Cyber defense will force governments to allocate GPUs

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 18:45

   I mean, look, I think we are, we always assumed in the sci-fi novels that we would be in an era of having robot wars. And I think we failed to recognize that the software came first. And we are actually having the AI wars. and there's a war that I think we're in the middle of right now where there's kind of one side that wants to centralize, control, manage, regulate, whatever the term of the day might be, the technology. And there's another side that I think is cognizant of the reality of the technology, which is that it is effectively ubiquitous. We talked about this in the past, due to open source open weight models proliferating, the fact that you can run models on small, local machines, that you don't need large clusters to do things. And that we're seeing model after model come out now, that are not under the jurisdiction of some centralized system or

2. 19:47

   agency, therefore, it becomes very hard and practically impossible to actually regulate the development of the software. That's the truth. It is what it is, you may not like it. but it is what it is. And those two sides are going to continue to fight. I think that the risks will rise. And as that happens, I'm going to make two predictions to your question. The first one is that I think we will see a massive rise in defense systems.

3. 20:17

   There is no turning off the open source, open weight models, or shutting down all the data centers are centralizing command and control. There are 197 countries on earth. They each have sovereignty. There's energy everywhere. They're all connected to the internet. Data centers will exist all over the world. So the attempt to try and centralize and manage the control of the software, I think is a fool's errand. But I do think that it's going to mandate because the risks will rise an increased amount of defense. And so I think, you know, companies like Palo Alto So networks, I think the investment in this space, what I heard in DC this week was some Fortune 50 CEOs are very worried about cyber defense.

4. 21:01

   That means they're going to start to make this their number one priority to address at the board level. That means the budgets for cyber defense are going to go through the roof. Now, let me ask you a question. When you have to do cyber defense in an era of AI, are you going to use old school software or are you going to use AI? Obviously. You're going to use AI. And because of that, because cyber defense is going to need to grow, the U.S. is going to need to provide competitive AI for cyber defense.

5. 21:31

   It is a classic arms race scenario. And so the U.S. cannot simply step back and pause and turn off AI. It is already everywhere in the world. Therefore, the risks are rising. Therefore, we need to increase our investment in AI to maintain our cyber defense capabilities. both at the enterprise level and at the government level, and potentially even at the civilian consumer level. So I think that the cyber defense necessity actually mandates that we continue to build out data centers to support it. The second prediction I'll make is I think 12 to 18 months from now, we're going to shift away from this conversation about managing models, regulating models, controlling models, estimating models, scoring models, all this sort of stuff that's going on. Because like we mentioned last week, every 11 days, we're seeing new models get

6. 22:19

   published. Now it is unbelievable. And this week, Google released their new model that also seems to be more performant than any other model. And every couple days, we're seeing a new one come out. So the model control systems are not going to solve any problems. What I think is going to end up happening, and this is a prediction no one may like, but I do think we're going to end up in a world where governments are going to regulate data center access and usage. They're going to say the data centers are here. But in order to maintain defense capabilities, we are going to decide who gets allocated a certain number of GPUs, certain amount of inference, a certain number of chips, a certain number of servers. Because at the end of the day, one of the key metrics of cyber defense is going to be how many machines do you have allocated to defend your systems against how

7. 23:01

   many machines are coming after your systems. And it is again, a classic arms race scenario, assuming model convergence, and even, you know, if models have slight advantages over others, it seems to be the case that we will maintain pacing. And if we're maintaining pacing, then the number of machines allocated to a specific task is going to be a key measure of how well it performs in that task relative to another set of machines doing the same task.

8. 23:25

   You're just going to need to allocate more machines to it. So if you're going to allocate more machines to it, you're going to end up having the government say, wait, wait, wait, wait, wait, we need the US financial services system to get X percent of the GPUs in America. We need to make sure that Y% go to defense. We need to make sure that Z%. You're saying it's all going to get nationalized and allocated top down?

9. 23:45

   I think that's where the regulators are going to end up having to play a role. You can kind of think about data centers becoming the utility, which is the clouds, the neoclouds. Effectively, the government will end up saying, wait a second, we're not allocating enough to JP Morgan. We're not allocating enough to the financial services system. And ultimately, the number of machines that are being allocated...

10. 24:03

    Why would the private market take care of that? Yeah, it's an opportunity to sell into the government. Now Google and Amazon Web Services, Caruso, whoever, can just sell more servers.