NVIDIA Reveals Rubin AI and Blackwell Ultra || Tech Field Day News Rundown: March 19, 2025

This week is Nvidia, GTC, and that's our closer look here on the gestalt It rundown, we are also looking at some other stories, including, uh, fricking lasers spun out of Google. Um, Intel's new CEO, uh, DevOps, empowered by Semaphore, uh, Microsoft's Quantum Chip, or is it, uh, solo iOS Cajun for agentic ai. And of course, the whole question of nuclear data centers.

Stay tuned as we dive in for this episode of the Tech Field Day rundown. Welcome to the Tech Field Day rundown, where each time we meet, we run down the IT News of the week with a variable degree of snarkiness. I'm your host, Steven Foskett, and joining me this week as my co-host since Tom is at Networking Field Day, is Mr.

Alistair Cook. Welcome to the show. Hey, Steven, it's great to be here, and I hope Tom is really enjoying Network Field Day, and I hope you are enjoying Certified Nurses Day.

Being a husband of a nurse and a father of a nurse, I think it's important that nurses are celebrated. Are you celebrating nurses today, Steven? Uh, I didn't know that I was supposed to be, but uh, sure nurses are good.

Um, I, I am instead celebrating International Clients' Day. I'm not sure what that means. Um, and, um, international read to me day, Al, can you read a book to me please?

When it's bedtime, I'll start reading you a, a gripping story, possibly a, uh, implementation guide for some sort of cloud technologies that will maybe actually keep you awake. Oh, I'm sorry. I'm sorry.

What did you say? Um, I was, I was hoping it started in a hole in the ground there lived a hobbit. That would've been a good story.

So It was a great story my father read to me when I was a child. Yeah. Um, so our listeners may have noticed that I said something different at the beginning.

We talked about this last week, but I think it's worth talking about again. Uh, this is the Tech Field Day rundown. So Gestalt, it is, uh, moving over to join, uh, the tech strong armada of websites.

Uh, that's something that we had hoped to do for quite a while, and we are very happy to have it achieved now. So, uh, gestalt, it is about to become Textron. It, it will become, uh, one of the flagship sites on the Textron network, which means that the rundown featuring me and Tom and Al is, uh, gonna be firmly rooted here on the tech field day side of things.

So going forward, we're gonna talk about this as the tech field day rundown. Your ears did not deceive you. It's still the same.

I mean, gosh, look at us. We got, ow we got Tom, we got me. It's gotta be the rundown, you know?

So just, just, just stick with us and, um, we'll try to bring you some snark. I get, I guess. Uh, let's, uh, get, uh, dive right in here.

Alphabet has spun off Tara, its laser-based internet backbone provider into an independent company focused on delivering high speed connectivity to underserved regions. I almost read that as undeserved regions, and that would be completely different. The newly formed company will continue developing its wireless optical communication technology, which aims to provide reliable internet with without the need for traditional fiber optic infrastructure.

Al um, fricking laser time, huh? Oh, yeah. But, uh, I don't think we're gonna be mounting these on any sharks.

These are lasers that'll be mounted on communication towers and a little bit of history. This came out of the alphabet project loon that you may recall, uh, canceled a little while ago. But this was the plan to have, um, balloons orbiting the earth and providing internet connectivity.

Uh, they will be interconnected between the balloons using this laser technology. Now, the balloons had some challenges, particularly some political and operational challenges, but the laser technology, the ability to maintain a consistent long range high data throughput link between two points, uh, through the atmosphere and the atmosphere is a pretty hostile place for, for sending lasers. So that, that technology for keeping that connection consistent is really awesome.

The kind of places that Tara now spun out of Alphabet, uh, is, is looking at deploying this as places where either the cost of running additional connections and additional wire is too, too high, uh, or even additional fiber if you're doing fiber to the home. So this will be a backhaul technology primarily provided to telcos, uh, but also for getting out into places where traditionally we might have used cell networks. So right through, uh, list developed countries where there was no copper wire infrastructure, wireless networks and cell networks were used for all of our, all of the internet connectivity, and they tended to have wireless backhaul as well.

Now with radio frequency, there's limited bandwidth, there's interference. If you have multiple sites using the same bands, this all doesn't apply where you have that directed energy link of, of the laser. So this laser technology is pretty cool for doing that backhaul without requiring additional, um, wireless bandwidth.

That's also interesting for getting backhaul into places where there's a lot of people. So one of the use cases that Taran talked about for this was providing additional bandwidth into event locations, maybe festivals, big sporting games where traditional cell networks get overloaded, and we need some way of getting those call connections from the onsite temporary cell sites back into the main network. And this is another of the places where not having to run cables and not having to, uh, trench through or manage wireless bandwidth is really beneficial.

So I think this is gonna be some pretty cool technology. The image that I saw of the actual transceiver that's currently in use looked to be about a four foot high. Um, I'm trying to think what, what that is in, uh, in the properly, uh, Americanized.

So it's about half of a washing machine in size to use US standard ways of describing sizes of things, or for those of us more metric, a a meter by a meter and a half kind of size. So not massive. And, uh, with some really smart technologies inside, I think these could be extremely useful for getting more back call.

Of course, there's a little bit of a challenge with doing back call through optical in that particularly, uh, obstructions might turn up, new buildings get built on the path your optical link follows. That's kind of a challenge for you. So it's not the cure for all problems, but it is going to cure a few problems around getting networking into places.

Intel's appointed, uh, lip Bhutan as its new CEO succeeding. Pat Gelsinger, who resigned, uh, in December, 2024. Tan was formerly the CEO at Cadence Design Systems and was also a board member at Intel.

Uh, he aims to revolutionize the company by streamlining operations and focusing on advanced chip manufacturing, including AI server chips to better compete with industry leaders like TSMC. Following the announcement, Intel's stocks surge nearly 8% reflecting the optimism about this strategic vision. Intel's had a bit of a rocky path for the last few years.

Steven, is this the way to get off that rocky path onto a smoother road? Well, it's a way to get off of the rocky path on a smoother road, and the Wall Street seems to think so. Um, for what it's worth, the the stock jumped quite a lot on the announcement of this last week, which by the way, happened just after we recorded the, the rundown last week.

Um, but that being said, I think that there's a lot of nuance here. Now, first off, I do have to disclaimer, I did talk about this on the, uh, Textron Gang on Friday, and, um, and there's a lot more insight there from the rest of the Textron crew, but if I can boil it down here, uh, the gist here is, number one, um, you know, Intel needed a new CEO. They needed someone who really understood the chip industry, someone who could really come in there as, um, you know, a bit of an engineer, but also a new, uh, a leader, uh, who could come in and, uh, direct things and, and somebody with vision, somebody who has a, a plan for what to do with this company going forward.

And Lipan, uh, basically checks all those boxes. Uh, he is someone who really understands the chip making industry. He, he led Cadence design systems, which was a, um, an EDA company.

They made the software that makes the chips basically, uh, think of it as a word processor for, uh, making computer chips. Uh, cadence, uh, is the undisputed leader. They're the Microsoft word of that industry.

And, um, they are basically an essential tool if you're gonna have a Foundry business. So if you want to have customers come in and manufacture chips, you really must have, uh, the support of the EDA tools. And so all of this is to say that, uh, this is someone who deeply understands this market, uh, but yet wasn't, you know, like an A MD or a Qualcomm exec, you know?

Uh, and, and so that's a, a pretty clever one. The other thing is that he did serve on Intel's board, as you mentioned, for a few years, which means that he understands, uh, the state of things at Intel. He understands.

He under, he knows the board members, he knows the, the executives, they know him. You know, all of this sounds great. Uh, what I'm hearing though, from insiders and from, uh, the analysts at the Futurum group and others who follow the company is that, um, there's a little more than meets the eye here.

Um, if we wanna read the smoke signals coming out of Intel, uh, it's important to understand lip Bhutan was on the, the group, uh, on the board. Apparently, uh, what I'm hearing is that apparently was, was, uh, the pro split part of the board. Essentially, they wanted to, uh, not have intel, uh, foundry and chip design and sales businesses be the same company necessarily that came to a head in August.

And, uh, pat Gelsinger seems to have won that argument. Uh, lip Bhutan, uh, left the board at that time, but of course, Gelsinger, uh, was forced out after. And, um, here he is back.

So, you know, if we're looking at what that means for Intel, it probably means that a split of the foundry and chip design business is much more likely to happen than we would've thought. And what might that split look like? Again, if we listen to the analysts and what they're hearing, uh, from their sources and, um, public information, um, what we're hearing is that it seems that it would probably be a consortium of, uh, money that would come in and maybe set up a joint venture or something like that, that would take the foundry business.

That's the Intel chip making business, including the brand new, um, you know, halo factory in, um, in Columbus, Ohio. And, um, split that off as a semi-independent company, uh, semiconductor, semi-independent, I guess you have to forgive my pun there, um, that Intel would still be involved in, but would be perhaps a minority shareholder alongside a group of customers, maybe to include companies like Nvidia or OpenAI, or even a MD. Um, if that happened, you know, that could be a really interesting outcome here.

Another partner that we're hearing getting involved in that consortium though, would be TSMC, which seems a little strange. That would be sort of like general mortars and Ford Foreign, a joint venture or something, you know, that, that, uh, would reduce competition, would probably raise some eyebrows. But as long as they're a minority shareholder, I think that it would probably pass, and that would allow, um, maybe a little bit more friendly competition between the future Intel Foundry and the uh, uh, TSMC established business.

So, we'll see where this goes. Um, like I said, so far, the word that I'm hearing is cautiously optimistic from folks within Intel, from the analysts who study the space and, um, from folks in the industry who are Intel customers. So all of this bodes, uh, pretty well for Intel and, and pretty well for, for lip Bhutan.

So stay with it here. We will be covering this story in much more detail as we learn, uh, the direction that Intel is gonna head. 0 license, aiming to provide developers with greater flexibility and transparency.

This move addresses the limitations of existing CICD solutions by combining enterprise grade reliability with the customizability of open source software, by making its code base publicly accessible, Semaphore empowers developers to explore, modify, and exchange the platform to suit their needs and requirements fostering a community driven approach to innovation. Uh, Alistair, I think that you've looked into this a little bit more detail. What do you think of semaphore?

Well, I did a, a quick look at Semaphore itself. I've not previously used semaphore or seen deployments of semafo, and it doesn't have a huge amount of market share, but Mitch Ashley says that this is really important, and Mitch Ashley is the, um, VP and practice lead for DevOps and application development at, at Futurum, of course, a friend to Steven and I as well. Um, he sees that this is a, a really significant move at a time that companies are moving from open source licensing towards more, uh, commercial derivative resistant licensing, let's put it that way.

Um, that this is quite a big move because the Apache two license does allow commercial use, uh, of, of the software. Unlike something like the new public license, uh, the APACHE two does, does allow for commercial use. So it's interesting move that it, that Semaphore is open sourcing the product.

Now, one of the objectives there is to get contribution from community, and they're definitely selling this as being a very community centered way of developing this SEMA SE four, uh, CICD tool chain. Uh, CICD tool chain is about software development lifecycle automation rapidly going from needing a new feature in the software to that software being that feature being safely available in production. And so this is central to business innovation around software.

SEMA four had done quite a bit of community engagement before this open sourcing, and I think that's, that's a, a good thing to see that there's a community edition, which is open source and free for small teams to use for commercial purposes. They provide a cloud edition as a service as well as an enterprise edition for people to run on premises. So there's a bunch of good things in there, and the actual source code is sitting up on a GitHub repository that's accessible and available.

The thing is though, CFO is not a big player in CICD, uh, one of the market surveys that I read that actually I don't have quite to hand, uh, listed them as under 1% of market share. So I'm not sure that's particularly significant, especially in light of the open source nature of Jenkins. So Jenkins is probably the best known, uh, CICD platform around it also is open source.

And so that visibility of what's going on inside the platform that that sefo is talking about is available to other people through who are using Jenkins as well. So I don't think that's a driver to move across to SEMA four, definitely see some value compared to using, uh, CICD platform as a service kind of tools such as the, uh, suite of tools around code pipeline that AWS has, uh, that are much more opaque in how they function and how they operate. And there's always a trade off between the, the value of using as a service components where you don't have to care about what's going on underneath and being able to fully customize the things that you need, the tools that you are using to build out your platform.

So it'd be interesting to see the progress, to see whether this change gets a, a much larger market share for SEMA four over the next couple of years. And we'll certainly be watching that for the, uh, AP dev field day. That's, uh, coming in July.

Microsoft's recent announcement of its major R one quantum chip claim to be a significant advancement in quantum computing that's been met with skepticism by the scientific community. And I raised some skepticism on the rundown when we first covered this experts question, the validity of Microsoft's findings citing a lack of peer reviewed evidence and unproven underlying physics leading to concerns about the legitimacy of the breakthrough. Is this a legitimate child of Microsoft, Steven?

Well, it is a concern for sure, and, um, this is something as well that we've talked about on the Textron gang. And, uh, we've been talking about it internally here at, uh, Futurum. The question really revolves around this whole idea of, um, major breakthroughs and how those breakthroughs reach the market.

Now, I'm going to say that I believe Microsoft has made a, uh, breakthrough, a technical breakthrough here with this quantum chip. I believe that the quantum effects that they're talking about are probably real, but I don't necessarily believe that this is a groundbreaking advancement. And I feel like, uh, they really did a disservice by making it such a hyped, uh, announcement saying, you know, a new type of physics, a new state of matter.

You know, we, we hear this a lot, especially with regard to quantum. Now, what I've learned over the years of watching, uh, technology develop from science to technology to products, is you can't really always predict where things are gonna go. Uh, what is gonna be important, uh, and when those things are gonna be reaching the market.

In fact, it's kind of the flip side of that. It generally comes at us by surprise. Um, raise your hand if you had, uh, graphics cards are gonna be the number one, um, money maker in the entire IT industry in 2025.

And yet Nvidia GTC, which is happening right now, is showing us that, uh, those, uh, humble little graphics accelerators that we used to buy 20 years ago are now the most important element in the data center. You could say the same about flash memory. Um, you know, if you thought that, um, you know, that that flash, uh, was gonna be the predominant storage media, you'd probably be laughed outta the room 20 years ago because it was so small and unreliable and expensive.

And yet it is the indis undisputed leader in storage. And frankly, when it comes to quantum, uh, I think we're not sure what's gonna happen. In fact, uh, you know, you could point out some of the other things that haven't come true in the IT industry, whether it's, uh, holographic, uh, storage, um, the rise of optical interconnects, uh, you know, all these things that were supposed to be revolutionary and are now, frankly, um, just small, uh, items that are sort of, uh, popularized to one extent or another.

Uh, quantum, you know, we're not sure. We don't know if this is really gonna happen, but one thing I can say is it's probably not gonna happen the way that people are predicting it. It's probably not going to be that the iPhone, I don't know, 30 is a quantum powered, uh, CPU.

It's much more likely that quantum is going to find itself a niche somewhere in the, uh, in the IT stack, and that that niche will be something, a workload that's particularly suited for this kind of chip. And furthermore, I think that it's worthwhile for us as an industry to be preparing for that moment. Now we have a good idea of the sort of problems that quantum computers are good at.

And I think that that is, uh, something where we're not gonna let ourselves be taken by surprise, which is why I am glad that, um, you know, NIST and others are working on quantum resistant cryptography, and that those things are being pushed into the market and, um, implemented today because it seems like quantum is gonna be particularly useful at breaking cryptography if we can get enough qubits put toward it. Uh, but that being said, um, problems like those are the problems that we're gonna be finding, uh, quantum looking at a, you know, productive use case. One thing I'll point out though is that there is an interesting, uh, collision, uh, between quantum and ai, and even NVIDIA is recognizing this.

They're doing an AI day at GTC this year. Microsoft, of course, is one of the big, uh, deployers and developers of ai. And I think that over time we may find that that's where quantum comes.

We may find that this technology is particularly well suited for the kinds of computations that, um, future AI models are gonna be doing. Uh, but we don't know that for sure, and it's certainly too soon to say it. And frankly, I doubt that the major on a one quantum chip is gonna be the, the chip that's gonna deliver on that promise.

It's most optimistically likely that this could be sort of one of those pioneering chips that that shows us the direction the market is heading. We'll see, uh, solo io, a frequent field day presenter has launched CAG agent, an open source agentic AI framework designed to automate cloud native operations within Kubernetes environments. Built on Microsoft's auto Gen.

Cajun integrates with existing tools to streamline tasks like configuration, troubleshooting, networking security, and so on, enabling DevOps teams to leverage AI driven automation. We've been impressed by solo io in the past. Uh, it's a very practical company of people who really understand the, uh, needs of the DevOps community.

Alistair, what's your take on Cajun Cajun or K Agent, or maybe it should be K four T uh, following the, the Kubernetes way of shortening things. I kind of like the idea that this is building AI into a platform, that this is the way of using AI to make Kubernetes operations better. And we know that Kubernetes operations can be a little troubling and, um, fragmented at times when you are working with lots of manifests and helm charts and trying to assemble everything together and solo.

As, as you know, from their appearances at Tech Field, they asked smart people, and they built this for themselves to make their own operation of their Kubernetes environments better, and decided it was so good that they could actually sell it. And this is essentially what we're, we're seeing here is a framework for building AI agents that will make the operation of your Kubernetes cluster much simpler for the humans. So the idea with AgTech AI is, uh, the AI identifies an issue and is then empowered for some issues to resolve the issue itself.

And that's a, a pretty useful thing. It's an important part of progression of AI from being something that's about interacting with a human, and particularly generative AI from being something that's about interacting with a human towards being. Something that's actually gonna make changes in your environment is, is going to take load off people.

Uh, the other interesting element in this is that it's, there's a, the whole declarative framework around it in a set of predefined functions and tools that are in here, it's not just here's an SDK for building your agent. It's, uh, it's much wider than that, um, Kubernetes operator within declarative API, um, and oriented very much towards building your Kubernetes or operating your Kubernetes environment, rather than delivering an application out to your end users that has this agent AI in it. So, pretty much a, a differentiated thing.

And as we've discussed many times, uh, with tech fields, say events, AI needs to solve problems within product. It's not a product itself, it's the thing that makes the thing better. As, as, um, I frequently quote Bob Allen, uh, having AI making Kubernetes operations better seems like a really good thing, particularly simplifying the life of the site reliability engineers or the platform engineers who have to maintain and, and operate a large Kubernetes environment that's hosting large cloud native applications.

So I think this is a, a really good thing, and I think we will see more organizations, more businesses that are moving towards this AgTech AI for systems on premises or systems within, uh, organizations. Amazon, Google and Meta are investing heavily in nuclear energy as part of their strategy to achieve net zero carbon emissions by 2050. These tech giants plan to power their operations with 100% renewable energy, nuclear energy will play a key role in meeting this ambitious sustainability goal alongside nuclear.

They're exploring additional technologies that can, uh, such as carbon capture to reduce their environmental impact and address the urgency of climate change and that not particularly great reputation that large data centers have for being good for the environment. Steven, is your data center good for the environment? Uh, nobody's data center's good for the environment.

Next question. Uh, no, um, it, it, it's, it's true. And, um, this, uh, honestly, this whole initiative, this whole, uh, story is frankly laughable.

Um, you know, does anyone really think that, um, Amazon Alphabet, uh, and meta have any idea what the data center is gonna look like in 2050? Do you have any idea what the data center is gonna look like? I mean, we just talked about quantum a minute ago.

Um, you know, we've talked about ai, we've got GTC running right now. Uh, we have no idea. We have absolutely no idea.

And, and that's why it's ridiculous to make these promises. Um, I guess, uh, if you wanna be cynical, you could say they've pushed it out past the current, uh, Washington DC administration. Uh, so at least they have some, um, cover in terms of not promising some kind of, um, ecological, uh, goal that's gonna get them in trouble with the, uh, right wingers in Washington.

But beyond that, oh, man, I will just remind everyone that nuclear, um, fission, although it, it seems to check all the boxes in terms of what we're looking for to power data centers, and especially AI data centers with reliable power, with consistent power delivery with, you know, highly scalable power with, um, you know, not a, a huge greenhouse, uh, impact, uh, has unfortunately been mired in, um, difficulties related to, uh, paperwork and, uh, red tape and nimbyism and, um, you know, cleanliness of, uh, disposing of the fuels and so on. Um, which means that unfortunately, uh, large scale nuclear fission plants, like we've got deployed all around the world just aren't being built. And the ones that are built are not delivering the cheap and clean power that you would think they are.

It's the cheap part that's the problem. Essentially. Nuclear, uh, electric power is much more expensive than fossil fuel power and much, much more expensive than renewables.

Now, renewables aren't as reliable and consistent as fossil fuel, but the fact that they're four times cheaper already than nuclear means that, uh, in terms of, you know, kind of what you actually pay when you go to get it, makes it much less likely that we're gonna invest in a lot of fission, um, going forward. Uh, maybe we should, but we're probably not going to. And then people say, oh, but small modular reactors and micro reacts and thorium and all this other stuff, well, that's all a bunch of hogwash.

Unfortunately, none of it exists. Small modular reactors are an idea that exists in a lab, and they have started to build a couple of prototypes. Essentially, they're the flying cars of the nuclear space.

Could they work? Yeah. Will they work?

Who knows? Certainly not anytime soon. So if you're gonna bet on small modular reactors and micro reacts that sit in the corner of the data center, well, I guess 2050 is as good a timeframe as any to say that we're gonna be using those things, because frankly, we have no visibility there, and nobody can say whether that's gonna happen or not.

As I mentioned, um, Nvidia, GTC is happening right now, and GTC is probably the biggest industry event out there. So Al uh, let's take a few minutes to look at what's being announced. Now, I do have to admit now GTC is going on live on Tuesday, uh, Wednesday, et cetera.

You know, the keynotes, uh, we're actually recording this Tuesday afternoon Eastern time, which means that the keynotes have just wrapped up and we've been watching and we've been following remotely, and we're talking internally and we're trying to keep up to date on what's going on. Um, al at the risk of being entirely outdated by the time this gets published, um, what's your take on, uh, GTC generally and the keynotes so far? Well, I think the, the core of what we see from, from GTC is a fairly standard.

Things are newer, faster, bigger, uh, beta elements. And we expect that, we expect that the engineers who are, who are designing these, these chips, these systems have been working really hard to make them more efficient and deliver more for, for their customers and deliver larger systems. Because trying to scale to thousands and thousands of servers with tens of thousands of GPUs, that gets to be challenging.

So trying to consolidate that, uh, that footprint is, uh, a significant element here. Uh, interestingly, seeing this shift from a, uh, biennial to annual release cycle, uh, I'm not sure whether it's the biennial that means every six months or the biennial, it means every two years. I just love the English language for being so, uh, so, uh, yeah, random about that.

Um, definitely seeing more chips being released, or newer chips being released. Uh, some thoughts around doing inference at the edge out there from, from Nvidia as well, because lots of data generated at the edge, and this is where we need to make fast decisions. So inference at the edge is definitely an, an interesting element of what's going on in the, uh, building AI solutions.

Uh, one that really caught my eye was groups or GR zero zero T to avoid patent issues is a, I am group iron gr, uh, which is a framework for having a humanoid ai. And this is the sort of bringing together of a couple of elements of the actual human shaped things by petal robots that we've seen around for a while, but the need for heck of a lot of processing to be done on those humanoid robots, and that they need to move to being a, a much more general purpose solution rather than being, you know, can we get a, a, uh, a bipedal robot to open a door and carry a, a hammer into a building. Those kinds of challenges that have, have been the mainstay of, of sort of the competitive building of, uh, bipedal robots.

Uh, I'm really interested to see how this framework, and we've gotta find some other way of, of naming it than group, uh, has, is going to come out into fruition and, and turn up. There's definitely some interesting blueprints around this. Isaac is the, the first blueprint of the first group blueprint, blueprint that is out.

Uh, and it's an open humanoid foundation model. So it'd be interesting to see how this all plays out. Uh, I am still waiting for my robot butler.

Uh, I would definitely like somebody to be, uh, doing the, the day-to-day household tasks that is not me, so I can focus on more creative things. Steven, it's your robot butler, um, serving well for you at the moment. Oh yeah, actually, uh, I just sent him out in the, um, micro Fusion powered flying car, um, to, uh, do some errands for me.

Uh, you know, Nvidia is strong with the code word name. So we've got Isaac, which is based on the Newton, uh, physics model. Um, we've got the next generation, um, you know, Grace Hopper.

Uh, you had the grace chip, you had the hopper chip, they combined to make Grace Hopper. We've got Vera Rubin, uh, she discovered black holes. Uh, that's pretty awesome.

Um, so we've got Vera is the next generation arm chip coming out of, um, Nvidia. Uh, it's their own home gray homegrown core, uh, which is fun. Um, just like Apple, they're creating their own CPU architecture, uh, or not architecture fundamentally, but their own CPU implementation of the arm instruction set.

Um, and, uh, seems a direction that a lot of companies are heading into, make some, uh, make, make a server chip of their own. Uh, and then we've got Ruben, which is sort of the successor to the, the next, uh, the hopper, uh, GPU architecture. And so combine those.

You got Vera Ruben, uh, maybe you discover some black holes using your AI infrastructure. Uh, they've also announced the next, next generation beyond Vera Ruben, did you hear this? It's, uh, surely you're joking.

Did you, did you hear it? It's Richard Feynman. Um, I, uh, by the way, if, if you're listening to this and you're saying, who the heck is that you owe it to yourself to go get, get more fineman in your life, and if you, and if you know Feynman, then you know that I'm right.

Um, it's interesting that essentially Nvidia is advancing everything, doubling everything. You know, we've got, uh, you know, the Blackwell Ultra, um, which is, uh, basically twice everything. Um, we've got Nvidia embracing, uh, joining chips together or joining, um, uh, wafers together to make a, uh, or dyes together to make a single, a single processor.

Um, so going forward, it's not gonna be, um, you know, completely separate chips. They're gonna be really kind of closely coupled together. Um, we've got a whole bunch of, you know, positive technology advancements on the hardware side.

We've also got some great software stuff. I'm not sure if, um, if Al if you had a chance to look into some of the, uh, the VMware of AI that, uh, Nvidia is putting together here, um, with, uh, their own, uh, software platform to run applications on. Um, ultimately though, you know, whenever I see Jensen Huang's keynotes, the one thing that strikes me is that I know, shocker, this guy really seems to know the market.

He really seems to understand what's going on here. And, um, and he, and he always gives me a new way to think about things. And, and, and so at CES it was talking about, um, basically moving beyond LLMs to having AI systems that understand the real world and can interact and, and exist in the real world.

And that's one of the prerequisites for this Newton, uh, physics model that under my, or underpins these, uh, robots, uh, you know, he expounded upon that in the keynote in a way. I, I, I really enjoyed, um, talking about the ways in which, um, you know, the, the, the advancements like deep seek with LLM, that's not reducing the need for ai, for Nvidia AI hardware. It's enabling us to go to the next level with, uh, entirely new models that are gonna be much more efficient on sort of a per operation basis.

But that means that we can use them to do more, because that's one of those, one of those laws of computing, right? The more power you have, the more ability you have to, to use that power deep seek just allows us to do a heck of a lot more and still continue to deploy more and more. And so I do expect that there's gonna be new AI models that will be doing things we, we never thought possible, just like today's LLMs are doing things we didn't think possible.

And again, that's kind of where, uh, Jensen Wang is talking about during the keynote, is that essentially we need to move beyond LLMs. We need a hundred times more power than we currently do in order to make even a more amazing advances in ai. And that's what NVIDIA's working on.

That's what the entire industry is working on, and that's why things are still going up into the right in the, uh, AI space, of course, you know, power cooling, manufacturing, uh, impact on society. All of these things are, are some of the downsides and some of the things we're gonna need to think about. I guess.

Al uh, what are your final thoughts on uh, GTC? Well, in terms of that power, I liked the discussion around the spectrum X silicon photonics ethernet switch. Uh, this is partly also great to see more commitment to ethernet from Nvidia.

Not just everything should be in finna band because we own it. Uh, this is talking about having a three and a half times energy savings. Uh, and we've previously in, in, uh, AI field day events, we've talked about what a huge amount of power is consumed simply by the optics that are transmitting all of the, the data in inside these AI computers.

And so moving towards the silicon photonics and, uh, a reduced need for those power hungry transceivers should be a, a really good thing. The other element you touched on was the, uh, Nvidia dynamo, which, um, Jensen talked about as being the, the operating system for an AI factory. And returning to the theme of ag agentic AI being used to operate it dynamo itself, the operating system for your AI factory is going to be using AI to operate itself.

Because operating this factory is so incredibly complex, and this is why I think it's taken quite a while for the concept of the IR factory to turn into products. There are just so many moving parts. Uh, so yes, lots of things moving, lots of innovation going on.

Definitely lots of work on pain points, both for, for the ability to scale. Talking about, um, the Jenssen's example was doing a seating plan at a wedding, very topical. My daughter's getting married next week, uh, next month.

Um, being, you know, being able to do this through AI without, which requires huge amounts, more context, knowing the people, knowing that you can't put these two people's at the same table, even though they've got the same last name because they're divorced, uh, and acrimoniously divorced. So that kind of intelligence is, is gonna require more resource and efficiently delivering it is gonna be absolutely vital. Uh, but the final final thought has to be your, Stephen, was your single biggest thought about GTC and what we've seen in the keynote today.

Well, I, I'm going to say that the biggest thing I think is, uh, this whole idea of bringing AI into the real world, whether it's through robotics or through manufacturing, or through, um, you know, autonomous flying cars, bringing AI into the real world is gonna be the next challenge. Um, we have only just scratched the surface of what, uh, all of this can do with these large language models. And next year, mark my words, we're gonna be seeing some absolutely incredible advancements that are not, not just chatbots.

We're gonna start seeing AI models that really do understand the real world. And I'm excited to see where that goes. So thank you very much for listening.

Um, before we go, let's quickly look at some of the things that are going on now this week as well. Uh, today and tomorrow, Wednesday, Thursday, we've got Networking Field Day happening. Um, we're very excited to be out there.

We've got, uh, BT talking about Quantum, which is, uh, really exciting. By the time this gets published, that will already be viewable on, uh, LinkedIn. We've got Selector ai, we've got Versa Networks.

Uh, we are also coming back in April, early April. We're gonna be heading to Houston, Texas, uh, to visit the campus of Hewlett Packard Enterprise. I'm pretty excited about this one because I'm a big server nerd, power and cooling nerd.

Um, and I can't wait to dive deep into Liquid Cooling and ILO and, you know, management and, and, and, and server interconnections and, and the physical boxes that make all this stuff happen. So again, keep an eye out. That one's coming in April and then, uh, later in April, we've got AI Infrastructure Field Day.

Do you wanna tell us a little bit more about the companies that are presenting there? Well, it's gonna be a pretty massive event. Uh, we have, uh, six at the moment.

So we have key sites, uh, who do lots of network validation testing. Uh, we have Net and Avis and Juniper also in the building, large networks and operating large networks. We have Fon and Ality both in the, the storage space, both the physical storage and storage for ai, as well as object storage for your applications.

A couple more people, we might have a, a really big announcement in in the near future of, uh, some more just holding onto the paperwork on that one until it clears out. But this is gonna be a pretty epic event and I'm really looking forward to spending some time with some great vendors, but also an amazing panel of delegates that, uh, have agreed to join me out there. Yeah, and I'll just, uh, give a quick shout out too, if you're listening to this, if you happen to be in the Silicon Valley area, we're gonna have a social event and you're invited.

So reach out to me or Al and, um, I would love to see you there in San Jose, uh, for our social on Wednesday of AI Infrastructure Field Day, uh, April 23rd to 25th. So thanks very much for watching the Tech Field Day rundown. You can catch new episodes every Wednesday on YouTube as a video or in your favorite podcast application.

If you were subscribed to the Gestalt it rundown, you're still subscribed to Tech Field Day. It's the same feed The Rundown is streamed on Techstrong TV as well. And you can often catch us on other Textron uh, programs, including you can find me every Tuesday on the Textron Gang.

And, uh, you'll see Tom and Alistair involved in other Textron and rum broadcasts as well. We'll be back next Wednesday to talk about all of the great IT news from the week that was. But until then, for Tom Hollingsworth and the Networking Field Day crew, for myself and for Alistair Cook, here's wishing you and you, or a super sparkly day.