporter Richard Quest, where he shared his views on AI. The key points are as follows:
<blockquote>
<ul>
<li>The definition of AGI (Artificial General Intelligence) is equivalent to the human brain, while the definition of ASI (Artificial Superintelligence) is 10,000 times smarter than the human brain, which will be achieved by 2035.</li>
<li>ASI requires regulation; otherwise, it will be very dangerous.</li>
<li>The combination of ASI and robotics will be a great product.</li>
<li>NVIDIA is undervalued, and its future will be broader.</li>
<li>Building superintelligent AI will require 400GW of power, 200 million chips, and $9 trillion in funding.</li>
</ul>
</blockquote>
The full translation is as follows (slightly edited):
Richard Quest: Over the years, I have conducted countless interviews, but I have never been more interested in the question I want to ask than I am now. Let's start with what interests you; I know it's AI—I'm obsessed with achieving artificial superintelligence; are you still fascinated by it?
Masayoshi Son: Absolutely, that's all I focus on.
Richard Quest: What do you mean, that's all you focus on?
Masayoshi Son: That will forever change the future of humanity. I want to make the future of humanity better and happier; that’s what interests me.
Richard Quest: So, what’s the difference between AGI (Artificial General Intelligence) and ASI (Artificial Superintelligence)? What does it mean for those of us born in different generations?
Masayoshi Son: The definition of AGI is the same as the human brain, while the definition of ASI—how super is it? 10 times super or 100 times super? My definition of ASI is 10,000 times smarter than the human brain; that’s my definition of ASI, and it will arrive by 2035, ten years from now.
Richard Quest: So what’s the difference? What will it do, and what should we expect or worry about?
Masayoshi Son: We should look forward to it, of course, but we must also be cautious; we need to regulate it. If such a superintelligence emerges without regulation, it could be very dangerous. Just like the automotive industry is very beneficial to humanity, but we have regulations; the two are the same.
Richard Quest: Generally speaking, no one in the world wants to build or intends to build dangerous AI; there may be countries that want to do so, and there may be independent actors who want to build AI that harms others. How do we prevent it? How do you prevent it?
Masayoshi Son: Of course, there will always be some bad people; you know, 1% of humanity is bad, but 99% of humanity is actually good. Of course, bad people may try to use AI to express malice. However, bad people may not necessarily have enough capital; it’s a huge capital requirement. To make superintelligence a bad entity, I would need to invest hundreds of billions of dollars. How can bad people sponsor it? They don’t have enough money to sponsor it
Richard Quest: Frankly speaking, you have previously admitted that you are not involved in AGI because you ran out of money; your money is gone.
Masayoshi Son: Of course, money is not infinite. I have to save some money for this big battle. For this great opportunity, I am currently saving, you know, hundreds of billions of dollars, so I can let Max move forward boldly.
Richard Quest: Do you know where that is? I mean, you might not understand these companies; they may not even have started yet. Do you have a very strong compass to know which way to go?
Masayoshi Son: I can't say.
Richard Quest: Come on, that's why we're here.
Masayoshi Son: I have my own views. Our company has nearly 100% market share of all mobile chips and 90% of any high-end AI, IoT chips, and various other chips. So I will soon become the owner of an AI-centered chip company. Well, what I am very interested in is AI robotics technology, not traditional robotics technology. The emergence of ASI combined with robotics technology will be a great product.
Richard Quest: Your charm lies in the fact that you have been there; you did it, you achieved great success, and you experienced subsequent failures, and now you are doing the same thing again.
Masayoshi Son: No, I made many mistakes in my work, just like everyone else does.
Richard Quest: It's annoying when people focus on our work; do you feel annoyed? Does it bother you?
Masayoshi Son: Well, you know, people should have the freedom to think. I have learned more and more from my mistakes. After the dot-com bubble burst, during the rebound crisis, I was almost bankrupt twice, but I survived. In fact, I lost 99% of the company's value, but somehow I came back.
Richard Quest: Did you lose that value because you took on too much risk? Or was it foolishness? Bad luck?
Masayoshi Son: I made that foolish big mistake. The dot-com bubble burst because people's expectations of the internal company entities exceeded reality, while the real value was still in the process. But ultimately, the internet was not a bubble in 2000; after that, the internet became bigger. I have no regrets.
Richard Quest: This is very important when we talk about AI. For example, NVIDIA; many people say, oh, NVIDIA is in trouble, but what actually happened is just that the market is overvalued.
Masayoshi Son: That's a good question; that's a good question. Many people will still doubt it, I believe NVIDIA is undervalued.
Richard Quest: What conclusion do you draw that from?
Masayoshi Son: Because NVIDIA's future will be greater.
Richard Quest: But the market is concentrated, as we heard in a previous panel discussion, even if NVIDIA tells us it will grow more slowly, we all understand
Masayoshi Son: What about the future of ASI? What is its future value? Is this a bubble? Some critics, the most negative ones, say that in 10 years, ASI may only have 5% actual utility, or, you know, replace the right people's jobs, but 95% will remain the same, unchanged. This takes time, so I say, okay, let's accept the most negative critics' viewpoint: only 5% change, 95% unchanged. But, you know, what measures are needed for ASI investment? How many gigawatts? Just before me, we were talking about gigawatts. Yes, I predict this will require 400 gigawatts of AI data center power, which is larger than the total power of the United States, and will need 200 million chips, with cumulative capital expenditures of $9 trillion. How do you recoup the costs? This is too large an investment for many, but I say this is still a very reasonable capital expenditure; $9 trillion is not too large, maybe too small.
Richard Quest: You like to think big. Before we wrap up, I just want to ask, let's ask our AI friend, the question I posed to our AI friend: Is Masayoshi Son a master of investment? Can we get an answer?
(AI robot at the interview site): Masayoshi Son is definitely recognized as a master of investment. His early $20 million investment in Alibaba grew to $75 billion in 2014, making a significant contribution to SoftBank's success. However, the results of some other investments have been mixed. Thank you.
Richard Quest: I have one minute left, one minute. I want to ask you, after these two difficult periods, you entered a time of conflict, reflection, and contemplation. What new understanding do you have about yourself?
Masayoshi Son: I think this has been a great experience. I was defeated, I was mocked, I was criticized, which made me think harder and made me realize I must become stronger. Everything is good; I love more, I think more. Before I go to sleep, I have many worries, but when I wake up, I have a brilliant smile.
Richard Quest: Ladies and gentlemen, Mr. Masayoshi Son, thank you very much!

NVDA

OpenAI

Masayoshi Son believes that the definition of general artificial intelligence is akin to the human brain, while the definition of superintelligent AI is being 10,000 times smarter than the human brain, which will be achieved by 2035. However, superintelligent AI requires regulation; otherwise, it will be very dangerous. Building superintelligent AI requires 400GW of power, which is greater than the total electricity consumption of the United States

- Masayoshi Son predicts that superintelligent AI will surpass human intelligence by 10,000 times by 2035, requiring 400 GW of power, 200 million chips, and $900 billion in investment.  
- He emphasizes the need for regulation to prevent potential dangers associated with such technology and believes that the combination of superintelligence and robotics will yield significant advancements.  
- Son also argues that Nvidia is undervalued and that the investment required for superintelligent AI is reasonable, despite concerns about its future utility.

Wallstreetcn

NVIDIA

Masayoshi Son: Super AI will be 10,000 times smarter than the human brain! But it will take at least 200 million chips and 900 billion dollars to build it!