AI Expansion: Infrastructure Shortages, Biotech Deals and Military Dogs

Hello, everybody. Welcome back to the Techstrong Gang. It's Thursday, and, geez, there's just no shortage of things to continue talking about.

But we have some of our Thursday regulars and some newbies, so let me walk around here and introduce everybody. But first, congratulations to the Milwaukee Brewers for winning the World Series. John tells me they've been in existence for, what was it, 57 years?

57 years. It's their first World Series, yeah. All right.

First title. Well, I wish them luck. I'm hoping that it doesn't become a dynasty, but if it does, well, hey, there could be worse teams, let's put it that way.

And of course, it's a celebration of small ball and small teams and low budgets. They can win. So with that, we're going to jump in to our topic of the day.

If you've been following the show all week, we've been talking about AI in the enterprise. And we started out Monday talking about things like ROI, and then we talked about whether or not people had the skills and the capabilities to actually implement any of this stuff. And then finally, we concluded that maybe they're just going to need help from the likes of SAP and Oracle to implement this stuff, because nobody seems to have the ability to leverage all their data in a way that's meaningful.

Of course, small companies are rocking and rolling with all this stuff because they don't have as much, shall we say, governance issues that they've got to navigate. But let's take a look into 2027 from here, and IDC has a report out that suggests that the cost of AI, while maybe the individual cost of a token is going to drop, but the total cost is going to continue to climb. And that begs the question: Will people have the budgets necessary to run AI?

Because it doesn't look like the supply chain for AI or any part of it is going to become any less constrained anytime soon. And that goes all the way down to the chips to everything cloud, including the servers and infrastructure and networking. So Guy, as you look out into 2027, what are you expecting to see here?

Well, our own research at Futurum in a lot of ways corroborates this research, and our forecasts are somewhat similar. Supply constraints are very often talked about. It used to be constraints in GPU supply, because there weren't enough fabs.

I think there probably still aren't quite enough fabs. But quickly it became apparent that power and maybe even water constraints and siting constraints in the United States and Europe, difficulty in finding sites for data centers. What's maybe being left out of this whole equation, now we are talking about 2027, which is coming up real fast.

But this is, as far as I know, still an economy that runs on standard economic lines. So where there are supply constraints, there can be a reaction and a lowering of demand. But specifically here, we're talking about a lowering of demand of the kind of AI we're doing now.

And I don't know if anybody has noticed. How old is agentic, like a year? Not truly a year, right?

But as far as a big movement, it's a year or less. Mm-hmm. My point being that this market has changed really quickly.

One example is the rise of non-foundation models or maybe foundation emulation, emulating models like Qini and whatnot, that do a lot of the same things but at a very lower footprint. We are still in this world where demand seems far, far greater than supply. My point here is twofold.

First of all, we are judging against a model of what AI should be used for, how it should be used, and where it should be hosted. That will change, and it will change quickly compared to many of the other technological surges that we experienced on cloud, internet, and so forth. And then the second one is that the kinds of supply constraints are experienced very differently in the hyperscalers and neo clouds, where it's siting data centers, getting power, but not so much getting GPUs, versus let's say in the enterprise, where it's a lot more about maybe getting GPUs, and not really so much about power.

On the third hand, everybody is suffering from the silicon and memory supply problem. I don't know if we have this model of AI that does not require a lot of context, a lot of KD cache or cache, a lot of memory needs. So if I'm looking ahead to 2027, along with my colleagues at Futurum, what we're basically saying is we believe enterprise adoption is not going to be as constrained as a hyperscale and neo cloud build-outs.

But I think if we're honest, we also have to say, like this is going to be a changing story every three months. Yeah. So let me ask Drew what he thinks about this.

But to your point, there's going to be thousands, maybe tens of thousands of AI agents, and they're all very aggressive when it comes to how much infrastructure they want to consume. Yeah. So as I look at the situation, we didn't have enough infrastructure for when humans were trying to use AI, so how are we going to support all these AI agents?

And Drew, to be frank about it, are we on the cusp of having to figure out how to ration our infrastructure to support what AI agents we are going to run? Well, let me say this. I think if you are running sort of like an AI forward operation, and you're not doing a ton of work to constrain your token usage, you should stop doing that business.

We are already at the point where if you are using AI to do work, it is essential that the usage is constrained, right? From a security perspective, this is very hard because anybody can run anything now, right? But I want to talk a little bit more about the economy of this, which is sort of the funny part.

Any econ folks out there might remember monopsony. Monopsony, yes. Monopsony, right?

Single supplier. Single supplier. We're close to that.

Single consumer, sorry. Single consumer. We're actually pretty close to that in terms of the use.

Maybe an oligop-- I can't even pronounce that word. Oligopsony. Thank you.

And so the economics are very strange, right? This doesn't follow normal supply and demand. The way it's funded is very opaque.

So even just trying to figure out what these costs are is really hard. One of the things that I read in Anthropic's recent report, where they revealed that they're spending an enormous amount of money on compute costs, 80% of that compute was for R&D. So 20% was all their customers running agents, right?

That is Anthropic's usage of their own compute to build the next model and do work, right? Their employees are using it. So the model now is as inference costs and training costs get lower, it's sort of like when you add a lane of traffic, just more cars show up, right?

Mm-hmm. And so the structure of it now is that any available capacity that comes out of more efficiencies will be immediately consumed by, "Hey, let's do some more training. Let's run some more agents," right?

There's just that much demand. So I think the economics, I think we're in a pinch for the foreseeable future. The prices of tokens are just the most bizarre, opaque things in the world.

And I think we're going to have a lot of fun trying to figure out why we're spending so much on AI for the next couple of years. " Have another. Yes.

There's some numbers that jumped out in this report that I just found startling, almost disturbing. 5 billion AI agents are going to be deployed by 2030, and 95% of enterprises already have an agent workload. Wait, that's it?

Oh, you think it should be higher? You think it would be higher, okay. That's way less than one per person on the planet.

By 2030? Oh, I think that's an underestimate. I genuinely think that's an underestimate.

Okay. Well, going on just a couple of other things. So this IDC report, and again, we can debate it, found that two-thirds of enterprises exceeded their AI agent budgets.

But also, there's no relief in sight, right? Basically, IDC expects lower cost for AI action are going to actually encourage far greater use. So we're going to drive up the total spending despite improvements in efficiency.

So this constraint issue is just going to not go away. Yeah, that's why it's structurally locked in. Well, I do want to rant a little because I haven't had a chance to do this in at least a month, about how freaking inefficient AI is at every level, at every stage, left to right, top down.

And by what I mean is the humans in the loop or the humans using it are using it grossly inefficiently. Mm. The context windows used are half unnecessary.

There's some good tools for that sort of thing now, but those are half unnecessary. The models are... We're still in maybe even baby stages of model development.

You don't need a foundational model for everything. Everybody knows that. But the theme I've had personally lately is AI is so easy to interact with.

It's this human interaction that comes from at least the LLM on top of whatever you're using, that people lose their heads. I think I actually wrote that in an article today. People lose their heads and abandon all of the kind of business and functional principles they ever had, and they just go for the easy button.

You can see this from AI engineers themselves who write a paragraph with a whole lot of begging and open-ended stuff in it, telling the AI to use a ton of tokens in order to create whatever, a comic book or something like that. It's absurd. I think go on, but go ahead.

So what you're just saying, Guy, really does make sense. Mike has mentioned this repeatedly, I think in the last couple weeks is the humans are using AI like they use apps, right? They only use like 20% of the functions.

So in a sense, it's a continuation of how we are using patterns. Well, so one of my sayings is, sorry Mike. One of my sayings is it takes five minutes to give somebody five things to do, each of which takes five months to accomplish, right?

And we are using AI in kind of the same way. We are trying to ask it something really simple without investing our time in trying to ensure quality and completeness and all that other sort of stuff. And then we're taking whatever comes out and going, "Oh, that looks pretty good," and passing it along.

That's true. But wait, there could be hope in here. Let me see if Drew can walk me through this.

The hope is in the inefficiency. I swear I'll shut up now. But hope is in the inefficiency because as we use it more- Let me finish the thought.

Go ahead. GPU utilization rates are atrocious. They are worse than any CPU utilization rate.

So is there, under the heading of necessity being the mother of invention here, can we, Drew, come up with more efficient ways to use AI so that we can support all these AI agents and get better mileage out of our existing infrastructure? It seems like that's the natural way to go, but that would be- Absolutely. And Guy's talking about it is exactly on the money.

This is a skills gap issue, right? So if you're a novice and you go to use your AI to do work, I'm not talking like use it to chat, whatever, that's different. If you're going to use it to do work, you have to learn how to get good quality out of that interaction because it doesn't come by default, and it's actually really hard.

You're not just begging your AI to do something for you. You're a psychiatrist, or you have to manipulate... No, I'm serious, like robo-psychology, Asimov, right?

You have to manipulate what it's thinking to get it to do what you want. Can I give a concrete example just from this morning? Mm-hmm.

I have terrible travel schedule in November, which by the way, ends in Ireland. Mike, I got to talk to you about that. I have a terrible travel schedule.

So, I used Claude to put together a calendar for the month of November. I had an image of it in my mind, and I would say that if I had just done it myself, it would've taken about the same amount of time, maybe a little longer to produce that calendar. But in my prompting, in my interaction with Claude, I said, "Create a system so that I can do this every month that I want to.

" And I was really specific about how I wanted it to look that matched my thinking about it. And so the amount of time, I didn't just toss it off. I thought through it, talked through it, worked through it, spent all that time.

And also I had in mind this idea of the fact that, and I think my first sentence was, or second sentence was, I'm going to want to do this for other months, not just November. And that comes from experience. I'm not a genius or anything.

This just comes from experience and from paying attention to what kind of output I get. And also understanding that it's going to take my time and involvement and massaging to try and get it somewhere. But I think to bring it back to the token cost, right?

Because Guy made that effort and created a skill or whatever prompt to do what you want to do, then the next time he goes to do that, he's going to use far less tokens, right? Exactly. It even created a script- Yeah ...

so that tokens would not be used for sort of foundational usage. So this is what I mean by the skill gap. If you use these a lot, you learn those tricks.

Mm-hmm. I'm sure there's books you can read. I've tried to put together some in-person training things because this really is a new way of doing work and requires different methods.

So to your point then, are we going to create leaderboards that show how good people are at using AI cost effectively? Is that where we end up with this? I'm doing that right now at a company I'm advising because there's a huge difference between the people who can do this and the people who can't.

So if you're a company, you need to figure out who's spending all your tokens, like full stop. And that can be really hard, but because the companies make it very hard, right? Let's just be clear about that.

It could be easier. Right. But all this is to say, the token costs are only going to go up, right?

I can't imagine them really getting that much cheaper while we're using less, right? They'll just raise the price if we all start using less. I think we're stuck in this, and I think there's two things that are going to break this.

One is, if you think about dark fiber in the 2000s and the build-out of all the telecom, what changed was lasers. They had better lasers, and so they were able to do a lot more on the same infrastructure. And that could very well happen here, right?

There could be some advancement in the technology that makes this a lot cheaper to do. Then the question is, does that new capacity go into reducing cost or does it just go into providing more or serving more demand, right? Because that's just going to be the market dynamics.

I hope that's the way it works out, but I got a feeling that whatever magical innovation is going to show up isn't going to show up in 2027. So I think we should just assume that we're going to have some significant costs. So if you are a GPU gas guzzler out there with the big tail fins, you need a different approach because we don't have enough gas to drive around town with all these AI agents.

That's just where we are. Necessity is the mother of invention. I agree.

I try to stay away from clichés. I'm looking forward to that innovation, which will probably arrive for 2028. But I would plan for the worst now.

All right. I'm going to switch some gears, talk to Jamie here for a minute. Jamie, how are you?

I'm great. How are you guys? We're good.

It seems like every AI company has discovered biology all of a sudden, and they've got a bunch of acquisitions and they're doing a bunch of work in this space, but can you give us kind of a summary of what's going on here and what you're seeing from the- Yeah. Go for it. Yeah.

Frontier AI companies are going from providing scientists with AI tools for research, to becoming participants in the actual infrastructure and the process of biological research and drug development. So Anthropic bought Mythril Medicine, I don't know if I'm pronouncing that correct, which used Claude to help coordinate research over, I think they said it was like a network of 11 different labs. And OpenAI bought Orchestra Bio to bring AI agents into the coordination of drug development programs, like managing schedules and setbacks and monitoring studies and things like that.

So, and then at the same time, the UK, as part of a larger AI investment package, they are investing in AI biosecurity with a new AI biosecurity lab that they're spending 100 million pounds on. Because the same capabilities that are useful for research with AI are also creating new security risks. Mm-hmm.

So the big story here is that AI is becoming the infrastructure of biology. Or that's what the frontier labs are working on, so. So, on the upside of this, and let me ask this to Jamie, but we could cure all kinds of interesting diseases, and maybe all live longer, but what could possibly go wrong if we're allowing AI agents to mess around with the building blocks of life?

Well, my question is, there are so many companies out there, like Sandbox AQ is a good example, where they've built models specifically around chemistry, biology, and quantitative scientific data. So my question is, with OpenAI and Anthropic and the more generalized models, they're starting with these very broad reasoning models and then adding biological expertise. And so I just wonder if that approach is better than the more specialized models being built from the expertise or, it's kind of an additive subtractive thing.

And so I guess to your point is, I'm watching to see which approach might produce the best science, I guess. Mm-hmm. Well, I don't know.

Drew, you got any thoughts on that? " And there's a piece to be made that says, maybe we do need specialized models that might be, well, to our earlier point, smaller and more efficient to run. 100% agree.

There's something here, though, that at first read, you might not pick up on this. Mythril is not doing any kind of drug discovery. It is not actually developing medicine.

What it's doing is automating the process of doing drug discovery approvals. Mm-hmm. And so I don't want to minimize this.

This is actually a really big deal because, forever pharma has used regulatory blocking to keep independent companies out, and this could break that chokehold. That's actually really important. And make far fewer mistakes.

Because the way medicines are designed now is you're just sending emails to different labs and stuff. It's sort of weird, right? You could have a computer tell you what DNA sequence, and then send an order to a lab, and they'll ship you the rendered DNA like in a week or something.

So the science that they're doing and the automation that they're doing is they're automating the practice of the science of medicine, which is very cool. Mm-hmm. Now that's separate from what used to be called computational molecular biology.

And this is using models to create drugs and discover compounds and the shape of proteins and things like that. And that's also happening itself. So, I'm really bullish on AI medicine.

I think it's going to be a real help. We really need it. Agreed.

But I've got to say, just stop by this little show that's happening in a few weeks called Supercompute if you want to see how scientists use AI as opposed to AI companies do science. I'm trying to think of an example. Maybe desktop publishing, which back when I first entered the workforce as a temp once or twice, I was just a typist, and I would work alongside...

I went to a few publishing houses where you had actual typesetters putting together the books or whatever. It was a real specialized thing. But with WordPress or whatever, no it wasn't WordPress, but Word something, whatever we had to begin with.

Somebody hit me with some ancient word processing or layout. ADEX. ADL.

We were producing newsletters, posters, and all this stuff, right? But soon enough, the graphic design industry caught up and just smokes the rest of us, and you can tell what I put together versus what any of the designers I worked with put together. And it's the same kind of thing.

The sophistication and depth and experience of use of AI in science is long and deep. Nice. They've been doing it for a really long time, and using it for a lot of really helpful elements that you might not even think of, which is things like job scheduling and test design and development.

What were you going to say, Mike? Well, Jamie, is the end game here, are we all going to wind up with our own personal DNA-based remedy for whatever ails us? Because we're going to use AI to kind of fine-tune whatever medicinal thing that gets created here, and whether it's a vaccine or whatever, but it's all going to be very much tuned to my DNA versus Guy's DNA, which is fundamentally different.

I think that- I hope ... that question is beyond my expertise. But I can tell you that to Drew's point, this isn't about AI discovering a molecule.

These stories are more about AI taking that molecule through a decade of research and development. That's kind of the main issue, is what happens to the molecule once it's discovered, and how does it get into a medicine to help people? I think that's what these platforms are more about.

Jamie, what do you think Anthropic or OpenAI's game is here? I don't mean game in the sense of they're playing with it. Are these proofs of concept, then it sounds kind of like a marketing expense temporarily housed.

Or are they thinking of building businesses and spinning them off and making even more money? From what I can tell, both OpenAI and Anthropic have contributed, or they're saying they're going to contribute each 100 million pounds towards the AI biosecurity lab in the UK. And so that makes me wonder if they know that these platforms and drug development is going to accelerate drug development.

But it kind of feels like they're just throwing their money at things that... I don't know. Maybe- To me, they seem like investments on their part, right?

Yeah. For eventual payoff, right? They're going to pursue a magnificently large market.

And I'm all for AI providers attempting to automate complex logistics and handoffs to multi-year clinical pipelines, but I always have pause when it comes to Anthropic and OpenAI, in particular, and given the history of their behavior and them acknowledging the risks involved in what they're doing. Anthropic, you read their S1, they talk about cataclysmic events that include things like infrastructure, et cetera. Right.

So I always want to step back and look at that. I think this is a math issue at the end of the day for them, because they're not going to make the return that they need to make unless they go branch out into these various other verticals and kind of go drive some of these issues, and I think they're going to be reaching out into all kinds of vertical industry sectors that- Absolutely ... is going to give some people pause that already exist in those sectors because- Yes.

Nvidia's doing it already. Nvidia's been jumping into as many vertical markets as possible, and I'm not going to go down the path, but that's getting more and more scrutiny from the government. The government's looking into the whole case acquisition.

They're investing in their own customers. This is kind of the problem. Yeah.

Right. " Drew? Mike, I'm glad you brought up moat, because I think this has a lot to do with that.

And just take a step back and the Mythril company that they bought was one person. One person. One person with an MD file Right?

That's what we're talking about. And how much did they spend, what, a couple billion for that? This is the most highly paid program manager in the history of the world, right?

Because what he did is he automated program management using Claude. Right? And actually anyone can do that.

You just have to know what to do. This person must have had a lot of experience within these fields and understood the steps involved, and was able to kind of coax the right prompt to get it to do it correctly, right? That's an AI engineer skill.

But- And it will actually do it the same way twice in a row? Because you know- No, I don't think so. No, that's actually part of it, right?

This is a platform, right? AI is a platform, right? And so what we build out of it is going to vary, and we might not even want it to be the same, right?

There's places where you want the determinism and places where you don't. So this is part of that art we're going to figure it out, right? So are you saying that, I forgot what they were called, but there was this moment in time where everybody was creating this digital artwork with GIFs, and they were going to be valued very highly.

ESPs, yeah. Yeah. So are these AI workflows going to be similar where people are going to go, "Wow, that's extremely valuable," until they figure out that everybody else can express the same intent and create the same workflow?

I'm dying to find out. This is actually, no seriously, this is a very critical question in the industry. Are we talking about existing people being multiplied and serving way more productivity than they were before, or are we talking about people just not doing that job anymore?

And those are really different outcomes if you think about it. Going back to what something Guy said, and I think it was you said it, Guy, but I actually do think the end game for some of these companies, especially Anthropic and OpenAI, is to figure out, let's buy something, let's develop it, and maybe we can spin it off as another company. Maybe.

I think that's one out. That's definitely one out. Or they go over to Pfizer and say, "Hi.

" So, it kind of says- Their tentacles are extending throughout all these industries. Right. Yeah, but either way, it's just such another Silicon Valley fantasy.

Seriously. No, it has been. Anthropic is being built out this.

They hire scientists to do this, right, Jamie? They don't have goons like you and you, John, not you, Jamie. Goons like you and me putting a bunch of prompts in to come up with new molecules and new drugs.

They're hiring genuine scientists who are part of the scientific community and all that. But that's the lipstick on the pig, again, because that does not make Anthropic into a research institution. It doesn't.

Yeah. There needs to be validation still. Biological validation is still...

You can compress the information processing all you want, but there still needs to be a scientist in the room somewhere. So, can I have a couple of trillion dollars to create a biology lab that will be on Mars in five, six years or so? You guys just write me that check, and I'll get right on that, and then you- You can have a couple of trillion rubles.

You don't need to create the lab, you just need to write the MD file. So you get paid a couple trillion dollars for that. There you go.

Which will run on my digital twin, probably down the street instead of Mars, but what do I know? All right. I'm going to shift the gear to another obscure kind of topic, but here we go.

So, the folks who do research on behalf of the Defense Department, or depending on what party you're in, the War Department, are researching how to augment dogs. And a lot of this is, if you are aware of these, we use dogs in various military roles, and we also use dogs for things that are related to the police department and all kinds of interesting things that go along with that. And they're talking about really a combination of some usage of tech and gene splicing to come up with a dog that's a little more robust.

And this dog will be connected to AI technologies and goggles and have some sort of sensors that wrap around it that give it direction or let it capture interesting information. This all sounds well and good. So, Drew, is this the future for how we're going to work with animals in these roles?

And let me ask you this, what's the fine line between doing this with a animal friend and a human? That is a fine line. I would say this story gives me sharks with laser beams vibes.

It's definitely out there. But unfortunately, it's actually a supply chain story. Military uses war dogs for stuff, and they are hard to get, and I guess they feel like they can't get them fast enough.

So there's actually a big part of this is really about creating breeding stocks in the US that's something to do with national security. But it sounds a lot cooler like this, so let's go with it. So first of all, I love this topic.

I love dogs. I love the kind of amazing things they can do. Right?

I'll stop right there. Not too long ago in our military, there was an alternative strategy to using animals in war, and it was building robots that could walk on difficult terrain and carry heavy things, and you didn't have to feed them. There's pros and cons to that, too.

" And it's not clear that this would actually yield anything useful, like in any kind of timeframe that would beat the robot dogs, which seem to be working pretty well at this point. So it is a lot of fun. It's a really silly thing to spend our military budget on.

The people involved are a little bit sketch when you go behind the scenes, so I- Homeworld lucky, right? I'm not going to say anything there, but this is a wacky idea. I can remember a time, and I'll ask John, maybe he'll remember this with me, because I can't remember the name of the movie, but it had George C.

" Mike Nichols directed this thing. That was his first film. Mike Nichols?

No way. You could communicate with the dolphins. They could speak, and they could carry weapons, or they could defend against weapons.

There's going to be a war among the dolphins. They were going to be android types. But they were actually real.

It was totally bizarre, and no one believed it, and no one went to see the movie. So- But it's happening now. And can I just mention one thing?

So this is a Frank story. Is Frank bizarre? I think it is a Frank story.

Yeah, it's Frank. It was interesting. So this company, what's it called, War Hounds?

So they're going to have to, kind of going back to our last segment, they're going to have to prove that gene editing and cybernetic augments can outperform decades of proven select breeding. So I think about the bioethical scrutiny and military operational testing. I think about those hurdles.

I mean, it's cool. Maybe it happens eventually. I'm not against progress, but some of the stuff when we first hear it is so out there.

I don't know. I mean, think about it. You could create an entirely new breed of dog, right?

We have seen new dog breeds come along over the years by combining different strains of different DNA, and they create some other new type of breed. But I don't know. Jamie, are we looking at we can create custom dogs here?

To my knowledge, genetics doesn't guarantee a better dog. Cloning one dog that's really successful, it doesn't clone its training or its experiences or behavior. So I don't know.

I'm skeptical. I feel like performance traits like sensibility and temperament and that kind of stuff, that's they're more complicated- I don't know. I'm skeptical.

So if this were to be successful, say, with war dogs, do you think this could lead to, like Mike alluded to, the super soldier, like the human cloning type of- Like a Captain America dog or something? Exactly. Yeah.

Yeah. Yeah. I don't know enough about genetics with dogs, but I know that there's a lot of genetic things happening in backyard breeders, and dog shows and things like that.

So maybe with the war, maybe war dogs, it'll translate. I think that's a more serious question than you may realize because there's an old saying that says when it comes to tech, if you can think about it, somebody's trying it. And I can bet you that somebody probably in places like China and Russia is playing around with this idea and probably seeing how far it might go.

I don't know. Guy, do you think that this is going to happen? Well, I do feel like we're in a timeline where the modern equivalent of dudes in basements with snacks, referencing their entire catalog of graphic novels, comic books, and grade B action movies- ...

coming up with all the new ideas. But here's what I'm wondering. We've been doing genetic engineering, humanity has been doing genetic engineering for thousands of years.

Cross-breeding, selection, that sort of thing, including with dogs. And that's a tried-and-true method. I'm wondering if those methods are being used or if it's more like this sci-fi fantasy concept of splice these genes into those genes and bibbity boom, we have a super cyber dog.

Because I think that seems likely that that's the approach. If I heard in the same breath with these breathless stories that they are going through a multi-generational breeding process, and it might take a decade or two or something, if not longer, then I could see dogs being bred to adapt to whatever cyber part there is to this. But that's not the sense I get.

The sense I get is this is like every tech startup, which has got some cool idea that's funded and they are pursuing, and it relies on a whole lot of experimental or not even tried and tested methods, unlike the thousands of years of crossbreeding that we are familiar with. In that case, it's just going to fail. It's going to fail over and over again.

John, what do you think the odds are that there is a multi-billionaire, maybe even a trillionaire at this point, that already has a large family and a large number of kids who might get another wife and try to do something that looks like- Yeah. He might be thinking about this as personal security detail. He has the guard dogs that are created in a lab, but then he's going to have his army of security detail around him.

He probably can- Can I just say that that person you're mentioning, Mike, is the modern equivalent of the dude in the basement with snacks, so anyway. Among other things that he consumes, yeah. He just also has a few trillion or billion, whatever.

Go ahead, John, sorry. Oh, no. I'm sorry.

No, I think definitely, he probably is thinking about it. That might could be a new venture, right? Sure he has.

Beard's guys already worked on it, right? They just got the robot that runs around that's quite impressive. So, we take it several steps along the way and there you go.

All right. Drew, I have to- I have one disturbing kind of direction where this could lead. And this is a bit of an ethical conundrum, but I imagine one of the things that they're aiming to improve with these dogs is their intelligence.

And at some point, that may be sentience. And then what do we do with those- Then where are we? dog people?

Do they vote? Whoa, right? It starts to go nuts.

What do you think about the future Westminster Dog Shows? They could take on a whole new category. They could be their own judges.

What are you going to do when the dog says the Kindle sucks? I think a realistic sort of issue with this whole thing is that soldiers don't like seeing their dogs die. And this has to do with a millennia of evolution that humans also adapted to by living with dogs.

And so I don't think we're bringing armies of dogs into the battlefield. Hopefully, it'll be just like rescues and stuff. Sure.

Mm-hmm. Well, whatever it is, we're going down this road, and I imagine that some governments will probably step in here and try to regulate some of this at some point. But that which was science fiction a thousand years ago is now somebody's lab project, so who knows how well this is going to play out.

But stay tuned, and if you see somebody walking down the street who looks a lot like you, it might not be an accident. Just saying. It may be you.

They live. You might be me. There you go.

All right. I want to thank John, Drew, Jamie, and Guy for sharing their expertise, as always. I want to thank you all for spending time watching the Techstrong gang with us today, and please stay tuned for the rest of the Techstrong TV lineup.

And of course, we have a word from our favorite sponsor, which is us, so please stay tuned for that, and we'll see you all tomorrow.