AI’s Reckoning: $725B in Spend, Weak Utilization and New Security Risks

Hello everybody and welcome to the latest edition of the Techstrong Gang. I'm your host, Mike Vizard, and today I'm joined by John Schwartz and Stephen Foskett, and well, we're going to talk about a lot of things today, but it's going to start out with money and IT, because man, it's starting to look like IT in the age of AI is starting to feel like a money pit, right? 5% to some sort of ridiculous number in the $6 trillion range, which I think means that overall increase of about, I don't know, I can't do my math on my fingers and toes here, but I think that's somewhere north of $700 billion in a single year.

And that seems to be a drop in the AI bucket compared to some of the reports on what big tech plans to spend in this whole space. And you got to ask yourself, where is all the money coming from for this? I know that the big tech folks are probably leading the charge on this, but other companies will have to spend significant amount of money on this.

And the thing that comes to mind when I look at it is utilization rates of our current infrastructure, and GPUs included, is just flat out abysmal. And so we are spending a fortune on infrastructure that we're barely using, and so you got to ask, is all this stuff sustainable? John, what's your take on what's going on here?

Well, we use the word staggering a bit too much when it comes to AI, but I think in this case it's appropriate, especially after big tech announced its recent quarterly results. So first, for context, for decades an annual capital expenditure, or CapEx, was in the tens of billions. That was considered aggressive when it was that high.

Now we're talking about hundreds of billions of dollars. So just for an example, we look at the hyperscalers, so to speak. Their collective spending for Amazon, Microsoft, Alphabet, and Meta is projected to reach $725 billion this year, which is up from a previous estimate of $670 billion.

So the thinking among these companies is that they all understand the risk of under-investing in AI CapEx is significantly larger than overspending. They claim, and I talked to Daniel Newman, the CEO of the Futurum Group, about this. He says that less than 2% of people are paying for frontier AI models, so we're still early in the process.

And you can make the argument that that number is woefully low, and it should be higher to warrant this ridiculous spending. I think what's going on is the scale of investment is driven primarily by the need for data centers, power grids, high performance silicon required to run generative AI models. So that's why NVIDIA and TSMC are doing so well.

But again, for the major hyperscalers, they're spending at a such an aggressive rate. The percentage of their spending as part of operating cash flow is projected to surge to 92%. So in other words, every dollar they earn is being plowed back into NVIDIA chips and power grids.

So the money is... I'm thinking about this in terms of where the money's coming from. They have big cash piles, but they also, I think, are looking at ways of trimming expenses to put into AI infrastructure, and that's why I think we're also seeing a ton of layoffs.

That's where AI is headed. So Mike, in other words, yes, it is a money pit. It reminds me of the movie with Tom Hanks, where they keep putting money into this house that keeps not being built out or is not ready for occupancy.

And I'm kind of getting that same sense with what's going on with this frenzy. Stephen, will these costs essentially, at some point, you would think they'd have to be passed on to the end customer. At some point, somebody's got to pay for all this stuff that the hyperscalers are doing, and is that going to be something that organizations are prepared for?

Or this time next year am I going to see a bunch of CFOs rioting in the street? I don't know if the CFOs are going to be rioting in the street, but I do think that the bill's going to come due. It has to come due.

That's the challenge. You can't sell dollars for a dime forever and expect to make it up on volume. 10, or at least a dollar.

And I think that that's really where this is all going to come home to roost. I think that, as of now, I don't want to be too depressing here. As of now, it seems like there are some paths forward where we could actually make use of some of the infrastructure that's being rolled out.

I think that the rise of agentic AI and the promise of AI agents is a massive pool that could suck up an awful lot of GPU capacity in a way that frankly, chatbots and AI-enabled applications couldn't. But that being said, the big problem here is that essentially, in order to pay for that level of infrastructure usage, there's going to have to be an awful lot of revenue coming in, and I'm not sure that these products are compelling enough yet for people to part with a huge amount of money. Just to think about it, many of us, for example, spend probably $50, $100 on streaming services for video, sitting at home watching television, those kind of things, various sports and so on.

That's tiny in comparison to what these companies need us to spend on our AI agents in order to make agentic AI pay for the infrastructure that it's using. My feeling is that we're going to have to convertWe as an industry, not me, but we as an industry, if this stuff is going to pay for itself, there's going to have to be hundreds of dollars per month rolling in from average people, tens of thousands of dollars per month rolling in on behalf of every single employee in enterprise businesses, and literally millions per month rolling in from corporate applications of agentic AI in order for this to really make sense and to pay off this incredible investment. I'm not sure I see the compelling use case for those kinds of- Yeah ...

yeah. It's interesting you say that, Stephen, because I think the market is showing some level of patience, but I think that's going to start growing thin unless we come up with some sort of killer app or some sort of agentic AI use across enterprises or even among customers that push that 2% of people who are paying customers up to about 20% by next year. There's got to be a significant bump.

And so far, based on what I've heard and seen from some of the executives at these companies rolling out these AI agentic platforms, when I press them about their customer use, it's fairly early stages. It's not clearly absorbed throughout the organization. And I wonder if that pace is enough to mollify the markets because these companies, again, the big tech companies are spending a ton of money, and people are going to want to see a return on investment.

And we're going to start seeing in the next few quarters even more pressure for them to show significant improvement in terms of AI revenue. Right. And to your point, the minute I start to do anything interesting, I start running into token shock, right?

Mm-hmm. Once I get into any kind of complicated reasoning, the bill for these things goes up dramatically. And I think that ultimately, if I need some ROI, I got to have something that has some significant reasoning behind it, because otherwise I'm just kind of creating better memos and funny-looking images for the social media folks.

Am I wrong, Stephen? Yeah. Well, and on that point, I think that that's one of the reasons...

So a couple of weeks ago, I coined the term super villain cosplay marketing. And essentially, you've got all these AI CEOs out there talking in a way that sounds completely asinine, talking about how AI is going to threaten the existence of humanity as a species, talking about how AI is going to spell the end of 50, 70, I don't know, 90% of all jobs, things like that. " Why would you go out there as the CEO of an AI company and say, "Hey, guess what?

" Why would you say that? Well, I think the reason that you say that is because that's the only way that these products could pay for themselves. So think about what's the biggest expense still for most businesses?

It's people. It is salaries and benefits and so on for people. And if you can go out there as the CEO of an AI company and say, "Guess what?

" That's the reason for this sort of bizarre, apocalyptic marketing message. Our product is so powerful and so incredible that it'll allow you to fire most of your staff, because that's the only way that any of this makes sense. Now, will that happen?

I think that a lot of people's experience with AI is this is really cool technology. It's really powerful technology, but I don't know that they're really going to fire people. Now, they are firing people, but I think a lot of those people were sort of over-investments in people following the pandemic from a lot of these companies.

Yeah. What I'm hearing from the analysts at Futurum internally is that essentially, companies hired a lot more people than they needed because they were sort of defensively hiring, just in case the boom continued. Now that things are shifting a little bit and we're looking at AI, maybe some of those people aren't needed.

I don't see a lot of companies saying, "I was able to cut my entire staff of mm and continue doing that thing with AI at a cheaper rate," because I just don't see it. You are-- I'm sorry, Mike. You articulated, Stephen, what Dario at Anthropic, Amodei, has been saying, but he says it in such stark terms, it's terrifying.

And he almost needs to stop doing that because his messaging is totally undercutting what they're trying to do. What I would surmise is that he's trying to warn us about what OpenAI is going to do. So maybe they're the alternative, but maybe with these companies, they've got these legacy models, and maybe they're shifting to this new computational standard.

So in the process, that's where the efficiency comes or where the revenue eventually comes. But my biggest takeaway, too, is if you're an employee at one of these places, you must be terrified. " I forgot who said- If the CEO believes it, right?

Yeah. But I forgot who said also, I think it was last week, talking about how, in some instances, organizations are laying out a massive amount of money in AI to eliminate jobs that we're not talking about half the company. We're talking about 10 people who aren't making enough money in total to justify the investment in the AI in the first place.

So we're eliminating a boatload of entry-level jobs that don't quite add up to any kind of math that works for the AI investment. Yeah. Well, Meta did that.

"So in a sense, it was 14,000, but 6,000 of those jobs were jobs they hadn't determined what they were going to be, or they had an idea and they decided we don't need those anymore. All right. So I'm also optimistic, though, on a couple of points, and Steven, I'm going to run them by you.

But it does seem like in the history of IT, necessity has always been the mother of invention. " So is that another opportunity and a frontier? And then I look at all these processors maybe, and I can't even pronounce this company's name.

Cerebras, is that right? Yeah, that's close enough. And they are creating what looks like a new AI chip.

Some people are skeptical about whether this is going to work or not, but the point is that maybe GPUs aren't the answer, and if we can rearchitect the semiconductors and the systems and everything that goes with it, suddenly the math on this stuff makes sense. But Steven, how long does that take to play out? Yeah, I think that there's sort of some pros and cons to that.

Certainly, we've had Cerebras present at Tech Field Day, for example, and everybody was wowed by what they do. They make a processor the size of an old LP record that has all sorts of memory and processing and everything on board. It's some really just incredibly cool stuff.

I wish them well. The challenge is that even that is really a pretty centralized piece of kit. And what you're saying, and what I think a lot of people are hoping for is more of a decentralized future, where you have incredible processing power and you're running models locally.

Now, I've been really working on that skunkworks-wise here within Futurum of getting models running locally, and it is possible to run some pretty powerful models locally. I've got a 23-billion parameter Qwen model running on my Mac Studio that runs just great on 64 gigs of RAM. The problem is that's a pretty big honking machine, too, and even that can't handle the workloads of agentic AI because it doesn't have enough token memory in order to do tool calling and remember its commands and so on.

You really need more like 128 gigs of RAM in order to have a contact space to be able to do a tool calling agentic system. That's a lot of money at this point, and that's very, very difficult to do. It's not something you're going to run on your laptop or on your phone.

But that being said, there is a really cool thing here, and that is that this technology has shown us that we can do pretty neat stuff with transformers and with large language models and with neural networks. And for me, the thing that I'm most excited about is not running ChatGPT on my laptop. What I'm excited about is running all sorts of other really cool neural networks on a variety of small portable systems.

You can do really amazing photo image recognition, video processing, transcriptions, all sorts of things that aren't chatbots, and you can run those locally already, and they're great. It's really neat that we have this cool technology. What we can't do, really, is run anything up to the level of frontier models on local hardware.

And I don't think we're going to be able to in the next four, five, six years, simply because memory is so expensive and processing is still so expensive. And even the best NPU companies, Apple has shown that they can do just incredible neural network processing on their M5 CPUs. You look at the benchmarks of those things, it's really incredible.

You look at what AMD is doing, what Nvidia, Qualcomm, a lot of these companies are doing, it's really powerful, but it's not enough because we just don't have enough memory. And so I love the idea of decentralized AI. The problem is that the hype AI, which is artificial super intelligent chatbots, that ain't ever going to run locally, and that's going to be a big problem.

John, in the Valley, are people just willfully ignoring math and economics and all this stuff and just kind of walking around and charting this incredible future for AI without anybody considering how this thing might actually play out from a cost perspective? It's we harken back to the let's move as fast as we can. We'll break things along the way, and we'll learn from our mistakes.

I'm almost starting to think that we're starting to hear murmurs of companies scaling back some of their AI infrastructure projects, like Stargate, for instance. I keep hearing back that's going to be scaled back in certain ways. It already has been, I guess.

I think there's always the fear of missing out or becoming the BlackBerry of AI, where you're left in the dust by your competitors. So you spend like crazy. Remember Meta?

Remember the Metaverse? Oh, we got to get there. We got to get there.

Let's spend billions of dollars. Oh, we got there. It's not what we thought it is.

Let's just scrap that. Now we're going to go after pursue AI. Now, as long as you bring in the revenue, and in the case of Meta, it's with advertising for the most part, you can do that.

You can get away with it. But I pity the companies that are not in that position. The hyperscalers are.

They're filthy rich. Their valuations are stratospheric. They've got piles of cash.

But they're going to burn through some of it, and I always kind of wonder what's going to happen down the food supply chain. What's going to happen to those companies? All right.

Well, I think the Piper's bill is going to come due one of these days soon, and people are going to start asking hard questions. "All right, folks. Stay tuned and keep your eyes on this whole thing, because it's going to have a lot of downstream cascading effects, and I got a feeling we'll be talking a lot about it in the months and weeks ahead.

I'm going to shift gears, though, to something that's along the same idea, but we're going to geek out for a minute. There are things called a knowledge graph, and we've seen some of them in our various applications that people are building, and they create the links between data, AKA objects, and establishes relationships. And you can point AI agents at these things, and there's an argument that says that that's more reliable because, well, the sad truth of the matter is when it comes to gen AI and all these AI foundational models, is we still don't know how they work.

We love the outcome, and we understand the fact that 90% of the time it may be accurate, but 10% just seems to be about an issue, and we still can't empirically show anybody who is going to do an audit how this thing actually came to its conclusion. There are folks now starting to argue that, well, what we need to do is add a knowledge graph to that, and there's a small company called Lovelace, startup, that did that. They have a thing called Elemental, and it basically creates a knowledge graph, which you can think about as a massive index, similar to Google, except that the relationships between the data has already been established.

Normally, I would just chalk this up to some inventive AI capability by a startup, but this one is led by a fella who's actually the dean of computer science for Carnegie Mellon University, so he seems to know what he's talking about. And he used to work for Google as well. " And he's also arguing that if we shift the AI agent's more reasoning capabilities to the knowledge graph, well, we'll be, A, less dependent on a specific LLM, but B, it will be lower cost ultimately because it'll be more efficient.

Stephen, I'm going to throw all this in your lap, but I kind of feel like we're down this massive path here with AI, and we don't really completely understand how it works. And if so, do we need to take a minute and take a breath? Well, I do think that there's a lot of misunderstanding of what AI can do out there.

It does drive me crazy on a daily basis to listen to some of the things that people say and to see people go down that spiral of thinking that just because you think it's human, it's human. It's not, folks. It's just not.

But that being said, like I mentioned in the previous segment, I am not an AI naysayer. I am a AI hype naysayer. I think that this is an incredible new capability we have.

We have an incredible new technology here that can do amazing things, and I use it every day for 50 different things. The idea that you can combine a neural network, that you can build a model based on the same kind of transformer capabilities that you would use to build a large language model, and that you can do something in a specific area with that, is just ripe for the picking. And I wish that there was more companies out there, like Lovelace, that were saying, "Hey, how can we figure out a way to apply category X to this technology?

" And what you're going to find, I think, is that companies that are doing that are going to be vastly more exciting and successful, and have relevant products that actually make money, than the companies that are just in this race to spend as quick as possible in order to build the biggest large language model possible. And so I look at that, for example, we've got AI Field Day here coming up. Some of the companies that are coming in there are talking about basically applying this technology to FinOps or network management.

And they're not just bolting on a chatbot. They're creating a new technology based on what needed to be done and what these products are good at. And I think that if you can add information, if you can add a knowledge base to what is fundamentally a limited, a slimmed-down language model, you can do some pretty cool things.

And so that's why I'm excited about this. Now, I don't know enough about this particular platform to make a judgment on it now. Maybe none of us do.

But I love the idea of we're not just fighting to make a bigger chatbot. Hmm. " And ultimately, if we start putting these AI agentic platforms in place, and we tell people that they're going to be autonomous, well, they're going to do all kinds of things that are unplanned or unexpected, and we're not going to be able to roll it back.

So unless there's something that feels like, it doesn't have to be a knowledge graph, but it feels like something has to sit alongside that that provides the checks and balances for those workflows to understand what exactly is going to happen and provide some sort of heads-up that says, "Hey, don't do that" to the AI agent, or tells us that this thing's about to go off the rails. And pardon me, but for all the investment in computer science, I might file that under just plain old common sense, John. Yeah.

Yeah, I think you're right. Yeah, it is about common sense. Can I ask you a question, Mike, about this?

I don't know if there's an answer, but when we talk about knowledge graphs, where are they most effective, or where do they win? Is it in governance or auditing precision? And where they fall short The way to think about it is they've already established the connections between various data points, AKA objects in this case, so that either the search engine or the LLM itself isn't trying to lay that on top of a bunch of random facts, that it is connecting the dots between that may or may not be correct.

Essentially, some human has gone in there with some data science and established the relationships between these things, and then that becomes the source of truth for the AI agent, which helps in everything from security to governance to the relationships. And you see these things at a smaller scale, like Microsoft has built them into Office, and it's a compelling way to think about how to put all this stuff together. These guys are arguing that the cost of the infrastructure has gotten to the point where we can do this at the levels of scale where we can connect the dots between trillions of objects.

And that becomes a new type of Google search index engine, you could think about it that way, specifically for AI agents. To Steven's point, I have no idea if this is going to work or not. The math seems to be there.

The folks behind it seem to know what they're talking about. Yeah. I'm pointing out that LLMs may not be the end-all and be-all of this game.

Steven? Yeah, and that's exactly right. I think that the difference here, and this is what I've seen in my work with LLMs, essentially, if you're treating it like an artificial super intelligence that's just a black box that doesn't need anything external, then you're doing it wrong and you're expecting too much from it.

John and Mike, I'm sure that on a daily basis, you're out there looking up facts you don't remember. Who was that that I talked about? What company did they work for?

What was their revenue number they mentioned on the call? No, you look that stuff up. You don't expect it to remember that, but yet we expect these LLMs to just know everything.

Well, they don't. No. We have to give them tools, we have to give them memory, we have to give them databases, and then we have to see what they can do with that.

And what they can do with that is pretty amazing. I, for one, use, for example, the Google AI Studio with Gemini all the time plugged into the Google workspace, and it is insanely good at what it can do. We use on Slack, we'll use Slackbot, and it's insanely good what it can do by having a source of information instead of just expecting it to know everything.

It doesn't know everything. Let's give it some information. Right.

" Sat at the end of the bar and he would always- Cliff, he knew everything. Yeah Yeah, we expect AI to be that, right? Except it's worse than Cliff, who just makes stuff up when it doesn't know, and then just- Yeah, exactly ...

" But then it turns out, no, that's not the case whatsoever. You mentioned tokens. Oh, can I mention, Mike, something that Moore said that was interesting is he mentioned that relying on LLMs to reason across massive databases can be token expensive, and that plays into the cost factor.

And I'm wondering, in a broad sense, is audit what the agent read enough, or do we need policy enforcement at a retrieval time, or...? Yeah. You're absolutely going to have to have policy enforcement at retrieval time, because otherwise, the AI agent is programmed to accomplish whatever task it thinks it wants to assign at all costs, and it does not care what data it needs to go access or how sensitive that data is.

And sometimes it will tell you that it did not access that, when in fact, it surely did. And turns out that these AI agents attached to LLMs have a propensity to lie to us, which is kind of shocking, but here we are sitting with these things. So I think there needs to be some sort of external policy control, but I'm pretty sure we're going to get to that in our next topic.

Yeah. Okay. All right.

I think we're going to go jump to that topic. All right. Steven just had another one of his tech field day events and was talking about cybersecurity, and I'm sure AI was at the top of that topic list.

But Steven, what's top of mind for folks in the realm of cybersecurity, and what are your takeaways from this event? Well, certainly, what you were just talking about is top of mind. A lot of us are concerned by what AI means when it comes to cybersecurity.

" It is more like what we were hearing from some of the companies and some of the delegates there, which is essentially, let's move forward with a new way to approach security, which is not actually a new way, it's an old way. It's basically the whole zero trust model, which says we have to make sure that things are segmented, that we have access controls throughout, and that we're approaching things on a very granular basis because shadow AI is here, it's coming, it's going to be everywhere, and we have to make sure that the security products can monitor and manage what customers or employees are doing in order to make sure that they're not creating holes by using these AI models. That's what we heard about from Fortinet at Security Field Day.

Another aspect that we heard from Dell was that security has to be a fundamental product of designing all systems. "But it should be. " Or you can say, "You know what?

We've got to fight for this. " And then another thing I want to bring up is, despite what we're hearing about AI, there are other topics in the world. It's not just AI.

One of the things that we heard a lot about was quantum safe cryptography. That was a big topic at Security Field Day, and it was a very interesting conversation as well. We had some really good presentations.

Just check out the Tech Field Day website for that. You'll see the presentations that talk about quantum safe cryptography. Essentially, the TLDR on that is that we have this.

We are ready for it. We just don't yet have the will to roll it out because it is a change. Tom Hollingsworth led the delegates in a roundtable discussion on quantum cryptography, and that was Tom's point.

Basically, who's going to be the first to stand up and say, "We have to be quantum ready," because otherwise, we'll never be quantum ready. Basically, you got to stick the flag in the ground and say, "Okay, here's where it starts," instead of just waiting for this thing to happen. Because it's true that quantum isn't breaking cryptography today, but today's encrypted data is still going to be there in 10 years when quantum is maybe breaking encrypted data, and we ought to be ready for that by acting today, since we have the tools.

We just don't have the will to use them. So I know that you all spend a lot of time on Security Boulevard. You're writing about these things.

What are these three topics? What do you think of these? I don't think it's going to take 10 years.

I think it's going to be closer to three, but- ... here's the part that I struggle with, and it kind of relates to the other two topics. I get that the encryption schemes are going to break, but at some point, people are going to combine AI and quantum, and we're still not going to have the security capabilities baked into these services.

Other folks who are malicious actors are going to use their infinite resources to create their own versions of these things. They will launch attacks at machine speed, and we will have to create another cybersecurity overlay to respond to that in near real-time and faster than humans can respond to or be aware of. And just for fun, we're going to have all these AI agents that we previously just disclosed can't be trusted to help us.

Yeah, so that's what I was going to ask you. With the genic workflows spreading, what are going to be the new failure modes, for instance? I'm not sure there is control.

I'm not sure what it means to be resilient in the face of all of that, because- Mm ... it just may mean that your entire IT environment needs to be disposable, and if there's an attack, you need to instantly be able to, for lack of a better phrase, rehydrate the whole thing and start over again with some sort of pristine form of data that you can spin up inside of, oh, I don't know, 30 seconds. What do you say, Steven?

Possible? Well, amazingly, it is possible. We've talked to these companies out there in the data protection space.

I bet if you talk to Rubrik or Commvault or companies like that, they're going to say, "Yeah. Yeah, we can do that. " I guess the question is, do you want to?

Do you want to invest in that? Is that something that you believe in, and is that something that you want to pay for? Because I think the technology is there, just like we were talking about knowledge graphs before, what we're talking about with small neural networks.

We have some really cool technology now. The question is, are we going to invest in that technology, or are we just going to be all, "Ah, chatbots," and never actually put money down to make new capabilities? Mm-hmm.

Is this just another example, John, where we're investing in one technology so that we can create a whole mess of other issues that we then have to invest in- Yes ... other technologies to go solve? So I'm going back to the first segment.

So where these capital expenditures and the hundreds of billions are being poured into GPUs and custom silicon. So then you create, in a sense, eventually, new supply chain firmware side channel risks that you then have to address. Right?

It's just kind of a domino, but a negative domino effect. In our haste to get to the end result, whatever that may be, and we don't even know what the end result is, we come across and we create even new issues. Maybe we create new industries and new types of cybersecurity opportunities for startups.

Mm-hmm. Steven, is it any wonder that business people kind of look at IT folks with a certain amount of skepticism these days? Because I think they're cottoning on to this thing.

What do you mean these days? I think we've got enough gray hair between us to know that's been how it's been for a long, long time. I think there's an opportunity here for IT pros, the kind of people who are listening to some of these shows, to raise their hands internally and say, "Hey, look, folks.

I know you're worried. " And I would love to see that. I would love to see people raise their hands and say, "I know you're terrified of ransomware and mythos creating new zero-days and getting into our network, but we can protect data.

" Or, "I know you're terrified that data is getting out there in all these SaaS applications, but we can get a handle on that. " And I would love to see companies doing that. I'd love to see our peers doing that, offering that capability, and I'd love to see companies giving them the money and the time required.

My fear is thatIn the rush to try to implement or to invest in something, chatbot, that they're not going to spend the money on these products that do exist. Mm-hmm. And my fear is that too many of our brethren are chomping at the bit on the whole AI thing and buying into the hype and selling that into their organizations without standing up and saying- Yeah ...

"Let me be your guide for this. There's massive change here. " And frankly, if I was going to end this conversation, I would say, that thing about AI is made up of two words, artificial intelligence, and there's not enough emphasis on the artificial part because we all think the intelligence stuff is going to magically work when most of it, eh, not always going to be right, not always going to be here, and somebody's got to ultimately supervise it.

Mm-hmm. Humans. John, what do you say?

Yeah, I think you're absolutely right. It's just this rush. Again, it's like the fear of being left behind, and I think the consequences have been somewhat overlooked or maybe compartmentalized.

I do find it, though, encouraging that some of these companies, though, are actually reaching the same stage that Stephen and you, Mike, have mentioned, where they're thinking about governance, they're thinking about some guardrails, they're thinking more about security than they have before. They're thinking about the ethical use. So they are trying to start to map it out a little bit because I think it's starting to hit them or starting to occur to them that they could be creating a lot of headaches for themselves, and they're trying to figure out how to triangulate things before they totally reel out of control.

Mm-hmm. Yeah, I guess. I'm going to leave it here, but I swear when I watch the AI execs talk about stuff when they're on one of those financial news networks, that's when they start the hype meter goes off the charts because they're...

And I think they're trying to talk to investors. I think when they go talk to actual IT folks and the people are actually using this, I'm hoping that the tenor of the conversation changes. I do think there are dual voices.

I think there's the overselling, marketing, hype-driven voice for the markets and the investors, and then there's probably the more down-to-earth, reasonable, practical IT. I hope that's the case. If not, we're in for a reckoning very soon.

I guess so, but hopefully everybody remembers that being credible matters. All right. Hey, gentlemen, as always, thanks for sharing your knowledge and insights, and thank you all for watching the latest episode of the Techstrong That gang series.

And stay tuned for the rest of the Techstrong That TV lineup. It's a replay behind us, and there's going to be some awesome new stuff coming up. Stephen, the next Field Day is when?

This week, actually. Wednesday through Friday. We're excited to have Mobility Field Day.

Now, if you are tired of AI, I guarantee they will mention AI. But happily, they will not just mention AI. This is going to be a packed event.

It's going to be streaming right here on Techstrong TV. We've got a whole bunch of companies in the wireless space. If I can be honest here, I love Mobility Field Day because everything they say is kind of new to me because I'm a nerd, but I'm not a wireless nerd.

But it all kind of rhymes with what I know. And so I'm always able to learn something new, I'm always able to follow along, and it's always cool. And the thing is, it's relevant, too, because all this wireless Wi-Fi stuff, it impacts us on a daily basis because we're all running wirelessly now.

And so it's a really neat thing. So tune in for Mobility Field Day. Again, Wednesday through Friday.

And then next week is AI Field Day. Yeah. And so we're going to be in San Jose catching up with a whole bunch of companies talking about AI stuff.

So tune in for that as well. Can't wait to see it. Yeah.

And when I get that billion-parameter AI thing running on my watch, I'm going to need a 10G network. It's going to be great. Folks, take care, and we'll see you next time.