NVIDIA CEO Huang Renxun AI will not take away your job, but the people who use AI will.

NVIDIA CEO Huang Renxun AI Won't Replace Your Job, But Those Who Utilize It Will.

Source: Newin

This morning, NVIDIA announced its Q3 financial report after the US stock market closed. The third-quarter revenue as of October 29, 2023, reached $18.12 billion, a year-on-year growth of 206% and a quarter-on-quarter growth of 34%. EPS profit increased nearly 6 times, surpassing analysts’ expectations by nearly 13% and 20% respectively. The revenue from AI chip-related business data center nearly doubled year-on-year, reaching a new quarterly high. The old yellow guy has won big again!

Old Yellow said, “Our strong growth reflects the wide-ranging industry transformation from general purpose to accelerated computing and generative AI. LLM startups, consumer internet companies, and global cloud service providers are leading the way. The next wave is beginning to take shape. National and regional communication service providers are investing in AI cloud to meet local demand. Enterprise software companies are adding AI Copilot and Assistant to their platforms. Enterprises are creating customized AI to achieve automation in the world’s largest industries. NVIDIA’s GPUs, CPUs, networks, AI foundry services, and NVIDIA AI Enterprise software are all engines of rapid growth. The generative AI era is taking off!”

• The end of CZ era at Binance may be the most magnificent 6 years for Chinese entrepreneurship.
• Chinese Police On the Lookout as CBDC-themed Money Laundering Schemes Gain Traction
• Aragon DAO’s Bold Move Backing Lawsuit Against Founding Team with $300,000 in Funding!

PS: Last weekend, in our column, we shared the over 10,000-word analysis of John Luttig, the investment manager of Funders Fund, on the current GPU market situation.

Last month, NVIDIA’s co-founder & CEO Jensen Huang also gave a very practical sharing session at Columbia Business School (CBS). Old Yellow had a conversation with CBS Dean Costis Maglaras to explore the digital future, including how Nvidia strategizes and operates, Old Yellow’s entrepreneurial experience, and how to become a qualified CEO, and more.

Here are some highlights from Old Yellow’s sharing session at CBS for everyone to savor:

Before making a specific decision, everyone should be clear about what they are doing and why they are doing it. And all this has nothing to do with choice.

From a personal perspective: three points need to be determined:

1) Things that are difficult yet right;

2) Things you are destined to do;

3) Things you enjoy.

From a company’s perspective: Using NVIDIA as an example, Old Yellow’s answer was very straightforward, explaining clearly the market choices, business models, barriers, and flywheel effects involved in NVIDIA’s pivot:

“The reason we don’t do manufacturing is because TSMC does it so well, and they are already doing it. Why should I take their work away? I like the people at TSMC, they are my good friends. Just because I have business, I can enter this field, so what? They do a great job for me. Let’s not waste time duplicating what they have already done. Let’s waste time doing something that no one else has done. That’s how you build something special. Otherwise, you’re just talking about market share.”

We observe two things: accelerated computing is a software problem, an algorithm problem, while AI is a data center problem. So we are the only company that goes out and builds all of these things. Part of what we do is make choices on the business model. We could have easily been a data center company, fully vertically integrated. However, we recognize that no matter how successful a computing company is, it will never be the only computing company in the world. It’s better to be a platform computing company because we love developers. It’s better to be a platform computing company that serves every computing company in the world than to be a standalone computing company.

We took this approach of taking this data center, the size of this room, with all the wires and all the switches and all the network and tons of software, and decompose it and integrate all of that into different data centers all over the world. It’s a crazy complex problem, and we found a way to have enough standardization when necessary and enough flexibility when necessary so that we could collaborate with every computing company in the world.

The result is now Nvidia’s architecture is in every computing company in the world. It creates a larger footprint, a larger installed base, more developers, better applications, happier customers. They buy more chips. It increases the installed base. It increases our R&D budget, and so on and so forth. It’s a flywheel effect. It’s a virtuous cycle, and that’s how it works. Simple as that.”

Furthermore, Mr. Huang also clarifies his views on AI and labor and workflows: “AI will not take away your jobs. People who use AI will take away your jobs. And if a company doesn’t have more ideas to invest in incremental productivity, then when the job gets automated, companies will have to lay off and join those companies that have more ideas and cannot afford that kind of capital investment. So, when AI automates their jobs, the situation is going to change. The nature of work is going to change.“

Below is the full dialogue between Mr. Huang and CBS Dean Costis Maglaras. Enjoy~

Costis Maglaras:

I want to start by having you walk us through Nvidia’s history, and then I want to talk about the leadership question we mentioned earlier, but you founded this company 30 years ago and led the transformation with different applications and types of products. Take us through that journey.

Jensen Huang:

One of my proudest moments was recently when I met the CEO of the first company I ever worked for, Denny’s. And he not only remembered me from being a dishwasher, busboy, and waiter at Denny’s, he remembers me from the first company, and I still know the menu. By the way, Superbird is fantastic. Anybody know Superbird? What kind of college students are you guys?

Denny’s is a restaurant in the United States, and Nvidia is a company that I and two other co-founders established outside the Denny’s near our house in San Jose. So they recently contacted me, and the booth we used to sit in has now become Nvidia’s booth, named Nvidia. This is where a trillion-dollar company was born, and it’s a very proud moment.

When Nvidia was founded, it was at the beginning of the PC revolution, and microprocessors captured the imagination of the entire industry. The world rightly saw how CPUs and microprocessors would reshape the IT and computer industries. Companies that succeeded before and after the x86 revolution were drastically different. During that period, we started our company with the belief that while general computing is impressive, it cannot be the solution to all problems.

We believed that there existed a form of computing that we referred to as accelerated computing, where you add an expert alongside general computing. If you will, the CPU is a jack-of-all-trades, capable of doing anything. It can do anything. However, obviously, if you can do anything, then clearly you can’t do everything well.

Therefore, we believed that certain problems were not suited to be solved by what we call conventional computers. Thus, we pioneered this company focused on accelerated computing. The problem was, if you want to create a computing platform company, I don’t know how many computer scientists are here, but since 1964, no such company has been born. That was the year I was born, and IBM’s System/360 perfectly described what a computer is.

In 1964, IBM described the System/360 having a central processing unit, I/O subsystem, direct memory access, virtual memory, binary compatibility across scalable architecture. It described everything we describe today about computers. 60 years later, we felt there was a new form of computing that could solve some interesting problems. At the time, it wasn’t entirely clear what problems we could solve, but we felt accelerated computing had a future.

Nevertheless, we set out to start this company and made a very good initial decision, to be honest. This decision remains incredible even today. If someone came to you and said, “We want to invent a new technology that doesn’t exist in the world, and everyone wants to build a computer company around the CPU, but we want to build a computer company around something connected to the CPU,” that was the first point.

The second point is that the killer application was a video game, a 3D video game from 1993, and that application didn’t exist. The company that built this company didn’t exist, and the technology we were trying to build didn’t exist. So now you have a company facing both technological challenges and market challenges, and the probability of success for this company was almost 0%, but nonetheless, we were fortunate because of two incredibly important people.

Frankly speaking, the three co-founders I’ve worked with were very important figures in the tech industry at the time. I called up the world’s most important venture capitalist, Don Valentine, and told him to give this kid some money and see if it works along the way. Luckily, they did, but that business plan, even today, I wouldn’t invest in it because it had too many dependencies, each with a certain probability of success.

When you add all these together, multiply them, you get 0%. Nevertheless, we envisioned a market called video games that would become the largest entertainment industry in the world, even though it was 0 at the time. We speculated that 3D graphics would be used to tell the stories of almost all sports and games. So, in the virtual world, you could have any game, any sport, and everyone would become a gamer.

Don Valentine asked me, how big would this market be? I said, in the future, everyone would be a gamer, which was also the wrong answer when founding a company. Honestly, these were all terrible habits, terrible skills. I’m not suggesting you do this. Anyway, the result proved to be true. Video games became the largest entertainment industry in the world. 3D graphics were successful. We found the first killer application for accelerated computing, which bought us time and solved a range of other problems using accelerated computing, eventually turning to AI.

Costis Maglaras:

This fascinating story is really amazing. Before we discuss AI, I want to ask about the time of cryptocurrencies. Obviously, gaming was a big milestone for Nvidia, and then at some point, the killer application became cryptocurrencies and mining. How did that development happen?

Jensen Huang:

Accelerated computing can solve problems that regular computers cannot solve. All of our GPUs, whether used for designing cars, buildings, conducting molecular dynamics research, or playing video games, have a programming model called CUDA that we invented. CUDA is the only existing and popular computing model today, just like x86, and it is used by developers all over the world.

Anyway, CUDA can perform parallel processing very quickly. Obviously, one of the algorithms we can handle very well is cryptography. When Bitcoin first appeared, there were no Bitcoin ASICs yet. The obvious solution was to find the world’s fastest supercomputer, and the supercomputer with the highest yield was none other than Nvidia’s GPUs, which were in millions of homes of gamers. So by downloading an application, you could mine cryptocurrencies at home.

This fact was the day my mom understood what I did for a living. One day she called me and said, “Son, I thought you were doing something about video games, but I finally understand what you’re doing. You buy Nvidia products, plug them in, and money starts pouring out.”

I said, yes, this is what I’m doing, this is why so many people are buying Bitcoin, leading to the rise of Ethereum, but would you have the idea to use a supercomputing system like Nvidia GPU to encode or compress, or do something to refine data and convert it into valuable tokens, do you know what this sounds like? Generating valuable tokens like ChatGPT.

So far, one thing has happened, if you extend the idea of Ethereum and crypto mining, it makes sense in a way because we suddenly created this new industry, raw data inputs, you apply energy to this machine, and money literally starts flowing out, and these currencies are of course in token form, these tokens are smart tokens, this is one of the major industries of the future; Now I’m just describing something that is very meaningful to us today, but back then it looked strange, you bring water into a building, heat it up, and what comes out is something very valuable and invisible, called electricity.

Today we move data into data centers, it refines and processes it and uses its capabilities to generate a large number of valuable digital tokens, in digital biology, they will be valuable, in physics, in IT and various computing fields, social media, various things, computer games, and so on, they appear in token form, so the future will be about AI factories, and Nvidia’s devices will power these AI factories.

Costis Maglaras:

So we’ve already jumped into neural networks, I think we’ve talked about parallel computing, like how to render graphics on a monitor, how to play games, how to solve cryptographic problems for Bitcoin. Please tell us a little bit about the use of GPUs in training neural networks, I want you to talk to the audience here about what it takes to train models like ChatGPT. What hardware is needed? What data is needed? How big of a cluster is needed? How much money does it cost? Because these are big questions, and I think it would be good for us to have an understanding of scale.

Jensen Huang:

Everyone wants you to think it’s a big problem, very expensive. But it’s not, let me tell you why, our company spent about $500-600 million in engineering costs to design a chip, and then one to two years later, I press enter, send an email to TSMC, send them a big file via FTP, and they manufacture it, this process costs our company about 500 million dollars.

It’s a total cost of $5.5 billion, and I get a chip, which is valuable to us, but it’s not a big deal. I’ve been doing this, so if someone says, hey Jensen, you need to create a $1 billion data center, once you plug it in, money will start pouring out the other side. I would do it immediately, and obviously many people would too, because who wouldn’t want to create an intelligent factory?

Now, 1 billion dollars is not really a lot of money. To be honest, the global expenditure on infrastructure computing is about 250 billion dollars per year. None of us are generating money, we are just storing our files, sending our emails, and that alone costs 250 billion dollars. One of the reasons we are growing so fast is because after 60 years of development, general computing is declining. It’s not wise to reinvest 250 billion dollars to create another general computing data center. It’s too harsh on energy and too slow in computation.

Now comes accelerated computing. That 250 billion dollars will be used to create accelerated computing data centers, and we are happy to support customers in doing that. In addition to accelerated computing, you now have an infrastructure for generating AI, like everything we just talked about. The basic way it works is that you take a lot of data and then compress it.

Deep learning is like a compression algorithm. You try to learn the mathematical representation, patterns, and relationships of the data you are studying and compress it into a neural network. So the input is, let’s say, trillions of bytes, trillions of tokens, so a few trillion bytes, and the output is 100 GB. So you’ve compressed all that data into this small file, 100 GB, like 2 DVDs that you can download and watch on your phone, right?

So, you can download this huge neural network onto your phone. Now, all that data is compressed and the compressed neural network model is an LLM, which means you can interact with it. You can ask it questions, it will go back to its memory, understand your intent, and generate text for you, have a conversation with you. So, that’s the core of it. It sounds magical, but for all the computer scientists and researchers in the room, it makes perfect sense. Don’t let anyone convince you that it costs a lot of money. I’ll give you a good discount. Let’s all create AI.

Costis Maglaras:

If I may ask a follow-up on that scale, you would need a computer that is essentially equivalent to a data center to estimate these models.

Jensen Huang:

GPT-4 requires 16,000 GPUs, which is the largest model anyone is currently using, worth 1 billion dollars. And that’s just one check, not even a big one. Don’t be afraid, don’t let anyone discourage you from starting a business and realizing your dreams.

Costis Maglaras: Let me ask you a question about the 1 billion dollar check and the growth you have experienced. I believe you were named the best CEO by Harvard Business Review. It’s entertaining. I will keep repeating that point, but in a sense, you are now leading a company through extreme growth, hypergrowth, something most companies never experience in their lifetime. I would like you to tell us some details, such as doubling in size within a year or managing the supply chain, managing customers, managing growth, managing funds. How did you do it?

Jensen Huang:

I enjoy management, only a part of it, which is counting money. It’s quite fun. When I wake up in the morning, I roll around in all the cash. Isn’t that why all of you are here? I understand that this is the ultimate goal, building a company is hard, there’s nothing easy about it. There’s a lot of pain and hardship, requires a lot of effort.

If it was easy, everyone would do it. And about all companies, regardless of size, whether it’s ours or other technology companies, you’re always dying, because there’s always someone trying to surpass you, so you’re always heading towards bankruptcy; if you don’t internalize this feeling, if you don’t believe it, you will go bankrupt. And I started with Denny, as you all know, Nvidia was built under extremely impossible circumstances. It took us a long time to get to where we are today. I mean, we are a 30-year-old company. When Nvidia was founded in 1993, Windows 95 hadn’t even been released. That was the first usable PC. We didn’t have email.

There were no laptops or smartphones back then. None of these things existed, so you can imagine how different our world was back then compared to now. We didn’t have LCD screens. Everything was Cathode Ray Tube (CRT). In that era, even CD-ROMs didn’t exist. In short, these things were the backdrop of the early days of our company. It took us so long to be recognized and become the company that reshaped computing for the first time in 60 years. Rapid growth depends entirely on people.

Obviously, a company depends on people. If you have the right system and outstanding people like me around you, the company will have the skills. It doesn’t matter if you sell $100 billion or $200 billion.

The truth now is that the supply chain is not simple. Does anyone know what a G-Force graphics card looks like? Raise your hand if you know what an Nvidia graphics card looks like. So you would think that a graphics card is like a cartridge that you insert into a PC Express slot on a PC, but our current graphics cards, used in these deep learning systems, have 35,000 components, they weigh 70 pounds. They’re so heavy that robots are needed for assembly. They need supercomputers to test them because they themselves are supercomputers, and they cost $200,000. With this $200,000, you can buy a computer that can replace several hundred general-purpose processors, which would cost millions of dollars, Every $200,000 you spend on Nvidia saves you $2.5 million in computing costs. That’s why I’m telling you, the more you buy, the more you save. Obviously, this strategy is very successful. People are really lining up to buy. That’s our job; The supply chain is very complex. We manufacture the most complex computers in the world. But how difficult is it really? It’s actually very difficult. The essence of it is, if you’re surrounded by outstanding people, the simple truth is, it’s all about the people; I’m lucky to have a great management team around me, and then the CEO would say things like “Make it number one,” or “Make it work.”

Costis Maglaras:

I want to go back to the AI trend and your perspective on the future, but you mentioned the word “platform” earlier and you referred to your software environment. So, you have the hardware infrastructure, you have a software environment that’s ubiquitous in training neural networks. You’re building data centers, or creating environments within data centers, which are essentially communication clusters between Nvidia’s hardware, software, and these resources. How important is it to have a complete platform solution versus just the hardware involvement? And how core is this to Nvidia’s strategy?

Jensen Huang:

I think, first of all, before you can create something, you have to know what you’re creating and why you’re creating it, and what are the first principles that it needs to exist by.

Accelerated computing is not a chip; that’s why it’s not called an accelerator. Accelerated computing is about understanding how you can accelerate everything in life. If you could accelerate every application, then that would be truly fast computing. So, accelerated computing starts with understanding which domains, which applications are important to you, and understanding the algorithms, the computing systems, the architectures required to accelerate those applications.

The proof is that general-purpose computing is a reasonable idea, and accelerating an application is also a reasonable idea. For example, you have a DVD decoder. You have an h.264 decoder on your phone. It just does one thing, and it does it really well. Nobody knows how to do it better.

Accelerated computing is a little bit like this odd intermediate state. There are many applications you can accelerate. For example, we can accelerate all sorts of image processing, particle physics, including linear algebra. We can accelerate many domains, and that’s one of the challenges because accelerating one thing is typically easy, you just run everything through a C compiler.

Accelerating enough domains, so that if you accelerated too many domains, you end up back on a general-purpose processor. Why can’t they just make a faster chip? On the other hand, if you accelerated just one application, then the market size isn’t big enough to support your R&D.

So, we have to find that middle ground, and that is the journey of our company’s strategy, that’s the intersection of strategy and reality. That’s where Nvidia got it right, and no other company in the history of computing has ever gotten it right; to find a way to have a large enough set of application domains that we can accelerate, that we can still be 100 to 500 times faster than a CPU economically. That creates a flywheel effect where we can grow the number of application domains, grow the number of customers, grow the number of markets, increase sales, and create even bigger R&D budgets so that we can create even more amazing things and stay well ahead of the CPU. Does that make sense?

Creating this flywheel effect is extremely difficult. Nobody has done it before, only once, and that’s capability. To do this, you have to understand the algorithms, you have to understand the application domains very well, you have to choose correctly, you have to architect it correctly. And then the last thing we got right, we realized that in order to have a computing platform, an application that you develop for Nvidia should run on all Nvidias. You shouldn’t wonder, “Does it run on this chip? Does it run on that chip?” It should run on every computer that has Nvidia.

This is why every GPU created by our company, even if no customers were using CUDA at the time, we dedicated ourselves to it. We were determined to create this computing platform from the beginning. Customers are not, but it’s the result of decades of hard work and billions of dollars. We wouldn’t be here without all you video gamers. You are our daily work, and at night we can work on solving problems in digital biology, help people with quantum chemistry, AI, robotics, and more.

We realize that first, accelerating computing is a software problem. And secondly, AI is a data center infrastructure problem, which is very obvious because you can’t train AI models on a laptop or a phone because they are not big enough computers. The data is in trillions of bytes, and you have to process these trillions of bytes billions of times, so obviously, this will require a massive computer distributed across millions of GPUs.

When I say millions, it’s because there are thousands within each of the 16,000 internal ones. So we are distributing the workload across millions of processors. Currently, there is no application in the world that can be distributed across millions of processors; Excel runs on one processor. So, this problem of distributed computing is a huge breakthrough in computer science, definitely a significant breakthrough, which is why it enables generative AI, enables LLM.

We observed two things: accelerating computing is a software problem, an algorithm problem, and AI is a data center problem. So we are the only company out there building all these things. Part of what we do is a choice of business model; we could have become a data center company, fully vertically integrated. However, we realized that no matter how successful a computer company is, it won’t be the only computer company in the world. Being a platform computing company is better because we love developers. Being a platform computing company that serves every computer company in the world is better than being a standalone computer company.

We took this approach, breaking down this data center, which is the size of this room, all the wires, all the switches and networks, and a ton of software, and we broke it down and integrated it into other different data centers around the world. It’s a crazy complex problem, and we found a way to have enough standardization where necessary and enough flexibility where required so that we can collaborate with computer companies around the world.

The result is, Nvidia’s architecture is now implanted in every computer company in the world, creating a larger footprint, a bigger install base, more developers, better applications, which makes customers happier, they buy more chips, which increases the install base, increases our R&D budget, and so on. It’s a flywheel effect, a positive feedback system, that’s how it operates, simple and straightforward.

Costis Maglaras:

One thing you haven’t done, I hope you can explain, is that you haven’t invested in manufacturing your own chips.

Jensen Huang:

Why is it like this? That’s a great question, and the reason is as a strategic choice, the core values of our company, my personal core values, our company’s core values are all about choice.

The most important thing in life is choice. How do you choose? Well, everything is, how do you choose what to do tonight? How do you choose? Our company chooses projects, purely for a fundamental goal, my goal is to create an environment, an environment where the world’s most talented people come here to work. An amazing environment, where the world’s most talented people, who want to pursue computer computing, computer science, and AI. Create conditions for them to come here to complete their lifelong work.

Now, if I were to say this, the question now is, how do you achieve this? Let me give an example of how not to achieve this. No one I know wakes up in the morning and says, you know, my neighbor is doing that. What I want to do is, I want to take it from them. I can do it too. I want to take it from them. I want to grab their market share. I want to undermine them in price. I want to kick them. I want to take away their share.

It turns out, no great person does that, everyone wakes up in the morning and says, I want to do something unprecedented. It’s very hard to do. If successful, it can have a huge impact on the world, that’s the core value of NVIDIA.

First, how do we choose to do something in the world that has never been done before? We hope that it is incredibly difficult. By the way, the reason you choose to do something incredibly difficult is because you have a lot of time to learn it. If something is easy to do, like TikTok dances, I wouldn’t worry about it, obviously because the competition is fierce, so you have to choose something incredibly difficult, and those difficult things themselves will deter many others because the person who is willing to endure the longest will eventually win, so we choose something incredibly difficult, you have heard me talk about pain and suffering many times, it is actually a positive trait, the person who can endure the most will ultimately be the most successful.

Second, you should choose something you are destined to do, whether it’s your personality traits, your expertise, or the environment you are in, your scale, anything, your perspective, something you are destined to do.

Third, you better really enjoy doing that thing, because unless it’s too painful and sufferable. Now what I just described to you is NVIDIA’s core values. It’s that simple. If that’s the case, why would I make smartphone chips? How many companies in the world can make smartphones? A lot. Why would I make a CPU? Do we need more CPUs? Is that reasonable? We don’t need all these things.

Therefore, we naturally excluded ourselves from the mass market. We naturally excluded ourselves from the mass market because we chose an astonishing market, chose to do something very difficult, and recruited extraordinary talent. Because we have the patience to let them succeed, to do something extraordinary. If we have the patience to let them do something extraordinary, they will do something extraordinary.

The formula is actually that simple, but it requires an incredible character to execute. Does that make sense? That’s why learning it is the most important thing. Great success and greatness are all related to character.

The reason we don’t do manufacturing is because TSMC does it so well, and they’re already doing it. Why should I try to take away their job? I like the people at TSMC, they’re my good friends. Just because I have a business, I can enter this field, so what? They’re doing it better than me, so let’s not waste time repeating what they’ve already done. Let’s waste time doing something no one has ever done before, something unique. Otherwise, you’re just talking about market share.

Costis Maglaras: When we think about the future, when we think about these 10 years.

Jensen Huang: The correct answer? By the way, I know I don’t have an MBA degree, I don’t have a finance degree, I’ve read some books, watched a lot of YouTube videos. Let me tell you, no one watches more business YouTube videos than me, so I can tell you, you guys are useless to me, but these are the correct answers, Professor Maglaras?

Costis Maglaras: You asked the wrong person. I haven’t studied business either, but they are the correct answers, haha~ What do you think will happen with AI when you consider AI applications and the changes we will see in the next three, five, or seven years? And where in our daily lives might we be affected?

Jensen Huang: First of all, I’ll say it straight: AI won’t take your jobs, people who use AI will take your jobs. Do you agree with that? Well, start using AI as soon as possible so that you can have meaningful employment.

Here’s my second question to you. When productivity increases, it means that we embed AI comprehensively throughout NVIDIA. NVIDIA will become a huge AI entity. We are already using AI to design our chips. We can’t design our chips without AI, and we can’t write our optimizing compilers without AI. So we use AI everywhere.

When AI increases productivity in your company, what happens next? Do you lay off people or hire more? You hire more. The reason is that increased productivity leads to profitable growth.

Why do people consider losing their jobs? It doesn’t make sense logically if you think you don’t have any new ideas. If you don’t have more ideas to invest your incremental income, what will you do when your job gets automated? You will lay off and join those companies that have more ideas but can’t afford to invest in funding, so when AI automates their jobs, the situation will of course change, the way of work will change.

AI will soon target CEOs, department heads, and CEOs, we’re screwed, sounds good, I think it’s first CEOs, then department heads, but you’re close enough, so you join those companies that have more ideas but not enough funding to invest, naturally, as the revenue increases, you will hire more people. Firstly, it’s a huge breakthrough, we somehow taught computers how to learn, represent information in a digital way, right? So, have any of you heard of this thing called Word2Vec? It’s one of the greatest things ever, Word2Vec, you take a word, learn by studying each and every word and its relationship with every other word, you learn all of our sentences and paragraphs, you try to figure out what the most relevant numerical vector for that word is, what numbers are most related to that word, so “mother” and “father” are close in value, “orange” and “apple” are close in value, but they are far from “mom” and “dad,” “dog” and “cat” are far from “mom” and “dad,” but maybe closer than they are to “orange” and “apple,” chair and table, it’s hard to say exactly where they are, but these two numbers are close to each other, far from “mom” and “dad,” “king” and “queen,” close to “mom” and “dad.”

Does this make sense? Imagine doing this for every number, every time you test it, you go, oh my god, this is awesome. When you subtract one thing from another, it makes sense. Well, this is basically representing learning information. Imagine doing this for English. Imagine doing this for every language. Imagine doing this for anything with structure, meaning anything that has predictability.

Images have structure because if they don’t have structure, it would be white noise, it’s essentially white noise, so there must be structure, that’s why you see a cat, I see a cat, you see a tree, I see a tree, you can recognize where the tree is, you can recognize where the coastline is, where the mountains are, where the clouds are, right? We can learn all of this, obviously you can convert that image into a vector, you can convert a video into a vector, 3D into a vector, protein into a vector, because protein obviously has structure, chemicals into vectors, genes eventually into vectors, we can learn everything as vectors.

If you can digitize everything and it makes sense, then obviously you can turn the word “cat” into an image, which is clearly not an image of a cat, but it has the same meaning. If you can convert words into images, it’s called stable diffusion in the middle of the journey. If you can convert images into words, it’s called subtitle production, like the subtitles below YouTube videos. Now, if you can convert, what’s your name again? If you can convert amino acids into proteins, that’s called the Nobel Prize, because it’s alpha folding, an incredible breakthrough.

So, this is an amazing moment in computer science. We can truly transform, transmit, and generate one form of information into another, so you can go from text to text, large amounts of text, PDFs to small amounts of text, summarizing archives. It’s something I really love, right?

We can ask it to summarize a research paper instead of reading every single paper. It has to understand images because research papers have a lot of images, charts, and such. So you can summarize all of that. And now, you can imagine all the productivity benefits. It’s actually an ability that you can’t do without, so in the near future, you will do it.

You might say, hey, I want to design, give me some options for cars. I work at Mercedes, and I care about the brand. This is the brand’s style. Let me give you a few sketches, maybe some pictures of car models that I want to produce. It’s a four-wheel-drive SUV, for example. And suddenly, it presents you with a complete 3D design CAD from 2010, 200 versions. Now, the reason you want this instead of just the finished car is that you might want to choose one and then iterate 10 times on that. You might end up choosing one and making modifications on your own. So the future of design will be very different. The future of everything will be very different. And if you give designers this ability, they will go crazy. They will love you. That’s why we’re doing it.

So, what impact does this have in the long run? One of my favorite fields is if you can describe a protein in language, you can find ways to synthesize proteins in language, then the future of protein engineering is right in front of us. As you know, protein engineering involves manufacturing enzymes to break down plastics, to capture carbon, to grow vegetables better with various enzymes. Your generation can create various enzymes, so the next 10 years will be incredible. We are the generation of computer chip engineering, and you will be the generation of protein engineering, something we couldn’t imagine a few years ago.

Costis Maglaras:

Okay, I think we will open up the floor for questions from the audience, so if you have any questions, maybe I can point, and we’ll have some microphones passed around. Okay, let’s start from over there.

Audience:

Thank you for coming tonight. Are you worried that Moore’s Law will catch up with the GPU industry like it did with Intel? Can you explain the difference between Moore’s Law and Huang’s Law? Jensen Huang: I didn’t propose Huang’s Law, and it’s not something I would do. Moore’s Law states that performance doubles every 18 months. A simpler way to calculate it is that it grows 10 times every 5 years, so roughly 100 times every 10 years. Now, if general computing is like a microprocessor and it grows 10 times every 5 years and 100 times every 10 years, why change the calculation to grow 100 times every 10 years? Isn’t that fast enough? Are you kidding me? If cars grew 100 times every 5 years, wouldn’t life be great?

So the answer is, in reality, Moore’s Law is very good, and I have benefited from it. The whole industry has benefited from it. The computer industry exists because of it. But ultimately, Moore’s Law for general computing is not about the number of transistors, it’s about how you use those transistors for the CPU and how you ultimately convert that into performance. That curve is no longer growing 10 times every 5 years. If you’re lucky, that curve is growing 2 to 4 times every 10 years. The problem is, that curve is growing 2 to 4 times every 10 years.

Our computational needs and the vision we have for solving problems with computers, our imagination in using computers to solve problems, has it exceeded the 4 times growth every 10 years? So our imagination, our needs, the world’s consumption of all these things, it has exceeded that limitation. You can solve this problem by buying more CPUs, you can buy more, but the problem is, these CPUs consume too much energy because they are general-purpose, like a jack-of-all-trades. A jack-of-all-trades is not as efficient as an expert, they don’t have the same level of productivity as an expert. If I’m going to have open-heart surgery, don’t give me a jack-of-all-trades, you know what I mean? If you’re around, just call an expert. So the journalist’s way is too blunt, and now it’s causing the world to consume too much energy, to spend too much, all just to forcefully drive general-purpose computing.

Now luckily, we have been researching accelerated computing for a long time. As I mentioned, accelerated computing is not just about processors, it’s actually about understanding the application domain and then creating the necessary software, algorithms, architectures, and chips. We somehow found a way to do this with an architecture that is both very fast, sometimes accelerating CPUs by 100 to 500 times, and even sometimes 1000 times, but also not so specific to just one activity. Is that reasonable? And you need to be broad enough that you have a big market, but you need to be narrow enough that you can accelerate applications. That delicate balance, that razor’s edge, is why NVIDIA exists. If I explained this 30 years ago, nobody would believe it, and in fact, if you honestly say it now, nobody would believe it either.

We’ve spent a long time, we’ve persevered, starting with earthquake processing, molecular dynamics, image processing, and of course, computer graphics. We’ve constantly strived, constantly worked hard, constantly put in the effort – weather simulation, fluid dynamics, particle physics, quantum chemistry, and then one day deep learning, followed by transformers. Next, there will be some form of reinforcement learning transformer, and then there will be some multi-step reasoning system. So, all of these things, we’re just an application.

Somehow, we found a way, created an architecture that solves all of these problems. Will this new law end? I don’t think so. Because, you see, it doesn’t replace the CPU, it complements the CPU. So, the question is, what comes next to complement us?

We simply connect it next to us, so when the time comes, we’ll know what other tool to use to solve the problem, because we are in service of the problem we’re trying to solve. We’re not trying to create a knife for everyone to use. We’re not trying to create a pair of pliers for everyone to use. We are here to accelerate computing to serve the problem, so that’s one thing for all of you to learn. Make sure your mission is right. Make sure your mission is not to build trains but to facilitate transportation, does that make sense? Our mission is not to build GPUs. Our mission is to accelerate applications to solve problems that ordinary computers cannot solve. If your mission statement is correct, focusing on the right things, it will continue to last forever.

Audience:

Thank you again. Now there’s a trend pushing for the localization of the semiconductor supply chain, as well as restrictions on exporting high-tech products to certain countries. What impact do you think this will have on NVIDIA in the short term, and what impact will it have on us consumers in the long term?

Jensen Huang:

That’s a great question. You’ve all heard this question, and I’ll repeat it. It involves geopolitics and geopolitical tensions, etc. Geopolitical tensions, geopolitical challenges impact every industry, impact everyone. Our company deeply believes in national security. We’re here because our country is secure, and we also believe in economic security.

The fact is, most families don’t wake up in the morning and say, oh my goodness, I feel so vulnerable because of the lack of military power. They feel vulnerable because of economic survival. So, we also believe in human rights, the right to create prosperous lives is part of human rights. As you know, the United States believes in human rights for people who live here and people who don’t live here, so this country believes in all of these things at the same time. We do too.

The challenge of geopolitical tension is that if we unilaterally decide, we decide others’ prosperity, there will be a backlash. There will be unintended consequences, but I’m optimistic. I hope, for hope’s sake, that those who are thinking about this issue consider all the consequences and unintended consequences, but it has led every country to deeply internalize its sovereign rights. Every country is talking about their own sovereignty, which is another way of saying everyone’s thinking about themselves.

As far as we’re concerned, on one hand, it may restrict our technology usage in China and export controls there; on the other hand, due to sovereignty and every country wanting to establish its own sovereign AI infrastructure, and most of them not being enemies of the United States or having difficult relations with the United States, we will help them establish AI infrastructure around the world.

So in many ways, this strange thing about geopolitics limits our market opportunities to some extent. On the other hand, it opens up market opportunities for us in other ways, but for people like me, I really hope so.

I really hope that we don’t let our tense relationship with China turn into a tense relationship with the Chinese people, and we don’t let our tense relationship with the Middle East turn into a tense relationship with Muslims. We can’t trap ourselves into that, I’m a little worried it’s a slippery slope.

One of the biggest sources of our nation’s intellectual property, as you know, is foreign students, and I see a lot of them here. I hope you stay here, that’s one of the greatest advantages of our country. If we don’t allow the smartest minds in the world to come to Columbia and stay in New York City, we will not be able to retain the greatest intellectual property in the world, so that’s our fundamental core advantage, and I really hope we don’t destroy it.

You can see that geopolitical challenges are real, national security issues are real, but economic, market, social, and technological issues are also equally real. Technical leadership is important, market leadership is important, all of this is important, the world is just a complex place, I don’t have a simple answer, we’re all going to be affected.

Audience:

I started as an engineer at a semiconductor company, got into entrepreneurship, and like you, as someone who is essentially a technical expert and engineer, successfully started a company. I learned about finance from YouTube videos, what are your thoughts on getting an MBA?

Jensen Huang:

I think that’s great. First of all, you might live to be 100 years old, so the question is, how do you plan to spend the last 7 years or 60 years? This is not just for you, it’s for everyone, do your best to care about education.

When you come here, you’re forced to get an education, how great is that? After leaving, like me, I have to search the world for knowledge, sift through a lot of garbage to find something good; in school, you have these amazing professors who filter knowledge for you and present it to you like on a platter, my goodness, if I could do it all over again, I would stay here as long as possible, absorb a lot of knowledge.

I would still be here, sitting with the dean. I’m the oldest student here. I’m just preparing for that giant leap on graduation day, to immediately step into success, but I’m just kidding. You have to leave eventually. Your parents will appreciate it, but don’t rush it. I think the more you learn, the better. There’s no one right answer to get there.

Obviously, I have friends who have never graduated from college but are very successful, so there are multiple ways to get there, but statistically speaking, I still think this is the best way to get there, so if you believe in statistics and mathematics, stay in school, complete the whole process, so I got a virtual MBA through hard work, not because of choice, but because when I just graduated from school, I thought I would become an engineer, no one would say, “Hey, Jensen, here’s a diploma, you’ll be a CEO.” I didn’t know, so when I got there, I had to learn.

There are many different methods for MBA and studying business strategies, obviously business issues are very different things, financial issues are too, so you have to learn all these different things in order to start a company, but if you are surrounded by amazing people like me, they will teach you along the way, so some things are crucial depending on the role you want to play, and for a CEO, some things are not just my job, but crucial, and I have to take the lead. That is character, your character has something to do with the choices you make, how you deal with success, how you deal with failure and huge setbacks, how you make decisions, these things are important.

Now, from a skills and craft perspective, the most important thing for a CEO is strategic thinking, there is no other choice. The company needs you to have strategic thinking because you see the most, you should be able to see the future better than anyone else, connect the dots better than anyone else, you should be able to mobilize; remember what strategy is – action! It doesn’t matter what you say, what matters is what you do, so no one can mobilize a company better than a CEO. Therefore, the CEO is uniquely positioned to be the Chief Strategy Officer, if you so desire. From my perspective, these two are the most important things, there are many other skills and things you will learn.

If I could add one more thing, I truly believe that a company is about a certain craft, making some unique contribution to society, you create something. If you create something, you should be good at it, you should appreciate this craft, you should love this craft, you should know something about this craft, where it comes from? Where is it now? Where will it go in the future? You should strive to embody a passion for this craft.

I hope today I have done something to embody a passion for and expertise in this craft, I know a lot about the field I am in, and if possible, a CEO should know about this craft. You don’t have to create this craft, but it’s best to be familiar with it, you can learn a lot, so if you want to be an expert in this field, these are some things. You can learn these here. Ideally, you can learn these at work, you can learn from friends, you can learn these by doing many different things.

– This HTML code represents an empty paragraph tag.

We will continue to update Blocking; if you have any questions or suggestions, please contact us!