𝗕𝗲𝘆𝗼𝗻𝗱 𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 — Jensen Huang (NVIDIA) and Alex Bouzari (DDN)

Welcome to Beyond Artificial J. Thank you. It's great to be here.

Alex: Thank you very much. Nice to see you.

Good to see you. You look great.

So do you, as always.

Well, thank you. Okay, I have to tell you how Infinia started, this new product. 2017, Nvidia said, “We want to stand up a reference architecture, superPOD, and we need the data part of it.” And we walked away from that meeting. I said, “There there’s got to be a different architecture for AI.” If Jensen’s far-reaching vision turns into reality over the next decade or so, a completely different approach is needed. It has to be something that scales efficiently for the training part of it; it has to be something that is very low latency; it has to be distributed on-premise and multi-cloud; and the payload cannot move because it’s too expensive—images, video, and all that. So it has to be metadata and tagging and so on. So that’s how it started. I drew it on the board and I said, “This is what the architecture should be for AI.” Yeah, and everybody looked at me and said, “You’re completely mad. You’re talking about things that are just cannot be done.” I said, “Well, let’s look at it differently. Let’s step out of the problem and let’s look at what we’re trying to solve here and forget about the past, forget about file systems, forget about all this.” And it took seven years.

Yeah, that’s what happens when two engineer CEOs get together.

I think the extraordinary journey you just described—of course, the journey from uh training a model to the journey now of us uh uh taking advantage of these incredible Frontier models and AI models and turning them to AI applications that are for inference and and solving solving large problems—one of the most important things people forgot uh is the importance of uh data that is necessary uh during application, not just doing training. And so so of course you want to train on a vast amount of of data for pre-training.

Mhm.

But during use, the AI has to access information, and uh AI would like to access information not in raw data form but in informational form. Yeah, and so this is the reason why the the uh the reframing of storage of objects and raw data into Data Intelligence, which is this new opportunity for DDN and providing Data Intelligence for all of the world’s Enterprise as AIs run on top of this fabric of information, uh it is just a a an extraordinary reframing of computing of computing and storage, um of the relationship between us and the the type products that you can offer to the world. So so the way we look at it, and again all of it is driven I think by your vision, and I mean you’ve basically created this industry, and in order for this industry industry to bear fruit, it it’s transforming everything we do—right at work, at play, Leisure, Health, security, safety. So so it’s profound and it’s Global. It’s a pivot of the global economy in every way. And so for that to happen, uh Enterprises need to adopt it, and they need to adopt it at an accelerated pace, which means the ROI has to work out. In order for the ROI to work out, the application layer has to be accelerated, and the infrastructure which resides in data centers or in the clouds has to be made efficient. That part of it we’ve done. Yeah. Uh now I think the big enabler and the big big accelerator is how do we supercharge the application layer, which you’ve been doing very very extensively with CUDA.

Yeah. I I really appreciate you using the word “accelerate,” and I think the at the at the computer science level, at the Computing level, of course, one of the things that you and I have enjoyed and observed the benefit of—Moore’s Law—Computing getting faster and faster all the time, um Computing getting faster all the time at the same price, at the same power, uh translates to Computing become more affordable. Yeah, and so we enjoy that for the first, you know, 30 years of our career. Then for the for the last 15 years of our career, uh we really saw plateauing. So so that Moore’s Law has has largely ended. MH. And the contribution we made to the computer industry, the First Fundamental contribution, is to accelerate to the extreme using a completely different architecture and refactoring of algorithms and and working together at the computer science level so that we take sequential processing and turn it into accelerated parallel processing. We’re now accelerating Computing to the extremes. Yeah, that allows us to bring cost efficiencies back to the industry, Energy Efficiency back to the industry, and of course acceleration of workloads. Yeah. On top of that Foundation, we invented or we made possible machine learning and artificial intelligence. We we now compute so energy efficiently, we now compute so incredibly efficiently that you can go to the extreme version of computing and let the computer figure out the insights. Yeah, so at the at the at the ground level, one of the things that we have to do is and and the the great work that we do together in this new product line that you have—the thing about Infinia is using accelerated Computing and using artificial intelligence to uh learn from all of the the data that you have, uh transform the raw data into Data Intelligence, yeah, embedding the intelligence into the model, extracting uh from all of your data the semantic of it, the intelligence of it, the information of it. Now, instead of serving raw data, you serve metadata, the knowledge, the the intelligence, the insights. The the incredible thing about that is metadata, the semantics of that, the semantic layer of data is in extremely compressed.

Yes. Yes. Yes.

The compression ratio is incredible, and so that’s why you can move it all over the world. And yeah, and in order for it to take place, so so multimodal I think is a necessity in order for Enterprises to really benefit from AI. And so with multimodal metadata tagging and the ability to move the metadata part of it becomes essential; that’s otherwise the economics simply don’t work out. There isn’t enough data center space in the world; there isn’t enough power in the world to get it done like that. So the metadata attributes are very key, and when we started out with Infinia, that was one of the things we focused on. We said, “It has to be a metadata-rich infrastructure that can handle at very low latency these objects that will have to be transformed to gain Insight.” Yeah, because at the end of the day, uh as you say, it is Data Intelligence. What is the point of bringing all this data into the environment to train the models if you don’t gain Insight from it, if you don’t gain business value, if you don’t get Leisure value for consumers?

That’s right.

So it has to work out; it has to result in benefits. And it’s a new way of interacting with your company’s data. You know, instead of retrieving data, you figure out what’s in it; you may be modified and stored back. You’re in a lot of ways talking to your company comp’s data. You have question. Yeah, you have questions for your company’s data; your company’s data speaks back to you, tells you what you need to know, uh you might have a fair amount of insight that’s uh distributed in your company’s raw data that is now in this uh semantic form, and uh you would like to have agents, AIs, yeah, that interact with data from different departments, query that data, generate reports that another AI agent then reads, understands, uh combines it with other data and other intelligence and produces ultimately some, you know, Insight. So basically, I mean, it it is domain experts; these agents, each one is a domain expert in a particular field, and they’re the advisors which now organizations can leverage in order to gain competitiveness, develop better products faster, bring more value to their customers, whether it’s products or services. And and you take that and you combine it with the Omniverse, which I think is the most phenomenal thing out there. Instead of doing things in in the physical world, you say, “Well, I’m going to develop a drug to solve some disease, but it’s very very costly; it’s billions of dollars; it’s years and years, and there’s FDA approval, and I don’t know if it’s going to happen or not.” And we have 10 different avenues that we need to explore; we cannot do them uh in sequence twins; we cannot do them in parallel. So we’re going to spin them up as digital twins in the Omniverse, and then the response will come back saying, “Okay, if you combine these attributes from pass number one with these other attributes from pass number four, that will maximize your likelihood of success, will compress the time to take that drug to Market, and will maximize the benefits Associated.” Now, for the audience is watching this, uh this is really interesting. You know, our journey started out in high-performance computing and high-performance Computing training Foundation models and Frontier models; that is the extreme version of high-performance Computing. And so so Nvidia and DDN um found our friendship and our long-term partnership through that origin. Now what we’d like to do as we train these models, these models can now uh turn the world’s Enterprises, all of our data from raw data into Data Intelligence. Yeah, and so now our world is moving into Enterprise together. Yeah, and so our arc from high-performance Computing to Enterprise, um from Cloud, public Cloud to now private cloud and public cloud, and then now when you’re in all of these companies, ultimately what they would like to do is uh live in the digital world, because when we’re living in the digital world, everything we do is faster, and this idea of Omniverse is to enable every company to be a digital twin. It is so profound; it’s so incredible, and I’m not really sure people grasp the importance of in and the and this is this is our journey which is the next click. So now we’ve gone from from uh uh supercomputing to Enterprise; now we’re going to go from Enterprise to the digital twins of Enterprise. Yes, and and in order to go from the digital twin the the Enterprise to the digital twin of the enterprise, we have to take the data of their domain, whether it’s uh the data of their domain could be 3D; it could be uh proteins; it could be chemicals; it could be information; it could be time series, you know, and it could be physical information. We want to take all of this data, extract the representation from it, the meaning from it, and once we could have the representation of it, we represent it in digital Omniverse, then all of these compan—the world’s companies—could have their representation in digital. Yeah, and once we are in digital, we can try a thousand experiments at the same time. So once we have Omniverse, we’ll have, you know, Multiverse of omniverses, and we could try all these experiments. S really, it’s living A Million Lives simultaneously, and you pick the one which is a combination of these various lives that will be optimal for you. And that applies to Enterprises and organizations; it applies to governments with Sovereign AI for the benefits of their constituencies; it applies to individuals, consumers, health. So you just spin all of this out and you optimize.

I love that you see that. That’s exactly right. But that is very profound. I mean, the first time I heard you speak of the Omniverse, I’m like, “Wow, this this is what AI has led us to; this is the explosion.”

That’s right. It takes AI to take the world’s raw data, yeah, extract its un its meaning, to represent it in a digital world, in a digital twin, because we have to compress time, and we have to lower the economic impact of it, so we have to do it much faster, much cheaper, and the only way to do that is with digital twins; there’s just no other way to do it.

Yeah, we’re going to digital twin.

So so that is the acceleration of uh the adoption of AI by Enterprises; it is helping them on this journey so that they can really compress time and and and gain significant benefits. That is the moment we’re in right now. Now we’ve uh uh taken the the data layer, yeah, and we’ve now built on top of it an intelligence layer.

Mhm.

And this is Data Intelligence. On top of that layer, we’re going to build an agentic layer.

Yeah.

Yeah. In the information world, we call them agentic AI. Yeah, in the physical world, we use physical AI that is embodied in robots.

Yes. Right.

And so so we have we now have another layer on top of it. What’s really exciting and and you’re probably you probably saw um what happened with DeepSeek MH uh the the world’s first uh reasoning model that’s open-sourced, and it is so incredibly exciting uh the the energy around the world as a result of R1 becoming open-sourced—incredible.

Why do people think this could be a bad thing? I think it’s a wonderful thing.

Well, first of all, I think from an investor, from an investor perspective, there was a there was a mental model that um the world was pre-training, yeah, and then inference, yeah, and inference was you ask an AI question and it instantly gives you an answer, one-shot answer. I don’t know whose fault it is, but obviously that Paradigm is wrong. The Paradigm is is um pre-training because we want to have Foundation; you need to have a basic level of foundational understanding of information in order to do the second part, which is post-training. So pre-training is going to continue to be rigorous, lots of data, multimodal data as you were mentioning; we’re going to learn from of course language, but language and images, video, sound, and we’re going to combine all of that together into our foundational knowledge, and that that’s going to keep on going. The second part of it, and this is the most important part actually of intelligence, is we call post-training, yeah, but this is where you learn to solve problems. Yeah, you have foundational information; you understand how vocabulary works and syntax work and grammar works, and you understand how basic mathematics work, and so you take this Foundation knowledge, you know how to apply it to solve problems, and so we call it uh you here it’s post-training; you could use reinforcement learning, human feedback; that’s another way of saying uh using human demonstration, or you could use reinforcement learning, which is self-practice, you know, or you could um uh use reinforcement learning uh with another AI, a coach. Right. And so so there’s a whole bunch of different learning paradigms that are associated with post-training, and in this paradigm, the technology has evolved tremendously in the last five years, and Computing needs is intensive, and so people thought that, “Oh my gosh, pre-training is a lot less…” They forgot that post-training is really quite intense. Yeah, and then now the third scaling law is the more reasoning that you do, the more thinking that you do before you answer a question, the better. Of course, there are many things that we’ve memorized. You know, how do you what’s a square root of 64? I mean, you just kind of memorize that; you could reason about it, m but you don’t have to, you know. And so ideally you’ve memorized a lot of basic things, but they’re most of the valuable intelligence you still have to reason through, and so you have to apply first principles; you have to break it down step by step; maybe you have to um try a whole lot of different experiments, and depending on the output of the results of one experiment, it might inform the, you know, the next experiment you do. And so reasoning um is a fairly compute-intensive part of, and so I think the market uh responded to R1 as in, “Oh my gosh, AI is finished,” mhm you know, it dropped out of the sky; we don’t need to do any Computing anymore. It’s exactly the opposite; it’s complete opposite because what DeepSeek is doing is is making everybody take notice that okay, there are opportunities to have the models be far more efficient than what we thought was possible, and so it’s expanding and it’s accelerating the adoption of AI. And so now we have R1, yes; we have Infinia Data Intelligence layer.

Yeah.

Yeah. The R1s can now be able to talk to intelligent information to solve problems. I mean, the other thing I think is extremely enabling is the CUDA ecosystem which you fostered and nurtured and and and help really people embark on. Now with CUDA obj, I think it is opening all kinds of possibilities because people cannot tie into this and apply it—combination of CUDA obj, NeMo, you know, the inference part of it for specific Industries—Life Sciences, Financial Services, autonomous driving and so on and so forth. You take all these things; you tie them together with the advances that will be made in the models, and again it’s an acceleration of the adoption of AI, which is a wonderful thing for NVIDIA; I think it’s a wonderful thing for DDN, but it’s all happening at the application. Yes. I think we need to continue to improve things as you’re doing with GPUs—faster and faster, lower power consumption, just driving the economics down in the data center and and in multi-clouds. But it’s the combination of all these things, you know, this Data Intelligence layer as as you suggested, which is tying into efficiencies in the data center and acceleration at the application layer. What’s really amazing is when you boil everything away, all of this is just software.

Yeah.

Yeah. But the representation of software today and the way we talk about software and the way we talked about software years ago—so profoundly different—incredible. One of the questions, Alex, that people have—should companies um build AIs or should they just use an AI That’s in the public Cloud? Yeah, and and I think that the answer is yes and yes. And the reason for that—if you could use a public Cloud AI, I would absolutely start yeah there. And so whatever whatever is available there for a lot of personal use, the general intelligence is very good and and growing and getting better at exponential rates. And so really really great to start with. However, there are many different domains inside a company. For example, the way we design chips, the way we write software, our CUDA programming, um our Verilog programming, the way we do Supply Chain management, uh it’s deeply domain specialized, and our expertise there obviously exceptional. Yes. And so so in that particular case, we tend to build our own AIs. We have all the necessary toolkits um from open-source models and uh open-source toolkits like Nvidia Nemo, um uh Nvidia Nims, uh and uh uh Data Intelligence layers uh like Infinia, um all of those uh pieces of Technology are available for you to uh onboard and curate uh the training curriculum and training your own AI. And so it will become the curation of all of these AI uh agents. Companies will consist of homegrown AIs, MH contractor AIs, AIs that come with third-party software platforms, um AIs that come with public clouds. And inside your company, you’re going to be an AI of AIs, and uh inside the AIs um a system of models, yeah, and you’re going to have AIs working with each other, solving large problems, yeah, sitting on top of your company’s Data Intelligence layer. And so so so so for organizations, I think it’s really okay as an organization, as an Enterprise, irrespective of the market you’re in, what is your mission? What is your purpose? Once you’ve defined that, then there’s you have to differentiate, and the differentiation can only come from specialized application of AI to your particular organization. So yes, you can consume AI, but that is just the consumption of it; the real value is, okay, your purpose in life as a life science organization or as Financial Services or organization is what? And so how do you differentiate? And that’s very special; that’s where you need the specialized names capabilities that Nvidia is delivering; that’s where Nemo comes into play; that’s where I think DDN’s Infinia comes into play. It is really enabling that differentiation to be more impactful, and it’s amplifying that differentiation.

Yeah. That’s exactly right. Uh again, I would like to thank you for being such a wonderful partner, showing us the way.

Thank you.

Uh embracing our technology internally, NVidia, which is wonderful. Can’t thank you enough for that. And and the NVidia is powered by DDN. You know, it is it is.

Yeah.

It’s great partnering with you, and you guys you guys have been you built amazing technology.

Appreciate it. Really amazing technology. And and without DDN and Nvidia, supercomputers would be possible. And so so I really appreciate that, and um uh and now we’re uh we’re taking our partnership into a new new frontier, into the world of Enterprise and Omniverse in the Enterprise with Infinia sitting on top of Infinia. It’s really great.

Thank you. Thank you. Thank you very much.

Yeah. Thank you.