By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
a16zJensen Huang & Arthur Mensch: Why Every Nation Needs Its Own AI Strategy
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
EVERY SPOKEN WORD
55 min read · 10,739 words- 0:00 – 0:39
The Impact of AI on National Infrastructure
- SSSteph Smith
This is the greatest force of reducing the technology divide the world's ever known.
- SPSpeaker
It will have an impact on GDP-
- SSSteph Smith
Mm-hmm
- SPSpeaker
... of every country, uh, into double digits in the coming years.
- SSSteph Smith
Nobody's gonna do this for you. You've gotta do it yourself.
- SPSpeaker
It's up to organizations, to enterprises, to countries to, to, uh, to build what they need.
- SSSteph Smith
Hmm. The stakes at play are basically the equivalent of modern digital colonialization. AI isn't just computing infrastructure, it's also cultural infrastructure. [upbeat music] Today, we're talking about sovereign AI, all things national infrastructure, um, and open source.
- 0:39 – 2:40
Is AI a General-Purpose Technology?
- SSSteph Smith
And so let, let's just start with the first question I usually get from nation-state leaders, which is: Is AI actually a general purpose technology? Right. So in the, in the history of humanity, we've had maybe a handful of these, 22, 24 economists call these specific technologies that accelerate economic progress broadly across society, right? Electricity, the printing press. And the question everybody's asking right now is: Is that the right way to think about AI? Or wh-why isn't AI just another important but ultimately narrow technology?
- SPSpeaker
It is a general purpose technology, uh, because, uh, it basically revisits entirely the way we are building software and the way we are using machines. Uh, and so in the same way that internet was a general purpose technology, AI is a general purpose technology here. It allows to build agents that are doing things on your behalf. And in that respect, it can be used in any vertical, uh, of the industry. It can be used for services, for public services. Uh, it can be used to change the life of citizens. Uh, it can be used for agriculture. It can obviously be used for, uh, defense purposes. So it covers everything that a state needs to worry about. And so in that respect, it's very natural that, uh, any state makes it a priority and makes it, uh, uh, and find a dedicated national AI strategy.
- SSSteph Smith
By the way, everything Arthur said is 100% correct. It is also exactly the reason why everybody's given up, and it's precisely wrong. And the reason for that is this: uh, if it's a general purpose technology and one company can build the ultimate general purpose tech-technology, then why shouldn't anybody else do it? And that is the flaw.
- SPSpeaker
Right.
- SSSteph Smith
But that's also the mind trick-
- SPSpeaker
Hmm
- SSSteph Smith
... to convince everyone that this is a technology that intelligence is only, you know, something that a few people ought to go build.
- SPSpeaker
Right.
- SSSteph Smith
Everybody ought to sit back and wait for it.
- 2:40 – 3:56
Why Nations Need to Engage AI
- SSSteph Smith
I would advise that everybody engage AI.
- SPSpeaker
Hmm.
- SSSteph Smith
And that it-- this is not, not for the privileged. Uh, intelligence is for everyone, and it is not just a few companies in the world who should build it. Everybody should build it. Nobody's gonna care more about, uh, s- the Swedish culture and the Swedish language and the Swedish people and the Swedish ecosystem more than Sweden. Nobody's gonna care about the, the ecosystem of Saudi Arabia more than Saudi Arabia. And nobody's gonna care about, about Israel more than Israel, despite the fact that the technology is general purpose and a-absolutely true. How could intelligence not be general purpose? Um, it is also, it is also hyper-specialized.
- SPSpeaker
Hmm.
- SSSteph Smith
And the reason for that is because, let's face it, I don't think I'm waiting around for a general purpose chatbot, uh, to be an expert in a particular area of disease. I still think that I would prefer to have somebody who is hyper-specialized in that field, uh, to fine t-fine-tune, to train, you know, and, uh, post-train, if you will-
- SPSpeaker
Ah, right
- SSSteph Smith
... an AI model that's gonna be specialized in that.
- SPSpeaker
It's a general purpose technology the same way a programming language is a general purpose technology.
- 3:56 – 7:47
AI and Cultural Infrastructure
- SSSteph Smith
Hmm.
- SPSpeaker
Uh, and in addition to that, it's also a culture, uh, carrying technology. So I think what that means is that there's an infrastructure. There are chips that obviously not every country are going to build. There are, uh, general purpose models like base models, compression of the web, uh, that are eventually going to be open source-
- SSSteph Smith
Hmm
- SPSpeaker
... and that can serve as the, the right basis for constructing, uh, specialized systems. Uh, but beyond that, uh, I think it's up to organizations, to enterprises, to countries to, to, uh, to build what they need.
- SSSteph Smith
Hmm.
- SPSpeaker
So the way to make it work is to take a general purpose model, uh, like an open-source model, for instance. And to get the knowledge you have specifically or ask, uh, your citizens or ask your employees to distill their knowledge into the systems, into the agents that are going to be working on your behalf, so that m- so that progressively, those agents become more accurate and more following the instructions, uh, and the, the, the specifications that a country or an enterprise may have when it's building, uh, AI systems. So you need vertical experts, or you need cultural experts, or you need people with a certain, uh, um, national agenda-
- SSSteph Smith
Hmm
- SPSpeaker
... to partner with, uh, technological companies that can expose the open-source infrastructure in a way that is easy to use and in a way that is easy to specialize. So I think that's where the, the frontier lies. It's a very horizontal technology.
- SSSteph Smith
Hmm.
- SPSpeaker
But to make anything useful out of it, you need the partnership between the horizontal providers and the vertical experts.
- SSSteph Smith
But unlike previous general purpose technology waves in history, like electricity or the, or the printing press, how, how is this one different? If I'm a, if I'm a nation-state leader and I'm trying to understand what the right framework is for me to think about AI in my country, should I think about it like digital labor, or should I think about it as, as-
- SPSpeaker
Uh, again, to bridges?
- SPSpeaker
I think it's similar to electricity in the sense that it will have an impact on GDP-
- SPSpeaker
Hmm
- SPSpeaker
... uh, of every country, uh, in the double digits-
- SPSpeaker
Hmm
- SPSpeaker
... in the coming years. So that means that from an economical point of view, uh, every nation needs to worry about it.
- SPSpeaker
Hmm.
- SPSpeaker
Because if they don't manage to set up infrastructure, to set up their own sovereign capacities at the right place, that means that this is money that might flow back to other countries, so that's kind of changing the economic equilibrium across the world. In the sense that's not very different from electricity, which 100 years ago, if you weren't building electricity factories, uh, you were, you, you, you were preparing yourself to buy it from your neighbors. Uh, which at the end of the, of the day isn't great because it creates some, uh, some dependencies. So that's-- I think in that sense, it is similar. What is fairly different, I think there's two things. First of all, it's, uh, it's kind of an amorphic technology.
- SPSpeaker
Hmm.
- SPSpeaker
Uh, if you wanna create, uh, digital labor with it, you, you need to shape it, you need to have infrastructure, talent, and software. And the talent needs to be created locally. I think this is-
- SPSpeaker
Hmm
- SPSpeaker
... quite important.
- SPSpeaker
Hmm.
- SPSpeaker
And the reason for that is that in contrast with electricity, this is a content-producing technology. So you have, uh, agents that are producing content, that are producing text, producing images, producing voice, interacting with, with people. And when you're producing content and interacting with the so-society, you become a social construct.
- SPSpeaker
Hmm.
- SPSpeaker
And in that respect, uh, social construct carry cultures and, and carry, uh, the values of, uh, of either an enterprise or a country. And so if you want those values not to disappear and not to depend on, on a central provider, you need to engage with it more profoundly than you would be to engage with electricity, for instance.
- SPSpeaker
Would you agree with that, Jensen?
- JHJensen Huang
Couple of ways to
- 7:47 – 11:49
The Role of the Digital Workforce
- JHJensen Huang
think about it. Your country's digital intelligence is not likely something you would want to outsource to a third party without some consideration. Your digital intelligence is just now a new infrastructure for you. Your telecommunications, your healthcare, your education, your highways, your electricity. Now there's a new layer. This new layer is your digital intelligence. It's your responsibility to decide how you want this digital intelligence to evolve, and whether you wanna outsource it so that you could never have to worry about intelligence again, uh, or this is something that you feel you want to engage, uh, maybe even control and shape, uh, into, into a national infrastructure. Of course, it has all the things that Arthur said, AI factories, infrastructure, et cetera, et cetera. There's another way you could think about it is your digital workforce.
- SPSpeaker
Hmm.
- JHJensen Huang
Now, this is a new layer, and you've got to decide whether the digital workforce of your country or your company is something that you decide to outsource, hope it evolves the way that you would like it to, or is it something that you want to engage, maybe even decide to control and nurture and make better? We hire general purpose employees all the time. We hire them out of school. Some of them are more general purpose than others. Some of them are more intelligent than others. But once they become our employees, we decide to onboard them-
- SPSpeaker
Hmm
- JHJensen Huang
... train them, guardrail them, evaluate them, continuously improve them. We make the investment necessary to make general purpose intelligence-
- SPSpeaker
Right
- JHJensen Huang
... into super intelligence that we could benefit from. And so I think that that second layer, thinking about it as digital workforce, in both cases, uh, it contributes to the national economy. In both cases, it contributes to s-social advance. In both cases, it contributes to the culture. And I think that in both cases, a country needs to play a very active role in it.
- SPSpeaker
Hmm.
- JHJensen Huang
And so I, I think it, it... It's back to your original question about, about sovereign AI, how to think about it. Yes, it is definitely a general purpose technology, but you have to decide how to shape it. Your country's digital data belongs to you. Your, uh, national library-
- SPSpeaker
Right
- JHJensen Huang
... your history. For so long as you want to digitize it, you could make it available to everybody in the world. You could also make it available to companies or researchers and institutions in your own country. It belongs to you. Of course, these are all vaporous things. They're very soft ideas. Um, but it, it does belong to you. And you could decide. It belongs to you in the sense that, that this is where you came from. It belongs you-- belongs to you in the sense that you could decide, um, how to put it to use for the benefit of your, uh, your people, and it belongs to you in the sense that, uh, it's your responsibility to shape its future.
- SPSpeaker
Hmm.
- JHJensen Huang
You know, sovereign AI. It's your responsibility.
- SPSpeaker
There are several other types of assets that nation-states fund and protect: the military, um, your, your electricity grid. If, let's say, I, I, I've understood now the criticality of AI infrastructure and sovereign AI, do I have to now take control of every part of the stack?
- SPSpeaker
Jensen mentions, uh, I guess digital workforce.
- 11:49 – 13:18
Specializing AI for National Needs
- SPSpeaker
Right.
- SPSpeaker
And I think it's a very good, uh, analogy- That you need an onboarding platform for your AI workforce, which means you need to be able to customize the models and pour the knowledge that are sitting, uh, in your national libraries into the model so that suddenly speaks better your language.
- JHJensen Huang
Right.
- SPSpeaker
Uh, you need, uh, to be able to put, uh, to, to get your systems to know about your laws-
- JHJensen Huang
Right
- SPSpeaker
... so that suddenly the guardrails that are set when you're deploying an AI software, uh, are, uh, compliant with, uh, everything that you've done before. And so that onboarding platform that requires to customize, to guardrail, to evaluate, uh, and then when noticing that certain things needs to be improved, to fix things, to debug things, uh, that's the platform that, uh, we are building.
- JHJensen Huang
Right.
- SPSpeaker
And once the custom systems are made, uh, it's important to be able to maintain them yourselves. So that means being able to deploy them on your own infrastructure, being able to ask your, uh, technological partners to potentially disappear from the loop.
- JHJensen Huang
Your IT department is going to become the HR department of your digital workforce, and they're gonna use these tools that-
- SPSpeaker
Right
- JHJensen Huang
... Arthur describes-
- SPSpeaker
Mm
- JHJensen Huang
... to onboard AIs, fine-tune AIs, guardrail them, evaluate them, continuously improve them.
- SPSpeaker
Right.
- JHJensen Huang
And that flywheel will be managed by the, uh, the new-- the modern version of the IT department.
- 13:18 – 27:47
Building a Digital Workforce
- SPSpeaker
Right.
- JHJensen Huang
And we'll have biological workforce, and we'll have a digital workforce, and it's fantastic. And so, so nobody's gonna do this for you. You've gotta do it yourself.
- SPSpeaker
Right.
- JHJensen Huang
That's why even though we have so many technology companies in the world, every company still has their own IT department. I've got my own IT department. I'm not gonna out-outsource it to somebody else. In the future, they'll be even more important to me because they'll be, you know, helping us manage these digital workforces. You're gonna do this in every country. You're gonna do this in every company within those countries. And so, so the space for what Arthur is describing, uh, to take this general purpose technology, but to really fine-tune it-
- SPSpeaker
Right
- JHJensen Huang
... into domain experts. They're national experts, or they're industrial experts, or they're corporate experts. These are functional experts. This is the future, the giant future space of AI.
- SPSpeaker
Well, so you both said something that I just wanna make sure I'm understanding correctly. You, you called it, uh, soft, a soft concept like your culture, and you said there are a bunch of norms that the training data has that you customize the models on. It sounds like you're saying, unlike compute, storage, networking, which have been-
- JHJensen Huang
By the way, you said norms. That exactly means it's soft versus-
- SPSpeaker
Right
- JHJensen Huang
... rules, which are more hard, you know?
- SPSpeaker
Ah, right.
- JHJensen Huang
Or, or algorithms and laws, which are very specific.
- SPSpeaker
Yeah. There, there's different things that you wanna incorporate into your AI systems.
- JHJensen Huang
Right.
- SPSpeaker
There are some elements of style and of, um, of knowledge that you're not going to, to enforce through strict guardrails.
- JHJensen Huang
Right.
- SPSpeaker
That you c- you can enforce through a continuous training of models, for instance. You take, uh, the national library-
- JHJensen Huang
Preferences. Right
- SPSpeaker
... yeah, you take preferences, and you, you, you, you distill it into the models themselves. And then you have a set of laws, uh, you have a set of policies if you're in a company, and those are strict. And so usually the way you build it is that you connect the models to the strict rules, and you make sure that every time it answers, it, you verify that the rules are respected. So there's really two things. So o- on one side, you're pouring and compressing knowledge in a soft way into the models, and on the other side, you're making sure that you have a certain number of policies and, and rules that are strictly enforced and that have a hundred percent accuracy. So o- on one side, this is soft, this is preference, this is culture.
- JHJensen Huang
Preference.
- SPSpeaker
Yeah.
- JHJensen Huang
Somebody's preference is multidimensional. You know, you know, what you prefer-
- SPSpeaker
But it's not-
- JHJensen Huang
... it depends
- SPSpeaker
... it's implicit m-many times in, in communication.
- JHJensen Huang
Well, there's just so many, there's so many features that, that defines my preference. It takes AI to be able to precisely, um, comply with the description that Arthur was describing just now. Could you imagine if a human had to write this in Python? You know, describe every one of these, capture every one of these things, uh, in C++? You know, based on this, I prefer that, but if you did that, I prefer that o- other thing. And I mean, the number of rules would be in- in- insane.
- SPSpeaker
Right.
- JHJensen Huang
Which is the reason why AI has the ability to kind of codify all of this. It's a, it's a, it's a new programming model that can deal with the ambiguity of life.
- SPSpeaker
Well, it sounds like you're saying AI isn't just computing infrastructure, it's also cultural infrastructure. Yes, it is.
- JHJensen Huang
Is that right?
- 27:47 – 31:02
Challenges and Risks of AI Adoption
- SPSpeaker
So let's talk about that for a second because change is hard. I've got an endless list. If I'm a nation state leader, I'm facing increasing amounts of geopolitical risk. My, my-- I don't know who my allies are. Elections are coming. There's any, n-number of things I've got to, uh, to deal with. But now I-- let's say I understand that, that this is important. You guys spend so much time talking to nation state leaders who are thinking about: what are the risks of adopting AI too fast? And, and you're right, the zeitgeist has shifted based on the Paris Action Summit. It seemed like people are... There's, there's a more-- a tone of optimism more than there was a tone of pessimism a year ago. But what are the most common, uh, questions you get from nation state leaders when they're asking about risks and how to think about them?
- SPSpeaker
So I've heard several questions. Uh, but one of the risks is to see your population start, uh, getting afraid-
- SPSpeaker
Mm-hmm
- SPSpeaker
... of the technology for fear of it replacing them.
- SPSpeaker
Hmm.
- SPSpeaker
And, and that, that is something that can actually be prevented, uh, if, uh, we collectively make sure that everybody get access to the technology and is trained in using it.
- SPSpeaker
Right.
- SPSpeaker
So the, the skilling of, of the, of the various citizens of the populations is extremely important. And stating AI as an opportunity for them to actually work better and showing actually the-
- SPSpeaker
Right
- SPSpeaker
... the, the purpose of it, uh, through applications, through things that they can actually install on their smartphone, uh, through public services. Uh, we're working, for instance, with the French un-unemployment system-
- SPSpeaker
Hmm
- SPSpeaker
... uh, to actually connect opportunities of jobs to unemployed, unemployed people through AI agents that are being actionated by, obviously, human operators within the, within the agency. And that is one opportunity. That's a very palatable opportunity for people to find a job better.
- SPSpeaker
Right.
- SPSpeaker
And so that's part of the thing that, uh, can make sure that population understand the opportunity and the fact that AI is really just a new change for them to adopt, just the same way they had to adopt personal computers in the '90s and internet in the, in the two-- 2000s. The common aspect with these changes is that you need people to, uh, embrace the technology. And I think the, the biggest problem that nation states may have is to see AI increase the digital divide, uh, that is a-already relatively big. But if we work together, and if, if done in this, in the right way, we can make sure that AI is actually reducing the digital divide.
- JHJensen Huang
AI is a new way to program a computer. It is because by typing in some words, you can make the computer do something. Just like we did in the past.
- SPSpeaker
Right.
- JHJensen Huang
We used to type words, and we made computers do something.
- SPSpeaker
And now you talk to it. Yeah.
- JHJensen Huang
Right now you can talk to it, you could do... You can interact with it in a whole lot of ways. You can make the computer do things for you a lot easier today than it was before.
- SPSpeaker
Right.
- JHJensen Huang
The number of people who could prompt ChatGPT and do productive things-
- SPSpeaker
Right
- JHJensen Huang
... just from a human potential perspective, is vastly greater than the number of people who can program C++ ever. And therefore, we have closed the technology
- 31:02 – 32:51
Is AI the Greatest Equalizer?
- JHJensen Huang
divide.
- SPSpeaker
It's probably the eq- greatest equalizer we've seen.
- JHJensen Huang
It is by definition-
- SPSpeaker
Right
- JHJensen Huang
... the greatest equalizer of technologies-
- SPSpeaker
Right
- JHJensen Huang
... of all time.
- SPSpeaker
But you still need to have, uh, citizens to know about it. I think that's the thing. Yeah.
- JHJensen Huang
I'm just describing the fact.
- SPSpeaker
Yes.
- JHJensen Huang
I'm describing a fact. The fact is, there are more people who program computers using ChatGPT today than there are people who program computers using C++.
- SPSpeaker
Right.
- JHJensen Huang
That's a fact.
- SPSpeaker
In four short years.
- JHJensen Huang
That's a his- that's a straight-up fact.
- SPSpeaker
Actually, three years.
- JHJensen Huang
Yeah. It's a straight-up fact. And so the fact is, this is the greatest force of reducing the technology divide the world's ever known.
- SPSpeaker
Right.
- JHJensen Huang
It's just perceived, and what Arthur's saying, the perception through, you know, I don't know who I'm talking about it, and I don't know how talking about it, but the fact of the matter is, it is not stopping.
- SPSpeaker
Right.
- JHJensen Huang
It's not stopping anything. The number of people who are actively using ChatGPT today is off the charts. I think it's terrific. It's completely terrific. Anybody who's talking about anything else apparently isn't working. And so I, I think people realize the incredible capabilities of AI and, and how it's helping them with their work. I use it every single day. I used it this morning. And so every single day I use it. And, and I think that, that, uh, deep research is incredible.
- SPSpeaker
Right.
- JHJensen Huang
You know? It-- my goodness, is-- the, the, the work that, that, um, uh, Arthur and, and all of the computer scientists around the world are doing is incredible. And, and, and people know it. People are picking it up, obviously, right? Just the number of active users.
- 32:51 – 34:23
Open-Source Models
- SPSpeaker
Um, let-let's talk about open-source for a bit, because both of you have talked quite publicly about the importance of open models in the context of sovereign AI. At DeepMind, you're part of the Chinchilla scaling laws, which were openly published. Your co-founder Guillaume created Lama. And then last year, NVIDIA and Mistral worked on a jointly trained model called Mistral Nemo. Why are open models such a big part of your focus?
- SPSpeaker
Because it's an horizontal technology and, um, enterprises and states are going to be eventually willing to deploy it on their own infrastructure.
- SPSpeaker
Mm.
- SPSpeaker
Uh, having this, uh, openness is, is important from a sovereignty perspective. That's the first point. And then the second point, uh, of importance is that, uh, releasing open-source models is a way to accelerate progress.
- SPSpeaker
Mm.
- SPSpeaker
And, uh, we created Mistral on the basis that what we've seen during our early career when we were doing AI in between twenty ten and twenty twenty, uh, was an acceleration of progress because every lab was building on top of each other. And that's something that kind of disappeared, uh, with the first large language models-
- SPSpeaker
Mm
- SPSpeaker
... uh, from OpenAI in particular. And so spinning back that open flywheel of I contribute something and then another lab is contributing something else-
- SPSpeaker
Right
- SPSpeaker
... and then we iterate from that is the reason why we created Mistral. And I think we, we did a good job at it because we, we started to release models, and then Meta started to release models as well.
- SPSpeaker
Right.
- SPSpeaker
Uh, and then we had Chinese company, uh, like DeepSeek release, uh, stronger models. And everybody benefit
- 34:23 – 35:47
Collaborative AI Development
- SPSpeaker
from it so coming back to Mistral Nemo of, um, one difficulty of, uh, creating AI models in an open way-
- SPSpeaker
Mm
- SPSpeaker
... is that this is more a cathedral than a bazaar setting-
- SPSpeaker
Right
- SPSpeaker
... uh, when it comes to open-source. Because you have large spend to do to build a model. And so what we did with NVIDIA team is really to, to, to mix the two teams together.
- SPSpeaker
Mm.
- SPSpeaker
Have them work on the same infrastructure, the same code, have the same problems, and combine their expertise to build the same model. And that has been very successful, uh, because NVIDIA brought a lot of things we didn't know. I think we brought things that NVIDIA didn't know. And at the end of the day, we produced w- something that was, at the time, the best model, uh, for its size.
- SPSpeaker
Right.
- SPSpeaker
And so we really believe in such collaborations, and we think that, uh, we should do them more and at a higher scale. And not only with only two, uh, companies, but probably with three or four. And that's the way open-source is going to prevail.
- JHJensen Huang
Yeah. I completely agree. The, the, um, benefit of open-source, uh, in addition to accelerating and elevating the basic science, the basic endeavor, um, of all of the general models and the general capabilities, is the open-source versions also activate a ton of niche markets and niche innovation.
- 35:47 – 37:17
Open-Source AI in Niche Markets
- JHJensen Huang
Uh, all of a sudden, in healthcare, life sciences, physical sciences, uh, robotics, transportation, the number of industries that were activated as a result of open-source capabilities that are sufficiently good is incredible. Don't ignore the, the incredible capabilities of, of open-source, particularly in the fringe, the niche, you know-
- SPSpeaker
Niche but mission-critical where data might be sensitive.
- JHJensen Huang
Yeah. It could be, for example, in mining energy.
- SPSpeaker
Right.
- JHJensen Huang
Who's gonna go create an AI company to go mine energy? Energy is really important, but the mining of energy is not that big of a market.
- SPSpeaker
Mm.
- JHJensen Huang
And so open, open-source activates every single one of them. Financial services, it turns out, activates them.
- SPSpeaker
Healthcare, defense.
- JHJensen Huang
You, you pick your favorites.
- SPSpeaker
Anything that is mission-critical and that requires, uh, to do once Own deployment, uh, that potentially requires to do on pr- on, on the edge deployment as well.
- JHJensen Huang
Right.
- SPSpeaker
Um, and anything that requires, uh, some strong auditing and the ability to do a thorough evaluation of it. You can evaluate a, a model much better if you have access to the weights than if you only have access to APIs. And so if you want to build certainty around the fact that, um, your system is going to be 100% accurate, I don't think you should be using a closed-source model.
- JHJensen Huang
And you have to connect it into your flywheel. How are you gonna connect-
- SPSpeaker
Your local data.
- JHJensen Huang
Yeah.
- SPSpeaker
Yeah.
- JHJensen Huang
You have to connect it into your own-- your local data, your own local experience. The more you use it, the better it becomes, that flywheel. You, you can't do it without open source.
- SPSpeaker
But
- 37:17 – 42:02
Balancing Security and Openness
- SPSpeaker
let's say I'm, uh, I'm a nation state leader. I've, I've been considering open source. I've-- I'm starting to hear things like, "Hey, open source is, is a threat to national security. We should not be exporting our models because these open models actually give away, uh, a ton of, um, nation state secrets, or more importantly, the bad guys can use these open models too. And so this is a threat to security. Instead, what we should be doing is locking down maybe development amongst two or three labs that have the infrastructure to, to, you know, get licenses from the government to do training, to do the right safety and certification." I-I've certainly been hearing that a lot. How should I think about that versus what you're telling me, which is actually, you know, open is, is better for mission-critical industries?
- SPSpeaker
Collaboration in between, uh, labs is going to be, uh, to be, to be critical for, uh, humanity's success. And if, if one state decides to lock things down, what-- the only thing that is going to happen is that another state will take the leadership.
- SPSpeaker
Mm-hmm.
- SPSpeaker
Uh, because, uh, cutting yourself from the open flywheel is just too high-- has just too high of a cost-
- SPSpeaker
Mm
- SPSpeaker
... for you to maintain competitivity if you do that. This is a debate that has occurred in, in, in the United States. And effectively if, um, if there's some export control over weights, uh, this is not going to stop any country, uh, in Europe, any, any country in Asia to, to continue its progress.
- SPSpeaker
Right.
- SPSpeaker
And they will collaborate, uh, to actually accelerate that progress. So, so I think we just need to embrace the fact that this is an horizontal technology, uh, very similar to programming languages. Programming languages, they're all open source, right? So I think AI just needs to be open source in that respect. We're glad to see that, uh, at the AI summit that occurred last week, for instance, uh, this was very much, uh, on the agenda.
- SPSpeaker
Hmm.
- SPSpeaker
Uh, this, uh, realization that, uh, we could accelerate together by being more open about, uh, the way we build the technology. Uh, and, and so it's great to see that open source has a, a lot of good days be-before it.
- JHJensen Huang
It is impossible to control. Software is impossible to control. If you wanna control it, then somebody else's will emerge and become the standard, just as Arthur mentioned. And the question is: Is open source safer? Open source enables more transparency, more researchers, more people to scrutinize, scrutinize the work. The reason why every single company in the world is built, every cloud service provider is built on open source, is because it is the safest technology of all. Give me an example of a public cloud today that's built on an infrastructure stack that isn't open source. You start from open source, you could customize it.
- SPSpeaker
Right.
- JHJensen Huang
Um, but the benefit of open source is the contribution of so many people and the scrutiny. You know, very importantly, you can't just put any random stuff into open source. You'll get laughed off, laughed off the internet. You've got to put good stuff on the open source because the scrutiny is in- is intense. And so, so I think the, the, uh, open source, uh, provides all of that. Great collaboration to accelerate, accelerate innovation, escalate excellence, uh, ensure transparency, attract scrutiny. All of that improves safety.
- SPSpeaker
In a sense, you're saying it's partly more secure because as we've seen with open source databases, storage, networking, compute, you get mass red teaming. The whole world can help you red team-
- JHJensen Huang
Yeah, right
- SPSpeaker
... your technology versus just a small group of researchers inside your company.
- JHJensen Huang
Yeah.
- SPSpeaker
Is that roughly right? A right way to think about it?
- JHJensen Huang
Yeah, yeah, yeah. Exactly. Exactly.
- SPSpeaker
By pooling a lot of organizations together-
- SPSpeaker
Right
- SPSpeaker
... to come up with a technology that they can all use, uh, and specialize on their own domains-
- SPSpeaker
Right
- SPSpeaker
... you're forcing the technology to be good for every one of them. And so that means you're removing biases. You're, you're really making sure that the general purpose models that you're building are, are as good as possible and don't have failures. And I think open source in that respect is, is also a way to reduce the number of failure points. If, um, as a company today, I decide to rely fully on a single organization and on its, uh, safety principles, uh, on its red teaming, uh, organization as well, I'm kind of trusting it a little too much. Whereas if I'm s- building my technology on open source, uh, models, I'm trusting the world to make sure that the basis on which I'm building is secure.
- SPSpeaker
Hmm.
- SPSpeaker
Uh, so that's a reduction of failure points-
- SPSpeaker
Right
- SPSpeaker
... and it's obviously something that you need to do as an enterprise or as a country.
- 42:02 – 48:16
Company Building Insights
- SPSpeaker
Um, we're gonna transition a little bit now into company building, which is something a lot of people are excited to hear from both of you about. So let's start with you, Jensen. You've remarked that NVIDIA is the smallest big company in the world. What allow-- enables you to operate that way?
- JHJensen Huang
I mean, our, our architecture was designed for several things. It was designed, um, to, uh, adapt well in a, in a world of change-
- SPSpeaker
Hmm
- JHJensen Huang
... either caused by us or affecting us. And the reason for that is because technology changes fast. And, and if you, if you overcorrect- On controllability, then you, you're, you're underserving a system's ability to, uh, become agile and to adapt. And, and so our company is, is, uh, uses words like aligned instead of use words like control. Um, you know, I don't, I don't know that one time I've used the word control, um, in, in talking about the way that the company works. Uh, we care about minimum bureaucracy.
- SPSpeaker
Hmm.
- JHJensen Huang
And, uh, we wanna make our processes as lightweight as possible. Now, all of that is so that we can im- enhance efficiency, enhance agility, and so on and so forth. Uh, we avoid things like, uh, words like division. When NVIDIA was first started, it was, it was, it was modern to talk about divisions.
- SPSpeaker
Right.
- JHJensen Huang
And I hated the word divide. You know, why would you create an organization that's fundamentally divided? I hated the word business units. The reason for that is because why should anybody exist as one? Uh, why don't, why don't you leverage as much of the company's resources as possible? I-- You know, I wanted a, a system, uh, that was organized, uh, much more like a computing unit, like a computer, uh, to, uh, uh, deliver on an output, uh, as efficiently as possible. Um, and so the, uh, the company's organization looks a little bit like a computing stack.
- SPSpeaker
Hmm.
- JHJensen Huang
And what is this mechanism that we're trying to create, and in what environment are we trying to survive in?
- SPSpeaker
Hmm.
- JHJensen Huang
You know, is this much more-- is, is this much more like a, you know, peaceful countryside, or is this, like, much more like a concrete jungle? You know, what kind of environment are you in? Because the, the type of, the type of, um, system you wanna create should be consistent with that. And the thing that always strikes me odd is that every company's org chart looks very similar-
- SPSpeaker
Right
- JHJensen Huang
... but they're all different things, you know. One's a snake, the other one's a, a elephant, the other one's a, you know, uh, you know, a cheetah. And everybody is, is supposed to be some-somewhat different in, in that, in that forest, but somehow they all same exact, you know, structure, same exact organization. Doesn't seem to make sense to me.
- SPSpeaker
Hmm.
- SPSpeaker
I agree that, uh, it feels like companies have personalities and, uh, and despite the fact that they're organized sometimes similarly. I should say that, uh, obviously we, we have a lot of things to learn, and I mean, the company is not, not even two years old. So I guess one, one challenge we have with Mistral, and I think, uh, our competitors have the same challenge, is that this is the-- one of the first time that a software company is actually a deep tech company, uh, that is driven by science. Science doesn't have the same timescales as software.
- SPSpeaker
Right.
- SPSpeaker
Uh, you need to operate, uh, on a monthly basis. Uh, sometimes you don't know, uh, exactly when the thing will be ready. Uh, but on the other hand, you have customers, uh, that are asking, uh, "When is the next model coming up? When is this capability going to be available?" Et cetera. And so you need to manage expectation. And I think for us, the, like the, the biggest challenge, and I think we're starting to do a good job at it, is to manage the, the hinge in between, uh, the product requirements and what the science is able to do.
- SPSpeaker
Research and-
- SPSpeaker
Yes
- SPSpeaker
... and product.
- SPSpeaker
Research and product. And you don't want the research team to be fully dedicated to making the product work. You need to work, and I think we, we, we've started to do it, on making sure that you have several frequencies in your company. You have fast frequencies on the product side, iterating every week. Uh, and you have slow frequencies on the, on the science side that are looking at why profoundly the product is failing on certain domains and how they could fix it through research, through new data, through new architecture, through a new paradigm. Um, and that's-- I think that's fairly new. This is not something that you would find in a typical SaaS company, uh, because, because this is inherently a science problem.
- SPSpeaker
I mean, NVIDIA is one of the most successful companies that have, over a thirty-year timeline, has figured out a way to keep science and research ahead of the rest of the world, whether it was CUDA back in 2012, where that was fundamental systems research, or Cosmos today, which is now saying, you know, is, is definitely state-of-the-art on how simulation should work out.
- JHJensen Huang
We've harmonized exactly what Arthur just said.
- SPSpeaker
Is, is that, is that heuristic right for you?
- JHJensen Huang
Sure. Yeah. We harmonize that inside our company. We have, we have, we have basic research, applied research, and then we have architecture, and then we have product development. And we have multiple layers of it. And, and these layers are all essential.
- SPSpeaker
Right.
- JHJensen Huang
And they all have their own time clock. In the case, in the case of basic research, the, the time-- the, the, the frequency is-- could be quite low. And yet, on the other hand, all the way to the product side, we have a whole industry of customers who are counting on us, and so we have to be very precise.
- SPSpeaker
Hmm.
- JHJensen Huang
And somewhere between basic research and discovering, discovering, uh, hopefully, uh, surprises that nobody expects-
- 48:16 – 54:11
Navigating Competitive Partnerships
- SPSpeaker
that are also your competitors, right? So-- And, and those competitors are huge and highly capitalized tech giants. NVIDIA sells GPUs to AWS, which is building its own chips called Trainium. And Arthur, your, you know, your training models that you sell through AWS and Azure, who have funded labs like Anthropic and OpenAI. So how do you win an environment like this, and how do you manage those relationships? Because we talked about company building internally, but now I'm curious externally, how do you survive in this, in, in this situation?
- SPSpeaker
Jensen said it well, uh, you give up control, but you work on alignment.
- SPSpeaker
Mm-hmm.
- SPSpeaker
And, uh, despite the fact that sometimes, uh, you have, um, uh, like certain companies can be competitors, you may have aligned interests, and you can work on, on specific agendas that are shared.
- SPSpeaker
Mm-hmm.
- JHJensen Huang
You have to have your own place. Um, obviously, obviously, these cloud service providers aren't working with Arthur because they already have the same thing. They just want two of the same things. It's because Arthur and Mistral has a position in the world that is unique to Mistral, and they add value in a particular place that is unique. A lot of the conversation we've had today, um, are, are areas that, that Mistral and the work and, and their position in the world, um, uh, makes them uniquely good at. And, uh, we are different. You know, we're not just another, um, ASIC or... And we, we can do things for the CSPs and do things with the CSPs that are not possible for them to do themselves. For example, NVIDIA's architecture in every-- is in every cloud. And in a lot of ways, we are their first onboarding for, uh, amazing future startups. And the reason for that is because by onboarding to NVIDIA, they're-- they don't have to make a strategic or business or otherwise commitment to a maj-major cloud. They could go into every cloud, and they could even decide to build their own system if they like, uh, because the economics turns out to be better for them at some point, or they would like access to capabilities that we have-
- SPSpeaker
Right
- JHJensen Huang
... that are somewhat protected within the clouds. And so, so whatever, whatever the reasons are, um, in order to be, to be a good partner to somebody, you still have to have a unique position. You need to have a unique offering.
- SPSpeaker
Mm.
- JHJensen Huang
And, and I think Mistral has a, has a very unique offering. We have a very unique offering, and, and our position in the world is, is, um, uh, uh, important to, uh, even the people we compete against. And so, so I, I think when we, when we are comfortable within that and comfortable with our own skin, then we can be excellent partners to all of the, all of the CSPs. And, and then we wanna see them succeed. Um, uh, I know that it's a weird thing to say when you see them as a competitor, which is the reason we don't see them as a competitor.
- SPSpeaker
Right.
- JHJensen Huang
We see them as a collaborator who happens to compete with us as well. And so, uh, you know, probably the single most important thing that we do for all the CSPs is bring them business.
- SPSpeaker
Right.
- JHJensen Huang
And, and, uh, that's what a great computing platform does. We bring people business.
- SPSpeaker
I remember when, when Arthur and I first met, we sat down in London at a late-night restaurant and kind of sketched out the plan for his Series A.
- JHJensen Huang
Mm-hmm.
- SPSpeaker
And we were kind of figuring out what he would n-why he needed so much capital for the Series A.
- JHJensen Huang
Mm-hmm.
- SPSpeaker
Which in hindsight was remarkably efficient.
- JHJensen Huang
Yeah.
- SPSpeaker
I think, uh, the Mistral Series A we put together was half a billion relative to other folks who had to spend multiple billions to get to the same place. But I asked him, "You know, what chips would you like to run on?" And I, I don't even think you-- You looked at me so absurdly as if I had asked you a question that how could it be an answer other than NVIDIA, other than H100s? And I think that ecosystem has been the startup ecosystem that NVIDIA's invested in, creates so much business for the clouds.
- JHJensen Huang
Yeah.
- SPSpeaker
What is the philosophy that led you to invest so deeply in startups and founders so early on, even before anybody knew about them?
- JHJensen Huang
Hmm. Um, two reasons I would say. One, the first reason is that I rarely call us as a GPU company. What we make is a GPU, but I, I think of NVIDIA as a computing company.
- SPSpeaker
Hmm.
- JHJensen Huang
If you're a computing company, the most important thing you think about is developers.
- SPSpeaker
Right.
- JHJensen Huang
If you're a chip company, the most important you think-- the thing you think about is a chip. And, and, uh, all of our, all of our strategies, all of our actions, all of our priorities, all of our focus, all of our investments, 100% of it is aligned with the attitude that is developer first. It's about the computing platform first.
- SPSpeaker
Hmm.
- JHJensen Huang
Another way of saying ecosystem.
- 54:11 – 1:01:20
Future Trends in Computing
- SPSpeaker
what are the most significant trends you see on the horizon? And in particular, if you could-- if-- for, for an audience of folks who might be leading-- who might be prime ministers or presidents or ministers of IT in some of the world's fastest-growing markets trying to understand where computing is going, how would you an- how would you guide them?
- SPSpeaker
We are moving to-towards, uh, workloads that are more and more asynchronous. So workloads where you give a task to, uh, an AI system, and then you, you wait for it to do, like, 20 minutes of research before returning. Uh, so that's definitely changing a bit the way, uh, you should be looking at, um- At infrastructure, uh, 'cause that creates more load. Uh, so it's, um, I guess it's a bull case for data centers and for NVIDIA. Uh, we-- As I've said, I guess, in the beginning of this episode, um, all of this is n-not going to happen well if you don't have the s- the right onboarding infrastructure-
- JHJensen Huang
Right
- SPSpeaker
... for the agents, if you don't have a proper way for your AI systems to learn about the people they are interact with and to learn from the people they interact with.
- JHJensen Huang
Right.
- SPSpeaker
So that aspect of learning from human interaction is going to be extremely important in the coming, uh, in the coming years. And there's another aspect which is around personalization of having, I guess, models and systems consolidate the representation of their users to be as useful as possible. I think we are in the early stage of that. Uh, but that's going to change, again, pretty profoundly the, the interaction we have with, uh, machines that will know more about us, uh, and know more about our taste and how to be as useful as possible t-towards us. And as, um... I guess to, to, to, to close on that, I think as a, as a prime minister or as a, as a leader of a, of a country, I, I wanna think about education-
- JHJensen Huang
Okay
- SPSpeaker
... about making sure that I have a, I have a local talent pool, uh, that understands AI enough, uh, to create specialized AI systems. And I want to think about infrastructure, both on the physical side, but also on the, on the software side.
- JHJensen Huang
Mm.
- SPSpeaker
So what are the right primitives, what is the right partner to work with that is going to provide you with the platform of onboarding? And so those two things, uh, are important. If you have this and you have the talent, and if you do deep partnerships, the economy, uh, of your state is going to be profoundly changed.
- JHJensen Huang
The last ten years, we've seen extraordinary change in computing, um, from hand coding to machine learning, from CPUs to GPUs, from software to AI, right? We're, we're across the entire stack.
- SPSpeaker
Right.
- JHJensen Huang
The entire industry is, has been completely transformed, and we're going through that still. The next ten years is going to be incredible. Um, of course, you know, we-- the, the industry has been wrapped up in talking about scaling laws, and pre-training is important, of course, and continues to be. Um, now we have post-training.
- SPSpeaker
Right.
- JHJensen Huang
And post-training is, is thought experiments and practice and tutoring and coaching and, um, all of the skills that we use, uh, as humans to learn. You know, the, the idea that, that, um, uh, thinking and agentic and robotic systems are now just around the corner is really quite exciting. And so, so the amount of what it means to computing is, is, uh, very profound.
- SPSpeaker
Mm-hmm.
- JHJensen Huang
And, uh, uh, and that's where we've spent the last, you know... People are surprised that, that, um, uh, that Blackwall is such a great leap over Hopper. And the reason for that is because we built Blackwall for inference and at a just in time because all of a sudden, uh, thinking is such a s- such a big computing load. And, and so that's one layer is there's a computing layer. The next layer is the, the type of AIs that we're gonna see. Um, there's the agentic AI, the informational, uh, digital w- worker AIs, um, but we now have physics AI that's making great progress, and then there's physical AI-
- SPSpeaker
Mm-hmm
- JHJensen Huang
... that's making great progress. And, uh, physics AI is, of course, uh, things that are, that are, uh, that obey the, uh, the, uh, uh, and understand the, uh, the, uh, the physical laws and, and the atomic laws and the chemical laws and, you know, all of, all of the various physical sciences that we're gonna see some great breakthroughs, and I'm very excited about that. That affects industry, that affects science, uh, h- affects higher education and research and, and then physical AI, AI that understand the nature of the physical world, you know, uh, from, from, from fic- friction to inertia, the cause and effect, and, you know, um, object permanence, those kind of basic things that, that humans have common sense, but most AIs don't. And, and so I think that that's gonna enable a whole bunch of robotic systems that are gonna, uh, have great implications in manufacturing and others. You know, the, the US economy is, is very heavily weighted on, on knowledge workers. And, and yet many of the other countries are very heavily weighted on, on, um, uh, manufacturing.
- SPSpeaker
Right.
- JHJensen Huang
And, uh, and so I, I think for, for many of the prime ministers and, and the leaders of countries to realize that, that, uh, the AIs that they need, uh, to transform and to revolutionize their industries that are so vital to them, whether it's energy-focused or manufacturing-focused, is just around the corner, and they ought to stay very alert-
- SPSpeaker
Mm-hmm
- JHJensen Huang
... uh, to this. And I would encourage people not to, not to, um, overrespect, uh, the technology. And, and sometimes when you overadmire a technology, overrespect the technology, you don't end up engaging it. You're afraid of it somehow. Um, don't, don't... This is, you know... Some of the things that we said today about, about AI closing the technology divide i-is, is gen-- is, is really something that ought to be recognized. This is of such incredible national interest that you have the responsibility to engage it. Anyhow, exciting times ahead.
- SPSpeaker
That was incredible. Thank you both so much for making time. If they want, if they wanna go learn more, they wanna figure out how to partner with, with the two companies-
- JHJensen Huang
Call us. You kidding me?
- SPSpeaker
You can call us, yes. [laughs]
- JHJensen Huang
Am I right?
- SPSpeaker
You can call us.
- JHJensen Huang
The two of us.
- SPSpeaker
Yeah.
Episode duration: 1:01:26
In the uListen app, the transcript lights up word by word as you listen. Get uListen
Transcript of episode Ww9SkW0Em58
