NVIDIA’s New AI Physics Models and the Rise of AI Ministers | TSG Ep. 970

Hey everyone. We're in Brooklyn. You're watching Techstar Tank.

Hi everyone. Welcome to this edition of Textron Gang. As I mentioned, Mike and I are in Brooklyn, New York today for a kinda unique, uh, events.

The first of this kind I've gone to, it's called AI Native Defcon, and it's in a place called Industrial City in Brooklyn. And I gotta tell you, I love the vibe already, though it's early in the morning. Um, I think, I think there's a future for this.

I think this thing's got less. We'll give you more of a report, probably not today show maybe tomorrow or on air. But in the meantime, Textron gang stops for no one.

And we'll keep going today. Uh, let me introduce you to our gang. We have the men up north.

Chris Blas, a new gang member, and we'll ask them to introduce himself here in just a second. Sid next Sidd, welcome to the Gang Sidd. If you wouldn't mind, take, take 15, 30 seconds, give people a sense of, of who you are.

Sure. Thank you. Uh, appreciate the invite and glad to be here.

Uh, so I'm currently the president and CEO of, uh, of Techon. Teon is a, is an advisory and analyst firm advising clients in the area of cloud and AI infrastructure and AI software and trends. Uh, prior to me starting off my own, on my own, I was a vice president of research at, at Gardner, where I tracked the same, uh, coverage space.

And, uh, prior to that I've spent my career as a practitioner at an executive positions at Cisco Dell, EET, bell Labs, and did my own startup company for for six and a half years, which I took to an exit. So my, my experience spans sort of practitioner research as well as being an analyst in the industry in those companies. Thank you and welcome to the game.

And then last, Lisa, I, uh, and I always try to get her name right 'cause I keep trying Chaya Chaya. Yes, Andy, I've been on the show for five times now. I know.

And I messed up your name six times, but Aki char till I get it right. Anyway, thank you. Good to be.

So, Mike, we're gonna lead off today. You know, this is a big week. Nvidia has earnings on Wednesday.

So the market, and, you know, the market is in a bit of a, of a, a, of a share, not a shambles, but it's retracting. And people are saying Nvidia might be a big reason why, but they're, they're keeping, they're trying to keep pushy envelopes. They're pushing the boundaries of visits.

Yeah. There's a super computing conference in, uh, St. Louis this week and NVIDIA's there talking about these new models that they're creating that are all wrapped around physics and letting people do some very, uh, intense research at the atomic level, right?

And they're talking about simulations or fusion reactors and helping people build new materials. And as I kind of look at it, I mean, and I forgot one other thing. They also are creating interconnects between supercomputers based on GPUs and what are gonna be quantum computers.

And they're saying that these supercomputers that they have will essentially be the control plane for the quantum computing. And we'll see how that plays out. But it feels like, you know, rather than just kind of focusing in on the GPUs, I can't help but wonder if I look at this and say, Chi let's start with you.

Are, are we on the cusp of some sort of new golden age and science and research? 'cause it seems like all these projects that people are talking about are gonna lead to all kinds of innovations. I think I'm aligned in terms of their next step.

They do have the domain knowledge of the hardware. They do have the people who have worked on it. So it looks like they're taking the next step in the direction, coming up with their own model.

And I think one of the ideology that will help here is machine learning for machines. So think about it that the GPUs, that when built by them, if something goes wrong, the model understands where they can perform better and what are the bottlenecks and give more insight about it. So I think the data that they have collected over the years and building the infrastructure for it, they're going to use the same data set to train these models to come up with the right logistics, performance metrics and benchmark that they claim about it, which people have not even scaled yet because everybody's buying it, but nobody has scaled the GPUs to up to that extent where they say, yes, it actually meets this many number of tokens that I wanted it.

So yes, I am very optimistic about it, um, in terms of model, but I, I also feel there are too many cooks in the kitchen now. So that's kind of where I lead others to kind of share my feedback on, I haven't tried the model myself, but that's what going to be in the next thing. But, uh, if it's coming from Nvidia, it seems to be that they have the, I would've thought about it and trained on the, in initial, uh, kind of trains for it.

Yeah. Sy, do you want in here? What's your thoughts?

Yeah, so Teon have been following, uh, in NVIDIA's progress over time, obviously. But, uh, uh, as far as Supercomputing is concerned, clearly, you know, the focus on quantum was something that, uh, was striking to me. Uh, but if I were to sort of summarize the takeaways, uh, you know, it's very clear that, you know, infrastructure is obviously advancing really fast, right?

The scale and ambition of systems being deployed with varied technologies, be it, you know, your traditional networking technologies with, with their spectrum announcement they made last at at in their own Nvidia, uh, event in DC a few weeks ago. And now with, uh, with Quantum, you know, things are really coming together and integration of quantum is real, right? That's the other takeaway I would highlight where you think like NVQ link shows quantum is entering infrastructure planning, so it's no longer a sort of research project in the labs of IBM and other places, right?

Um, so it's becoming more and more real. And then, uh, I heard one terminology that jumped at me that NVA talk Nvidia talked about, which is AI for science, right? So things like life sciences, climate materials, manufacturing, uh, you know, focus on those kinds of things.

So the high level narrative that jumped at me, I think we ought to track that as to what they're gonna be doing there. And then, you know, in, in terms of risks and governance, they talked about, you know, the whole quantum integration, quantum safe, the whole supply chain issues associated with quantum, the energy and thermal footprint, you know, the whole sustainability narrative is highly irrelevant as well. So those things jumped out at me.

So I think going forward, we wanna make sure you keep an eye on how NVQ link develop or deploys and develops over time, right? Tracks or the commercial aspects of these technologies, because oftentimes we hear companies like Nvidia and others talk about what's today on the truck, what they're gonna be doing in the future. But it's important to understand, you know, in the, what the future's going to hold for them to be making revenue out of their futures, right?

And then evaluate the software stack maturity. They've already been in the market, withing lada, but with quantum SDKs, you know, what does it mean for the developer community and how do they build this thing through their whole ecosystem partner play? So those are the things I would highlight.

You know, I I, protein folding sticks with me as I look at this era of ai, and it was a couple years ago, right? You know, we basically solved the protein folding thing completely. And if you're just a general science nerd like me, this is one of those things where the possibility space is so enormous that in the past we've forecast, you know, taking the computers we have now and forecasting 'em forward, it's still 10 trillion years until we could possibly do it that way.

And we're done. You know, protein folding is done. Amazing, right?

And as I've learned in the last year, more about specifically how these systems work, you know, LLMs and the, and everything we're talking about right now, you would think that if you didn't know much about it, if you think about the old way, the digital way, we think that maybe it's vector math with GPUs or something is making it faster, and it's not, it's semantics, right? You know, because what those researchers basically did is they just asked the AI could do these things, right? And what does that even mean?

And we were stuck in this debate space around this, but it's not that complicated, right? You're, you're, uh, uh, you're reducing the possibility space, you know, like a human. If you were talking to them and said, Hey, could you research this?

You know, they would say from this point, the likely things are this direction. And not try to, you know, at, at every step, do every digital fill, leverage digital pothole, and the savings in time and energy are hey, transformational. That's why this is, you know, such a big deal.

That's why, you know, to, you know, Helen and, and Sid, you know, uh, analysts like you, the business investments where people are putting the money right now, ah, I don't feel that comfortable about it, but the overall trend, yes. Because if you could speed up these things like protein folding, like physics research, not just by sheer CPU horsepower, but by semantically narrowing the space and sheer horsepower. Holy cow.

Exactly. I think, I think that's spot on. You know, I read an article in the Wall Street Journal on Friday about this guy Jan Koon, who's a, a chief scientist at Meta, and he was a contemp labs, and he's talking about how the new world models not large language models, because LLMs are not grounded in physics.

So to your point, Chris, you know, pro about the protein fold, the analogy, right? Uh, world models are really going to be the next big thing because LLMs don't know basic these two sequence transaction, right? If A then B, because they have a pattern of text and they do NLP and, and come up with the logical answer to a prompt, but there are no physics, they don't know things that, things like balls roll downhill or liquids spill, you know, unless such pattern actually appear in text, I think you're spot on in, in, in the knowledge That that's very true.

Chris, let me, let me get, so my, my take when I hear these things though is, you know, I, I'm gonna borrow a nursery right from the old lady who lives in the shoe. There was a company called Nvidia who invented the GPU. They had so much money they didn't know what to do.

Okay? When you're a $5 trillion company, you can't, and you wanna keep growing that market cap and in pleasing the street, you can't be doing incremental growth. You gotta make some holy, you know, uh, deep passes, right?

Like Google did. Remember when Google was untouchable, it, it controlled 99% of the, of the, of the search market. They were, they were printing money and they, they, they set up a bunch of Hail Mary's, right?

And some of them, like Google fiber, uh, you know, really civilization changing big things because you need that kind of stuff to get from 5 trillion to 10 trillion. It's not gonna be about selling a couple more GPUs. You've gotta invent game changing new markets.

And so kudos to them, not kuda, that's something else. But kudos to them for doing this. But that's the game here.

They need to find where's my next $5 trillion market gonna be? And, and so that's what this is about. It might be protein folding, it might be the, as Mike say, the control plane for quantum, right?

And this way they ride that quantum because, you know, hey, Google's live, they have the willow as the quantum chip or any one of these players are making, you know, that are working on quantum chips and qubits and so forth. But Nvidia is not gonna abandon that market. So they're, they're making a, a, a, you know, a bit of a Hail Mary there and a hail Mary here.

They don't need all of these to hit. They need one or two of 'em. And it's almost like the VC gig.

Yeah. I think Nvidia jump in Nvidia is at an inflection point right now, because what you just said about their lineage, I mean, how do Nvidia get into the AI business? It was an accidental discovery, right?

GPUs are basically designed with the purpose of accelerating computer graphics and image process and render pixels on a computer screen. Now that, that function requires the ability to multiply very large matrices, right? Turns out that Transformers, which is based on NLP and sequence transduction, also requires the ability to multiply large matrices.

So somebody came up with that idea, Hey, why don't you use GPUs for processing AI workloads? And, and then, and then we know, we know what happened thereafter. But I think as a company, Nvidia really ought to think about what is the, to your point, what is the next big thing, right?

Is it protein folding? Is it something else? You know, is it gonna be learning processing units?

So lpu versus GPUs, where these things are built for inferencing in custom environments for small language models. I think that's the, that's the challenge they are currently faced with. And it's almost like changing the engines of a 7 47 in flight, right?

And many large companies go through this inflection point 'cause they're crossing the cat classic crossing Cham, right? So it remains to be seen how they sort of do both things at the same time and make money out of that, that, that transition, right? But Sid, let me ask you this question.

I do have one, like Nvidia is one of the biggest players in the chip, right? There's no competition around it. Why?

Why can't they continue to work what they're good at it and scale and reduce the cost of chips and GPUs instead of a market, which is already a multiplayer, Correct? You're right. So they're gonna milk that as they can Go ahead.

No, you go, you go. So they're gonna milk that as, as long as they can. Obviously I would do that because that's bringing me revenue, right?

But, but they have to think about brand new architecture that's different from the GPU, right? If they don't, then there are others who wanna come into this. There's already lot of development going on in pockets within the hyperscalers that are building their own, you know, specialized ships that they're not even, that are, that they're not even, uh, selling to the, in the open market, right?

I know that some of the hyperscalers already doing that. I have information under NDI can talk about it here. But that's what's going to cause a problem for them if they don't evolve as a company.

Go ahead, Mike. Sorry. Yeah, yeah.

No, if they, you know, they will be kings of the GPU and that might take 'em from 5 trillion to 6 trillion. Jensen Wong didn't get here thinking small. No.

So let's, let's play that out a little bit, right? In my mind, there's no rule that says that Nvidia has to continue to only make GPUs in theory, they could go make another processor that is optimized for AI models as opposed to the one that they developed that was a happy accident. And we've already seen them starting to build CPUs and they have dpu, and it seems to me they're a lot more ambitious than just looking at GPUs and They want quantum, right?

So, you know, I think that, you know, they're gonna be so far down the road on the model side that, you know, if they come out with some additional processors, 'cause they learned how to optimize those models, they might be in a better position than folks who are just pure play prosor companies. You might wanna look at Nvidia and think about it now as trying to own the entire stack. I mean, heck, they sell entire systems now, You know, they, there's going to be bigger markets.

I mean, a AI is huge and they, they've capitalized on it, but this is not the time to take your pedal off the, take your, you know, take your foot off the pet. This is the time where you now have resources to double down and, and, and make some big passes. Some Hail Mary kind of attempts to, you know, who knows what, what can come of it, right?

And they're already talking about the next two generation of GPUs beyond Blackwell. So, you know, and if they're on a cadence for delivering those, I think they said, uh, somewhere between a year and two years. I mean, the pace of this stuff is gonna just dramatically accelerate and maybe we'll all be living in a different world by the end of the decade because town's point some of this stuff, if it ever hits, we'll change the way we let.

And I think also another thing we need to think about is what are they gonna be doing in the area of sort of pervasive computing, right? Because it's one thing, building a chip and putting all the intelligence in the chip footprint of real estate. But if I need one unit of compute to process, uh, an AI workload, do I need a pervasive computing environment?

Which means I have GPU clusters or whatever, D-P-U-T-P-U cluster that are distributed in the distributed computing analogy. So should they be focusing on, uh, connectivity of these cluster, of the compute clusters, which gives that one unit of compute from wherever is the most efficient resource for that, right? So now suddenly networking is becoming sexy all over again, right?

In the world of ai. So like what Nvidia do in the world of networking beyond just computing, right? It is gonna be an interesting challenge for them.

I don't know what, I don't know how you guys feel about that. That's a good topic to talk about. You know, they say Nvidia made hardware sexy again, so now they'll make networking sex sexy again.

I hope so, because a lot of the systems they're currently selling look and smell like made friends to me. But wow, We're not outta time for this segment. Let's think quick, Frank, and we're going to come back and, and talk about, uh, our next segment.

More ai, probably you're watching text on game. You've earned it. The spotlight, the responsibility, the weight of teams, companies, and entire industries fall on your shoulders, lives depend on your decisions, your home life included, that work your protected physically and digitally.

Nothing gets through your team without a fight. But in a globally connected world, everyone sees you, including those who mean to cause you and your organization harm. And now home your sanctuary attackers see an opportunity.

Your digital front door is wide open. And what compromises your home can breach your boardroom. Because the devil's greatest trick isn't targeting your workplace firewall.

It's convincing you that your personal life isn't at risk. Black clerk, digital executive protection, defending the new attack surface your personal life. Hey folks, we're back.

And this is a little unusual. It may be scary and something right outta 1984, but Albania has a minister for ai, and I guess there's gonna be more and more government decisions are gonna be fed through the AI first, and maybe somebody will check on this, or maybe they'll just say, Hey, that looks good to me, and away we go. But Chris, I know you've been kind of monitoring this kind of thing and, and have talked about this in the past, but it looks like it's here.

Well, more countries essentially have ministers of AI and we'll just all, you know, wake up in the morning and be told what to do. I I wouldn't, uh, end it that way at all. But, uh, yeah.

So the last segment ended up being sort of a market ana, you know, business analytic and the architecture. Lemme talk about narrative infrastructure, right? You know, so Albana is not the first, you know, here in Canada, just this year we have our first, uh, minister of ai, ev uh, minister of Solomon, Evan Solomon.

But last year, mark Sean, the was the first deputy minister. And we know in cybersecurity, we've seen this happen before. Mark Weatherford, who's now at, uh, doing policy at Nvidia, was the first under secretary for cyber in the, in the beltway, in, in the US federal infrastructure.

And Albania is an interesting case because, you know, they, they punch above their weight in EU and in, uh, a number of ways in this, in this case. But we're going down this path, you know, these are important topics that need to be addressed at this level, at the policy level. Someone has to be responsible for it.

You know, Mike, if it ends up being authoritarianism, I guess then, you know, democracy has failed, but I don't think we're going there, right? And this, you know, Sid, you touched on something in the last segment that I really wanted to, to to pivot on because, uh, the Meta's AI chief, um, Yann Ko, you know, leaving and his statements about needing to move AI into a, not a, not just a word prediction, but a world map, right? Models exactly what, but I, and we focus on in this attest civic AI world, and it's not that far reach.

And it's interesting, uh, to see folks like, uh, like, uh, like, like Lako making those statements and those moves right now, right? And just just to map this out for folks, LLMs are a language modeling thing that's modeled after what happens in our brains. But that's not the only thing that happens there.

And mammals, 200 million years ago started building a world map that's literally the term used in cognitive science. So they could, it's like a, like an eight big video game, like a really, really simple one. You can navigate in 3D find the food and find the food you found yesterday, right?

The world map that Launa is talking about, that that folks like myself and, and, and our allies are working on these days isn't about a whole new different ai. It's about taking this capability we have and having it operate in a relational environment so it can basically have that world map. And in that it touches on the things that, like Mike and Alan and you guys, we've talked about week to week throughout this year, you can do so much more without so much energy, time costs, so on and so forth.

So that's just, you know, that, that move, you know, out of, out of meta AI this week and this happening in Albania, you know, modeling what's happening here in Canada, that maps a narrative arc that's fairly straightforward, I think, I think we'll see the business and technology architectures follow from that. Yeah, I think, I think the, and back to the world model conversation, I mean, the way I understand it is it's an AI system that essentially builds in, you know, an internal representation of an environment and then how it's used to ate future states, right? So it's more of a outcome-based predictive model rather than LLM, which is a pattern recognition based technology, right?

So, so I, I see that whole world model that Koon talks about, as, you know, the predicting model, predictive modeling of physical systems like infer, object permanence, motion collisions, the physics thing I talked about, you know, gravity, things of that nature. Uh, it also has a relevance to autonomous planning so it can run thousands of simulated rollouts, right? Uh, it has, it's more grounded in decision making rather than sort of like, you know, in the case of sequential transaction, you know, it's very, very one way, right?

So, I mean, I can say I, I know based on text collecting the LLM database, if you wanna call it that, that Tom Cruise with mothers Holly Cruise, but it can't tell me, Holly, the sun is done, right? Because it goes only in one direction sequential, right? So, so world models will kind of overcome those kinds of things.

And you know, it's also gonna have things like embodied intelligence, so robotics, autonomous driving, manufacturing, drones, gen software, all sort of require this, right? So I, I look at sort of LLM use for reasoning word models use for simulation and a chance and agent for action, right? That's what the triad of future of AI architecture, in my opinion.

Well, and, and the, the, the, the value of the world map for now, m is i, is is is it self transformative? And that's why I think this era is, you know, historically we may back, look back at this, the hype cycle wasn't big enough, but it was the wrong hype cycle up to this point. But I think we're starting to get it and, and we build LLMs modeling ourselves.

Like that's literally what we're trying to do, extrapolate a little bit farther back, why are mammals, how do they do it? And we're talking little mousey creatures who got a, uh, evolutionary advantage by having that really simple world model. I think that's what we can do today, right?

It, uh, and, you know, attestation channels and the systems we build and demonstrate all this all the time, it, again, it's just happening this year. So it's not like we need to create another LLM industry, another multi-trillion dollar thing. We need to take these tools and work them in a relational environment that allows, you know, the result to act as if there is a world map like mantels do it, you know, nothing like we do at this point ourselves, but that's the path we're on.

It looks like, uh, not the Albania, like what, just one point on the Albania thing. Looks like they came up with something called dla. Are you guys familiar with that?

It's sort of a special, I don't know if it's LLM that's developed by the National Agency of Albania in cooperation with open AI at, uh, just I aspect, sorry, go ahead. Chai, you point. Uh, no, I was just saying that going back to the original point of we are having a minister, which is a ai, um, power, I've been a citizen of that country, I'll be excited, I'll be excited to see the politicians are getting a revamp or a new perspective to make the decision, and it might take the country the direction they want it to.

So I think as I'm, I'm not talking about the technical aspect of it, but I'm just saying as, as the humans who are taking this decision into consideration, is a good move. Um, politically, I think it'll, uh, give the governments the insight in terms of the directions where they should invest in take, take the insight, take the inputs, but can they rely on it completely? I think it's too soon to do that.

Yeah. I think, I think also, you know, just because we have an AI that spits something out, well, when we determine that whatever that truth is, that it's inconvenient, we're still gonna ignore it. Yeah, I, yeah, I, I think a lot of these countries have essentially, like Albania have long faced systemic issues with corruption, you know, nepotism, weak public procurement practices.

So, so the more they do in, in, in, in terms of openness and creating these AI initiatives that span obviously within their, their own country, but overall across integration with the eu, let's say, or the European Commission, right? Uh, I think those things kind of give, give an optics that they're more open to sort of collaboration and interoperability the rest of the world and creating these AI initiatives is probably one way to address that. Uh, so I look at, I don't, maybe, But I'm a huge fan of the, of, of capitalism and democracy and freedom of speech and so forth for the reasons of evolutionary pressure, right?

Because I honestly believe that the better systems are better people and ethical and, and so forth. And again, if I'm wrong, let's find out, we'll play it out in the markets. But, you know, to Mike, to your, you know, and I, and to be clear, you know, we, we, we, uh, uh, uh, spar on this one, but that's good.

You know, I need a razor to go against 'cause I'm an optimist, but it's just, if I, I right now looking around the world, I'm seeing countries like Albania, like Canada, where I have good faith. It to be reasonable effort to have an AI minister and do it properly. I expect a number of countries will do it wrong.

I think it will cost them, it'll cost 'em economically, it'll cost 'em in global power. It'll cost 'em in trade and relationships. And while I could be wrong, I think this is the kind of role like cybersecurity.

I was glad to see that get into the public sector because it's a technically intrinsically kind of more honest than just I said so kind of thing. Those of us in this, in cybersecurity know that that's not entirely true all the time. But I think this topic does drive it down further and further if your actual, you know, let me try to end on this, right?

We talk a lot about canon, right? Canon in our, in our terminology means what you actually do, what you say you do, right? And AI tools are actually really, really good at this right now, right?

And lots of use cases and, and we're seeing in real world. But I think the countries, governments, as they semantically analyze their actual canon, they'll find out whether or not they're doing what they said they do and opportunities to be more or less corrupt. Pick your, pick your choice.

I think you're looking at it wrong in all honesty here. Here's the deal. I, I've said it before, I'll say it again.

We're at the beginning of the beginning of the AI story. We're still at the beginning of the beginning. We don't know where this is gonna go.

Will it turned out as, as that chief scientist in the Wall Street Journal article said that LLMs are sort of a dead end and we need to go to these world models to really make this work better or to recognize our dreams. I don't know, but it's gonna be big enough where we need a government element. You probably need a world government element to it as any one country alone.

But it's not just to regulate ai, right? North Korea will regulate ai. The old Albania will regulate ai.

You know, the people who sit clutching their pearls saying, oh my, what could AI do bad? They're dinosaurs. They're already road guilt.

Forget about it. AI is going to go as fast as people can make it go. No one's gonna stop it.

No government, no nothing. It's going go, what needs to happen here is governments are gonna recognize how can we use ai? How can we harness ai?

Not how we can control it, right? So it's not just a regulatory function. And, and quite frankly, and I am no fan of Donald Trump, anyone who knows me or has heard this show knows I'm not.

But what he did early on in that administration is he appointed a so-called czar for ai. I haven't heard much from him since, but I assume he is working behind the scenes on all these deals that they work. We need a government policy that probably says, Hey, AI is going to be world changing, civilization changing, perhaps we need to be in on the action.

We need to understand what's going on. We need to try to influence it the best we can, given our resources. The resources of the US are very different than the resources of Albania.

And they're very different than the resources of Canada. But government resources will be brought to back here. And, and rightfully so.

It has to, if it, if it's as big as we think it's going to be, it naive to think the government's not gonna be involved in, I think it's that somebody may try to control it. 'cause they will, because for humans, Well, it will it be the government or some strong man or who that, you know, who knows? I mean, I like to, I, I like to see the government.

I like to see the government, uh, Be providing some guidance on responsible ai, right? I think that's key, right? Because we can't have all these private companies go burst.

So again, I'm, I'm a, I'm a capitalist, don't get me wrong. We, but we can't have chaos, right? And there's too many competing positions, ones being taken by Elon Musk and Xai and Grok and that stuff.

The ones being taken to open ai. And you've got a whole host of other positions, right? Meta and so on and so forth.

Like, so we have to be somewhat cognizant of how that is managed, for lack of a better term, uh, from a responsibility at respective. But I don't wanna see government interfering in ai, right? So I don't want government come stepping in and say, if you sell this chip to China, gimme 15%.

I mean, to me, that's socialism, right? Like, how is it a different? Absolutely.

Either you're regulating, you don't want China to have it, or you do, but you don't, it's not Hand. You criticize socialism. On the other hand, you participate to socialism.

Agree. That's a problem. We, we all, and then what the other thing is, I like to see government actually deploy AI with their own agencies.

I mean, if I have a government minister, that person should be responsible making sure government is using ai, right? I'm not seeing any of that happening, right? Well, Elon must supposedly, but guys, we gotta take a break here.

We're overtime. We've gotta come back for our third segment today. It's a lively discussion.

You're watching Techstrong Discover Textron Group, the epicenter of tech innovation. We are your go-to for reaching IT leaders and practitioners worldwide. Our secret impactful content that sparks awareness, engagement, and top quality leads with us.

You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more. Join our satisfied clients. Let's revolutionize your tech journey.

Contact us today and tell your story to the world in the most powerful way with Textron Group. Hey folks, we're back and we're gonna have one more little chat that's AI related at least. But we have talked about one of the biggest issues with AI is energy.

We don't have enough of it. The grids kind of, well, shall we say, shoddy. And so now a lot of folks are saying, well, we're gonna have these massive batteries that are gonna enable us to store energy so that we can mitigate some of the impacts these data centers are having on our grids.

And it's an interesting theory. I just don't know if the science behind the batteries is up to the task or not. But Alan, what do you make of it?

Can batteries kind of solve this issue for us? It it, it could help if you believe we can can get there, you know, on yesterday's gang, we spoke about JP Morgan Chase put out a report that now they expect 5 trillion, 5 trillion. I'm back to that 5 trillion number.

I love that number $5 trillion of data center build out over the next couple years, right? What's the biggest bottleneck to bringing these data centers online and operating them energy? We don't have the energy to operate those data centers.

We can't build nuclear and get it approved fast enough. We, what? Are we gonna burn coal for this?

Uh, the government is not a big fan of solar and wind, and that's probably our best bet. So much. And this is again, the beauty of AI and, and the, and the first segment and video making some hail Mary, that's the beauty of AI is much like the NASA space program to the moon in the sixties, right?

They knew to get to the moon, they had to invent some things that didn't exist right now, right? They had to do better computations, they had to make materials better. They, they needed mathematics.

And you know, they, a lot of technologies got invented and commercialized as a result of the mood program. We are in the same boat here. We know the key to making our, you know, to solving the energy issue for data centers and, and for beyond data centers is to have that a battery storage, right?

We could then really store all that solar and wind that gets generated and, and stored for a rainy day, so to speak. So I think this is gonna spark better battery technology. And I think we've already starting to see these kinds of announcements, right?

And that's what this is about. We need better batteries to make our data centers and our world better. You got trillions of dollars at stake here.

Let's go build better batteries. I just wonder if these batteries are gonna be kinda, you know, a substantial impact or is it gonna be like, you know, when I get that electric vehicle and if I pump the brakes, the battery will recharge. But it never really does.

I don't think that's, I don't think that's what we're talking about. I, I think we're talking about, I think we are not talking about the recycle waste this battery is going to generate over time, right? Because if you see Tesla is the one who entered this space years ago, and now this is the time where they're changing the batteries because their battery's supposed to last seven years.

So where is this battery waste going to we get disposed of? So it's not just about data centers getting electricity. I mean, you just said Alan few minutes ago that we are still at the beginning of the beginning.

So when, when we are not sure with AI where we are heading to, and we are building data centers and we don't, we do not have a clear concrete plan in terms of what are we going to do with this batteries when they're run out of life. We are still in that circle where we are building things. We don't know what use case they're in.

And we're probably producing more environmental hazards, which will be more dangerous than the benefits of ai. The last I check this, government wasn't very concerned about environmental hazards. Wait, wait, you now have some, all those rockets that Elon Musk and Jar Bezos are building, Right?

Send them up in space. I i, I think it's important to recognize what is the role of the battery. The battery's not gonna be the primary supplier of electricity gives data centers, right?

It is gonna be that power buffer, right? And in the grid and the racks. And the reality is that GPUs have very high peak average power issue.

So, uh, the, it, it's a very spiky and sort of a denser, less predictable power profile, right? So AI dense racks, I think the last time I read was like, consume 40 to 60 kilowatts today and then going up to 120 kilowatts, right? So the question then becomes like, what is that?

What is the role of the battery that's going to be, uh, going to be relevant or this buffer power buffer as they call it, right? And already the hyperscaler, like Amazon, Google and Microsoft are deploying what they call battery energy storage systems, or best they call it, which not only generate the power, but also self capacity back to the grid during peak demand. So I think all of that is an interesting topic, but equally important that someone touched upon is the environmental aspect of these batteries.

Like, what are we going to do with all this battery waste, right? That gets, uh, generated over time by powering creating those power puffers for these ai AI data sectors, right? So I think those are some of the interesting things to, Well, I I I, I've really enjoyed the, the, the power, you know, when I look at these shows over the last year, right?

We keep coming back to this issue. And I, and this is one of these issues that comes up and we're mostly talking about ai. We talk about power, but I've been involved with the grid for decades and the last 15 years particularly has seen a lot of work in this direction.

And this use case is just for these sort of, you know, from a power engineering perspective, you know, one of these CEOs slap your forehead. Of course they, you know, of course we're gonna throw this in so into the specifics, kinda like the last segment. You know, I am on an international level.

I am looking for countries to disprove things that seem popular, you know, draconian, authoritarian, whatnot. This particular, uh, legal move in this particular jurisdiction I think is a good idea. Cattle just this week, you know, the largest battery produced on earth is coming out, has announced that the production level of their next generation of the lithium iron batteries, and that should be a, a big quantum step here in Canada.

Volkswagen is building a giga plant, you shovels in the ground, uh, battery plant, uh, that'll be producing, uh, platforms in 27. And, and, and, you know, take everything else we're talking about here over those timeframes, if you know, hyperscale or consuming energy data, uh, in bigger data centers with more Nvidia GPUs was gonna follow exactly the curve that it's on right now. That's physically impossible.

What are we going to do instead? Right? And, and whether it's these individual things with, with best, with battery storage systems, I, I saw a neat little thing that, again, I don't think it's the answer, but I love these sort of things in the uk they have a program now where you can run a micro data center in your shed, literally in your backyard, and it'll provide the heat for your house and lower your costs, You know, well Finland, Finland is building underground data centers and using the heat to, to heat how host old house.

So yeah, This winter, I have enough GP news running at home right now this winter. I'm, I I will in fact, you know, lower my heat bills. So anyways, you know, this is a complex issue, but there's pragmatic realities in, in power terms.

We talk about, you know, rotating mass, right? You, you don't think about this. If you plug something in, there's some spinning tons of steel that feels that momentum.

So we have battery systems now that can mimic that, which is bloody amazing. But we're still so early in this that again, grids and jurisdictions will make bad choices. I have a lot of popcorn.

I think we'll all learn from mistakes, but yes, this is where we're going. So maybe, maybe, maybe the future is where all these big corporations that are building these AI data centers will work with consumers and install GPU clusters in every home power with batteries, which can also power their refrigerator, generate enough heat and save money. So maybe that's new Steady kind of thing.

So, So Sid, are you saying that the energizer bunny's gonna save ai? Yeah, Exactly. You, Mike, you summarized it very well.

That's Why I love 'em so much, right? I mean, I don't wanna raise my kids under the house where GPUs are running. I don't know, what are the ultimate race gonna do to this American life?

Well, You put it in, you put it in a shed in the, in the, in the yard. You know something. I mean, still, still around.

Yeah. Yeah. I think on that note, we we're gonna end today's text drug gag gag.

Thanks for joining us. This was a great, great discussion on some great topics. Thank you for watching Micah.

And I'll be reporting here from, uh, AI Native Deron in Brooklyn today, tomorrow. So stay tuned for that. Stay tuned for the rest of the text on TV coming out here right after this.

But for now, half of the gang live in Brooklyn, we're out.