AI’s Closing Window: Tariffs, Cyber Defense, and Killer Drones

Hello, everybody, and welcome to Friday. I would normally say, well, happy Friday, but my Fridays are all the same. " So here we are, and we got all this stuff going on, and well, there's just a ton of stuff out there, but let me introduce our attendees today.

John Schwartz joins us once again. How you doing, buddy? I'm doing well.

I'm glad it's Friday, but you're right, Mike. I'm also dreading it, because they're going to be a news dump probably in about two hours. There you have it.

There you have it. Usually bad news starts Friday. Always.

Always. Yes. All right.

" Jack, how you doing? I'm doing good, and at this point, I think I want to make Claude bleed a little bit. I've been fighting with Claude all week.

Oh, man. Yeah, that's just a losing battle. You just wipe the memory and start over.

Exactly. Drew, good to see you again. You're becoming a regular here on the end of the week show, but welcome as always.

Yeah, great to be here. Looking forward to the conversation as always. All right.

Tracy Reagan, good to see you as always. How are things? Good, and I'm always happy to be on the end of the week show.

All right. It's a good way to start my weekend. And Jeff Rich, how you doing, my friend?

Doing well, thanks. If you're having a bad week or bad day, it's National Cherry Turnover Day. Go do that.

Go do that. And of course, apparently there's some libations you can have with that, correct? Oh, and it's also National Red Wine Day, so you should do both.

Just don't wear a white shirt when you're doing it. Exactly. That's good advice.

All right. Well, we are chock-full of security related issues, and so we're just going to dive in. But the Trump administration is threatening to levy some additional semiconductor tariffs, and apparently what they're trying to do, in the name of national security at least, and maybe manufacturing, is they want more of these semiconductors to be made here in the US.

So, John, walk us through what they're up to. So they are looking at expanding semiconductor tariffs to bring back manufacturing to the US, which as we know, is an obsession, and it's a goal. But it's raising alarms among folks in the AI industry, obviously, about AI growth and if the supply chains get disrupted in the process.

So Politico first reported on this, and the idea is to pressure tech giants into shifting their manufacturing back to the US, but it's going to come at a cost, literally a cost to consumers, because if this were to happen, the reasoning goes, or the argument goes, this price is going to be passed on and extend to laptops, gaming hardware, and most importantly, data center servers. The details are sketchy, but evidently, Commerce Secretary Howard Lutnick favors this phased-in approach that ties tariff relief to a foreign company's investment in domestic chip manufacturing. So, we will see how this pans out, but it's interesting that this is happening at the same time we're debating or talking, honestly, almost every day, about data center expansion and scaling back of some enterprises because of costs.

And it's kind of this collision course that we're heading towards, and the fallout, unfortunately, it always happens, it always seems to be that the consumer probably is going to pay in the end. So, let me ask Jack this. How legitimate a concern is this from a national security perspective?

Do we really need all these chips to be made here in the US, or is this, as some folks would say, yet another tariff shakedown? Well, I think the answer is, I believe that it is important that we have this capacity in the United States. Currently, most of our advanced chip manufacturing takes place in Taiwan through TSMC, and in Korea, South Korea I should say.

Now Taiwan is under continuous threat from mainland China and the People's Republic Communist Party of China, and that could have a very big impact. Now, if China either militarily or economically gets control of Taiwan, that would be a big impact on the United States and on the United States military and on their economy. So I think it's important that we do expand our manufacturing capabilities here in the United States.

Unfortunately, I don't necessarily agree with either the way we're going about incentivizing companies to do so, and the timeline that people are thinking about this, right? I was in California for 25 years. You can throw a tilt-up warehouse, build it in three months.

You go from scraping the ground clean to a building that people are occupying. It's amazing how quickly warehouses go up. Mm-hmm.

A semiconductor fab is not a warehouse, and it takes a considerably longer time and a hell of a lot more money to build a fab and get it up and running, and to producing chips in volume. We're talking years to a decade to be to the point where you're producing advanced chips in volume, and tariffs aren't going to change that speed one little bit. So I don't understand what's the motivation behind using taxes or tariffs or any of the other things as a incentive to hurry up the equation.

I don't know, Drew, but isn't it also fair to say that if I'm taking more money out of the ecosystem, it's going to take longer to build the new manufacturing facilities that probably aren't going to be available for five years anyway, because that was the money I was using to fund those things, right? Yeah. The timeline doesn't make any sense with respect to tariffs.

It's just not a proper incentive for this kind of activity. There is one, though, in the CHIPS Act that was passed several years back that provided incentives for companies to build data centers in the US. And there are projects ongoing that have started and they're part of the way through building it out.

This could definitely disrupt that, and make it take longer perhaps, or increase the cost because the equipment they need to import is now tariffed. I think this administration loves to create tariffs, but what they love even more is creating exemptions to those tariffs. There's always a carve out.

You're right. Yeah. I think there's maybe something of that going on here, because it doesn't really make any logical sense.

All right. Tracy, under the heading of maybe perhaps shooting one's toe off, if the cost of the processors goes up and people are already complaining about the fact that AI is getting expensive, won't that just take more money away from those AI projects, and there'll be less money for software for that matter because the underlying hardware will get more expensive, and then we'll have fewer projects, and it becomes a somewhat self-defeating proposition to begin with? Absolutely.

And it's not just AI becoming more expensive, it's everything becoming more expensive, right? It's already happening. So, let's just say that when we think about how tariffs could raise the cost of GPUs and servers and other AI infrastructure, as well as other products that use chips from medical devices to appliances in your car, the problem is that trying to restore tomorrow's supply chain by making today's AI build out more expensive could be devastating to our position as an AI leader because it will slow us down.

It definitely will slow us down. And I find this to be kind of ironic, because as Drew pointed out, the Biden's CHIP Act brought, I don't know, it was like $50 billion in incentives, and we do have now manufacturing plants starting to be built, and this is actually starting to happen. But the Republicans hated it because they talked about it being subsidies and government picking winners.

And the carve outs are exactly that, picking winners. And we know that that will happen. So, this is an annoying announcement, and if it's true, it will be impactful to the overall economy, and it will be impactful to our position as AI leaders.

There's no doubt in my mind. Literally, you can't reshore a semiconductor supply chain overnight, but you can make AI data centers and technology at least 15% more expensive overnight. And that's what these tariffs do.

So when that happens, the cost of everything goes up, and we're already looking at very expensive data centers that's causing everything to explode. And we need these chips to be manufactured right here in the US. So if Trump has decided that he wants to repeal the CHIPS Act and move forward in a tariff model, it will stop the manufacturing plants that we already have.

I believe that. All right. Well, Jeff, let me ask you this.

If you're sitting in Taiwan and you're watching all of this, what would be the implications? Because I think what the president is trying to say here is that we are going to manufacture this stuff here in the good old USA, and once we have that capability, it's going to be, "Good luck, Charlie. " Right?

Well, yeah. There's a couple different vectors at play here. If I'm in Taiwan manufacturing chips, first of all, Tracy said it.

There was one word she used that I think is key if this happens, and this could simply be yet another round of tariff whiplash. All right? First of all.

So, let's see, because we're at the point now where anything that's announced like this still takes time to, is it even going to happen? And if it is, is it different, and who's going to be exempted, and everything else. But if I'm in Taiwan making chips, what I'm thinking is they don't have the capabilities, they don't have the knowledge skillsets people.

I'm going to start putting people in the US to do this. And I'm going to continue my revenue stream, deal with this. It'll probably get me an exemption.

Life's good. Now I know I need to go deal with whatever Beijing's going to do to me as well. But I think that's the issue, and we need to recognize it is important to have these manufactured in an environment we can trust, which we're saying is the US, and I agree with that.

But just because it's being done in the US doesn't mean that we have to give it blind trust. Where did the knowledge come from? Who's working on it?

What's the ultimate supply chain for it? So I think it's more complex. If you're flying a plane and you want to go to a higher altitude, you don't just pull back on the stick.

You'll stall and crash if you do that. You have to increase your speed. We need better incentives.

Well, I believe everybody who's involved in this in Taiwan will have the option at least of buying their way into a US citizenship program, right? Mm-hmm. That's probably what's going to happen.

John, as you kind of look at all of this, the cost of living will go up if this gets instituted, and midterms are not that far away. Yep. You read my mind.

I'm thinking about the timeline here. It couldn't be worse, right? So Apple's going to announce their new phones in a couple of weeks Usually it's the second Tuesday in September.

Their prices are already going up. We're going to see gaming console prices go up. At the same time, there's this huge resistance and pushback against data centers.

Here's another thing to lob on top of that. And also, as I think Jack or everyone else has mentioned, the timeline is nonsensical because by the time any of this is achieved, Trump's going to be out of office. So in the short term, we're going to get gouged, and there's going to be even more resentment, and there's going to be another talking point to hurt Trump and his followers.

I just don't understand it. Hopefully, it's a trial balloon, but I don't think it is. I think he's so obsessed with tariffs that he's not going to let go of this one.

No. And some of those new data or manufacturers are in states like Indiana, right? Mm-hmm.

I think there was a company that just got spending like $4 billion on chip manufacturing in Indiana. And that's coming from some of the CHIPS Act. If that money gets pulled out, he's going to have problems right in his MAGA backyard.

That's a great point. You think about Louisiana, Texas, Indiana, Ohio Idaho ... down the line.

Yeah. Yeah. Because the CHIPS Act did put money into building these plants, and much of it went into red states.

Yeah. Yeah. So, one last question on this, and I'll throw this to Jack.

Jack, will consumers, and corporations for that matter, just start buying up as much gear as they can now at a lower cost because then they'll store it somewhere and hoard it because they know that the cost is going up, and we'll see this kind of spike in activity, but we'll trail off at the back end? But I believe that's what happens when tariffs are... Have you looked at the price of memory?

Regardless of tariffs, I think people are doing that already. We went through this with disk drives a couple of years ago. We're doing it with memory now.

And it really depends on what you consider perishable or not, right? These days, for instance, the modern processors are so powerful, they're more than sufficient, and they last five to eight years without a problem. So you can buy now and not worry about it for another five years.

If you're looking at your server processor for doing AI processing, you're always going to want the latest and greatest and the fastest. And there is so much cash flying around that you're willing to eat the increased costs to get the latest and the fastest thing. So it really depends.

But yeah, I'm sure there are a lot of companies right now doing lifetime buys or, Jeff brought up airplanes, we used to call it bunkering fuel, right? When you fly, you would fly to where the fuel is cheapest, and you load the plane up where fuel is cheapest, even though it costs you more to fly that fuel around because in the end, the math works out. And I think for some organizations, the math will work out to do pre-buys and big buys now and get both discounts.

And for some orgs, it won't. So, I hate to use the analyst's favorite phrase, but it all depends. Is that like my cousin who drives 30 miles to get cheaper gas?

Is that the thing? When I had a diesel pickup that had a very, very big and very expensive fuel, 25 cents a gallon makes a big difference. And you're willing to go out of your way when you're loading 60 gallons of gas.

When it's 10 cents a gallon difference and you need 10 gallons, who cares about the dollar? Right? It depends, right?

But yeah, some people are going to do that because it makes sense economically for them. All right, Drew. And Jack makes a point on the type of chip that we are in need of right now, and maybe it's not even developed yet.

Maybe next year it's going to be the chip that we need. In particular, Jeff and I on these calls, we've been talking about distributed LLMs. Mm-hmm.

And my guess is if we really address the problem with these LLMs and hyperscalers and data centers, it's going to be distributed back down. It feels like we're going back to the mainframe in a way, and we're going to go the other direction, which means that everybody's going to want better, faster chips on their home machines. And so we could potentially change the way we're doing AI but not have the chips to do it.

There you go. So we've got to make a decision here. We've got to figure out how to do this the best way.

Right. Data centers, they're going to be on the ballot in November, there's no doubt. So we're in a weird place to have this come up.

All right. We've got to move on here because, well, there's just a lot to talk about. But yeah, this is just going to have all kinds of fun cascading effects going from now till well in the next year, I think.

So, if you are planning on buying any kind of consumer electronics for Christmas, now is the time to go shopping. Right. You heard it here first.

Next topic. There was an open letter sent out from the folks who build these AI models, the frontier models, basically saying that the window to defend ourselves from these AI-powered cyber attacks that are coming our way is closing. And, I'm of two minds with this thing, and I'm going to go to Drew first on this.

Is this some sort of cynical marketing ploy to get everybody to buy their services, or is this a real thing? And I ask whether it's a real thing because I was talking to Palo Alto Networks about this stuff this week- And they pointed out that while, yes, we are seeing more of these AI-enabled malware, the malware itself is not novel and the existing security frameworks are adequately defending us against them. So what's your take at what's going on here?

So look, I think it's all of the above. The only thing that can stop AI is more AI, is sort of mutually assured destruction logic. And it definitely serves the purposes of AI model creators.

But it's probably not wrong. I think there is a window where we can now start using AI to fix our systems before the AI can break them again. But it's a crazy letter.

I mean, we're going to create products that are going to break your whole system. Please buy more of it so we won't break you, is kind of the message. It's nuts.

The other thing, I think this kind of, something you had said earlier was machine speed. What does that mean? What is machine speed?

Everyone's talking about how AI is so fast, but AI's really, really slow, right? The tools that go fast are the existing toolkits, scripts, and platforms for causing malicious events. This is a sort of offline thing, this war, right?

It's happening in the development environment. It's not happening in real time, in a way. So, yeah, I think they need it.

I think we all need it. And I think there's another thing in there that was a little bit interesting as well, is this concept of when do we get to that stage that I think Tracy was referring to of we don't need an AI in a data center anymore. We can run it locally or distributed.

And there was a bit in the article that talked about NVIDIA looking to buy Hugging Face. I thought that was really fascinating. NVIDIA will make money, whether it's in the data center or in your office.

Seems like a logical thing. But this could be part of that prep, right? There's going to be a, I don't know, a phase change or something where a lot of work will be done by models we run locally.

" Yeah. Well, Jeff, I'm reminded of one of the "Godfather" movies. The guy shows up, says, "Yeah, nice store you got here.

" Mm-hmm. This letter feels like that. Oh, it feels a bit like a protection racket, yes.

And I'm guessing it was written by AI- ... for what it's worth. I'm trained on Mario Puzo novels.

Yes. Just because the bad guys are getting more guns doesn't mean you need more as well necessarily. I'm not saying pro or anti-gun, but whatever the weapon is, and AI happens to be a tool or a weapon, depending on which end is pointed towards you.

So, I actually had a conversation with someone this morning saying they really want to put in a security program. What tools should I buy? And I said, and I think it applies here, "That's not your question.

Your question is, what do you want to accomplish? What do you have that's really valuable to you? " And that's being missed.

We can't simply say buy more AI to protect against more AI attacks, because all we're going to do then is open up a vulnerability for a low and slow traditional attack that's going to go in unnoticed, because you're so focused on AI and spending all your money there. So I really think this is an introduction to a protection scheme to say, "Buy more of what we have, or what you have is going to go away," rather than saying focus on the real security issue. I think the tell there is in sort of the transparency, right?

So I've been trying for over a year now to get real solid numbers on attack frequency and success rates and impact, so we could actually model the threat like we do all our other threats. But the AI companies are keeping this very close to their chest. They're not informing the larger security community of what attacks are successful and what is a good control to stop.

Because the guardrail systems that they're pushing are just not really fit for purpose. They just will not provide the level of protection that we need. Right.

Tracy, there is clearly more vulnerabilities than ever that should be remediated, so we should take that part of the equation seriously, and we need to pay down that debt. But I have to wonder, to what degree are the security companies actually relevant in this conversation? Because it's the app dev team that's got to go out and fix all that stuff and write the code and put the patches in place, and the security people are there cheering them on, pointing to a bunch of CVE scores that are kind of irrelevant anyway.

So, what's going on? So I don't think that the AI cyber race isn't really about attackers versus defenders. It's about automation versus traditional DevOps processes.

Mm-hmm. So if attackers operate, they use AI to find these holes, and developers are still working through vulnerability tickets manually and then releasing them as standard bug fixes, the attackers win. And this is my big beef with that letter.

It doesn't matter about knowing that there's a vulnerability. We already know there are many vulnerabilities out there. AI changes the question from can we find them before release, to can we fix them once they're found as fast as possible?

Using AI to just produce another 100,000 findings doesn't solve the problem. It doesn't matter. We're never going to be able to get through the list of vulnerabilities we have We need to disrupt.

Now, we can use AI, and I totally believe in this, to do more work around reachability and more work around de-bloating. But what developers and teams need to know today is not the vulnerabilities that are hitting their code repos. Because they're doing a pretty good job on finding those.

What they really need is the ability to figure out what's hitting their production systems today that wasn't announced yesterday. Mm. That is what we have to be able to do.

We need to be able to defend our operational endpoints and focus on doing that as fast as possible. None of what we heard from these AI companies talked about disrupting the way we manage vulnerabilities. And there are going to be hundreds of thousands of vulnerabilities that we find over the course of an-- We'll probably be at 100,000 this year.

I think we're at 49,000 already by July, something like that, the last time I looked. Mm. Developers can't fix that many, and they shouldn't even be worried about fixing that many.

They should be worried about fixing the one that are exposing endpoints, and we don't do that. And Palo Alto doesn't do that for everybody because it's expensive, and not every system is going to have Palo Alto running on it. " But not everybody is going to use something like Palo Alto Networks or Aqua or Rapid7 to do that.

So we have to disrupt it. And I get so frustrated when we talk about finding more vulnerabilities, and we're going to find more vulnerabilities for you. Who freaking cares?

It's not about the number, it's not about finding them, it's about knowing what's being exploited on my endpoint today. Mm-hmm. And that's where the discussion has to go.

And hopefully getting to some sort of fix. Jack, the relationship between cybersecurity people and application development teams, let's just say there hasn't been a lot of love lost over the years. Yeah.

The developers get these lists of vulnerabilities from the security people. They go chase them down. They find out they're not running in memory.

They're not reachable. There's 100 reasons why it may not be relevant. And to be honest, though, the security people do have issues, and there are issues with applications, and often the security people view the developers as, well, the root cause of all their troubles.

So, how can that relationship evolve as we get into this AI era and maybe get better? Well, it's developers, but I've said this before in these discussions. We had this discussion when Anthropic announced Mythos and yada, yada, yada.

And there are two sides to the equation. There's the producers of software and the consumers of software. And AI is really great at finding bugs, feeding it to the producers, and they fix the bug.

Right? Now, what happens next? That's now a patch that gets propped in IT's hands, right?

" And they would tell you, "I've got anywhere between 1,000 and a million things that I have to patch on a daily basis. " Right? So just as an example, Microsoft's Patch Tuesday over the last three months went from 20 or 30 patches they issued in one cycle to 200 the next month, to 600 the following month.

Right? Now, that's just what lands on your Windows PC. You do an automatic update.

It's great. It's not a big interruption. If PC has to do a reboot, it's not a big interruption.

But now apply that to all the servers that are running out there and say, "Okay, now I just got listed 600 patches on a server. What is it going to do when I apply those to all this old, crusty software that's running underneath, on top of this server that's just got patched? " So people are very hesitant to apply patches.

They get bigger and bigger backlogs, and that's part of what Tracy was alluding to is it doesn't matter how many more we find, we don't have an ability to manage it today. Yeah. Now, I'm going to rant for a little bit because this is the same thing that Anthropic did, or, sorry, OpenAI in, quote, the Hugging Face incident, right?

" Well, I don't know about you, but I can create an isolated sandbox very easily with a pair of wire cutters and cut the ethernet cable, right? It's not that hard. Right?

It just isn't. We can do the same thing, both physically and logically, to create isolated networks that have zero reachability and isolate, and microsegment and isolate these tools. There's lots of techniques to do this, and it's not that hard.

Now, if you are operating a critical infrastructure thing like a water plant, which has been in the news recently, and I think the letter mentioned, right, it's a warning to critical infrastructure operators. Hey, do you really need internet connectivity to the outside world from your water plant? How important is that?

Yes, your guys can have remote access. They don't have to show up on site. Make them drive in.

It's safer, right? I'm sorry, but it's safer. There are well-understood things to protect ourselves that we seem to ignore because they don't involve AI.

We don't necessarily need to involve AI to solve part of the problem. Right. Hey, Mike.

Rant over. Oh, sorry, Jack. This preoccupation that OpenAI and Anthropic have had of late about security, I just think of it cynically in terms of fear sells And both of these companies are gearing up for their IPOs, so we're going to hear a lot of this.

I think everything that they talk about has that in mind in the back. And it's not subtle. It's going to become more overt.

But it is part of a marketing narrative that they're going through now in terms of responsible AI use and trying to put themselves in that position, which, good luck to them. I don't disagree with the premise of what they're saying, but I think the intent is looking closely at IPOs, basically. Just like NVIDIA's intent now is to steer us away from thoughts of antitrust.

Right. All right. I thought Tracy and I were the designated cynics for the program.

Oh, no. Well, yeah. I looked through things like gray-colored glasses sometimes.

Yeah, I think it has spread because they're using FUD, John. Yep. And FUD doesn't resell, but FUD does sell.

And the other, there's a point with NVIDIA, if I may, that hasn't been mentioned here. The CEO of NVIDIA made a statement that AI runs on debt, not tech. Hmm.

This is true. And there's a whole lot of that behind this as well that we can't ignore. I think what we learned from all of this is the most important thing about AI is to know where your cables are and where they're connected to.

That's really not cool. And have a good pair of clippers. I think what Jack was alluding to earlier around isolation is so important, and something that a lot of companies haven't really invested in, but can be done.

And I think, Tracy, you mentioned before, the patching controls are just so difficult, right? So we have this thing where it's not really our fault that there's a vulnerability, it's just a function of software, right? The more software you have, the more vulnerabilities.

But then there's this really expensive process for alleviating that. And I think this is a space where the controls are not really mature enough yet. Right?

There are ways to prevent attacks without having to deploy. You can use different kinds of sandboxing things. We can put it in the hands of operational teams.

They don't have to necessarily push a new software update, right? So there are some tools that we don't have yet that I think are really needed for this problem. Right.

And habits. It's our habits, too. We have habits on the development side.

" But developers don't want to do that. I wouldn't want to do that. I don't want them to just assume they're all, because the newest one could have worse vulnerabilities than the older one.

You don't know. Mike, we're not going to let you do it, Mike, because I want to add on one more thing to it. Sorry.

All of these water systems that are being warned, ask them how long they've been dealing with their ICS devices that haven't been patched or are vulnerable. How many years has that been? Why are you worrying about this now?

All right. We do have to jump, but I will point out one thing. Somebody told me the other day we spent all these last few years telling everybody how important it was to stay current on their patches.

" But there you have it. All right. Let's shift over to there was some interesting news out of Ukraine.

I guess they were poking through some drone debris and lo and behold, they found some processors that came from NVIDIA. GPUs, AI-driven drone attacks. And then, there was a separate story where basically Ukrainians were showing off their software systems for managing drones and the stuff that they use, and it looked pretty advanced.

So hats off to them for that. But Jack, what's your take on what's going on here? How come we've got some GPUs that nominally belong to an American company that were made in Taiwan showing up in a battlefield in Ukraine?

I'm shocked. Shocked, I say, to find that there's AI in our drone systems. Yeah.

This is the natural evolution of a lot of research that has been done over the last 20-plus years that's now being accelerated on the battlefield. We've been looking at AI swarm technologies. Years ago, they called them buddy systems, where you would have a pilot in a fighter aircraft would be followed by a swarm of unmanned aircraft that are effectively disposable but could attack something en masse.

And that's research that has been done since the '90s, early 2000s. There's been a lot of stuff being done like that. Now with drone technology and compute technology, yeah, we can apply AI to it.

So the Ukrainians found a downed Russian drone and inspected it and it had an NVIDIA Jetson processor in it, so they assume. It didn't have encrypted code in it, so they were able to look at the code in the systems and see what it was doing, and it was doing some rudimentary target identification and navigation. Essentially, drones use AI for semi-autonomous or autonomous navigation for a long time, and we wouldn't recognize it as AI today considering what we call AI.

But a cruise missile from the 1980s to 2000s, a US-based cruise missile, had a lot of processing capability in it to do a map of the Earth's terrain following. So this has been around for a long time. It's just like everything else, getting accelerated, moving faster.

Drew, I thought that these GPUs were hard to find and they were scarce. So how is it that these Russians are able to find some and stick them in a bunch of drones? Because the IT guy down the road here can't get any.

Yeah. I don't know where they're sourcing their chips, but it definitely is not surprising that this is developing. And I think having this after the last segment about AI's going rogue is sort of the kicker.

We spent so much time developing precision targeting or precision strikes and munitions that can go down a chimney or whatever, but we're not doing precision target assessment. " It's really terrible. Where are the three laws?

Why don't we have Asimov's three laws of robotics here? I don't know. Because all of these conversations, there's never even a mention of some kind of circuit breaker or something like that.

It's pretty scary. One thing that I thought was really interesting about the Delta AI that Ukrainians are deploying is what's the old saying? Armies always march on their stomach.

Mm-hmm. We're now in a phase where armies march on their data. So the fact that they can arrive with a data center and have it operate in the field without needing to call back home, that's a huge capability.

I think that's something that is going to make a big difference in especially the drone battles to come. John, what do you imagine that call from President Zelensky to Jensen Huang sounded like? For all we know, Jensen was probably there on stage talking to somebody and advocating the customers they've won over, and the NVIDIA dream.

Yeah. " They're everywhere and everything, and I don't mean to be an alarmist, but I'm starting to think that they're becoming much more powerful, more quickly than even they anticipated. Where in fact, Jensen is like a world leader.

He is on a par. He's probably more powerful than Zelensky- Right ... honestly.

Right? He's probably one of the five most powerful people in the world. So you've touched on it, and here's the thing.

GPUs are munitions, and we have no control over the supply chain of said munition. This is frigging insane. Jeff, am I crazy?

Well, those are two separate questions. But I'll support you here, Mike. It is insane, and I have to play this sound clip because I think it applies for those that were around when this movie was out.

Shall we play a game? " It did not turn out well. Of course, there was a happy ending, kind of.

How close are we to this? How close are we to having an autonomous decision made to say it's time to deploy a weapon? In my opinion, we're there.

If we have to ask the question, we're there. But the bigger question is, who's controlling that, and do they have any guardrails? And John, I think you're right.

The technology is really driving a lot of those political and military decisions right now, and when does it get out of their hands? Remember even back to the military industrial complex with Eisenhower? Remember how we didn't want any soldiers on the ground, we wanted to do everything from the air, air superiority, et cetera?

It just keeps escalating and escalating, where there will be no humans involved in the fighting. And listen, we're going to have these AI drones, AI technology, munitions. Maybe NVIDIA becomes like an arms dealer in a weird way.

Yeah. They don't want to admit to it, but in a sense, whether they like it or not, this is a first indication that something like this could happen. So the only human interaction is casualties.

Wasn't that a point of- None of us should be surprised ... wasn't that a point of contention, though, between Anthropic and the Department of War? Right?

Yes, it was. Right. Yes, it was.

Yeah. Now I think they've kissed and made up, but yeah, this whole thing is just escalating, and every time the technology gets better and better. We went from these land wars in Korea and Vietnam, so to speak, to this fear of thermonuclear war, to the point of people are not even involved.

I will throw this out. I'm going to take this too far perhaps, but there's a story. "Wired" talks about OpenAI and this persistent mode AI that never sleeps.

And the idea of that happening in a situation like this, in a warfare situation, is terrifying to me. All right. So let's walk that through a little bit, and Jack, help me here.

We're going to have drones, swarms of drones, being launched, and we'll have swarms of drones fighting against them, but eventually, some swarm is going to get through some defense somewhere, and those drones are, in effect, going to carpet bomb civilians? Is that where this is all going to go? What's going on here?

That's where it is right now. So take AI out of the equation for a minute. Right now, when you have a hot war between Ukraine and Russia, and both sides are using drones.

Now, Russia hit a shopping mall recently in Ukraine, and with drones, and I think it killed five, seven, some small number of people, but any loss of life is not to be minimized. Ukraine is targeting Russian civilian facilities. Ukraine took out the vast majority of the Wildberries warehouses in Russia.

Wildberries is essentially the Russian equivalent to Amazon. And destroyed almost all of those. One or two drones with incendiary devices torches a warehouse.

It goes up in flames very quickly and very prettily. A lot of nice pictures about things on fire. So it's happening today with or without AI.

And you can compare the Ukraine environment to the United States versus Iran, where the US has 100% total air superiority, whereas in Russia and Ukraine, neither side has air superiority and isn't able to take out the drones of the other side 100% of the time. They get through, and we have these things. Now, will AI make it better?

It's not the wrong word. Will AI change the equation? Probably, but since both sides are using it, it's yet another in a many centuries long history of weapons leapfrogging each other in capability.

When you go from sticks to guns, to artillery, to airplanes and tanks, et cetera. We've had always been this leapfrogging environment in the art of war, and I don't see that stopping any point in time. Well, we got to end the show, but I sure hope that something gets done here one way or the other, because no matter which side is doing it, no matter whether they're good guys or bad guys, last time I checked, carpet bombing civilians is a war crime.

Yep. " And maybe we should make sure that those semiconductors are not being used for that particular purpose. But hey- And before you stop, I just have to get my two cents in on this topic.

Okay. Well, first of all, there should be no surprise that we found an NVIDIA chip on a Russian drone. No surprise.

The foreign adversary laws really don't apply to certain companies. They just don't. Look at SpaceX.

Okay. The second thing is we're spending most of our money on Golden Dome right now. The US is not putting that much money into drone technology, which is a problem, because what we're doing right now is spending a lot of money on what we believe, what maybe the Trump administration believes that war looks like, using advanced giant missiles as opposed to drones.

So it feels like we're kind of stuck in the past, and the US is going to get left behind because we're not putting that much time and effort into understanding and how to use drones, how to defend against drones, and Golden Dome won't. So there you go. All right.

I definitely have to end this here, but it's pretty apparent that I have no control over this conversation anymore. I thought that was the whole point. I want to thank you all for sharing your thoughts.

I want to thank everybody for watching the show, and I hope we'll see you Monday, and have a great and safe weekend. Take care.