WEBVTT 00:00:00.000 --> 00:00:02.319 OK, so let's try to unpack this. We are looking 00:00:02.319 --> 00:00:04.719 at a future where the most valuable real estate 00:00:04.719 --> 00:00:07.639 isn't on the ground. It's, what, 300 miles above 00:00:07.639 --> 00:00:10.320 it. The sheer size of the financial bet here 00:00:10.320 --> 00:00:13.500 is just staggering. We're talking a potential 00:00:13.500 --> 00:00:17.879 $1 .5 trillion IPO. And it isn't just about launching 00:00:17.879 --> 00:00:20.059 rockets anymore. This is about building real 00:00:20.059 --> 00:00:22.660 functional infrastructure, literally a cloud 00:00:22.660 --> 00:00:25.210 in the clouds. Hmm. This is where high stakes 00:00:25.210 --> 00:00:27.649 sci -fi really meets, you know, global compute 00:00:27.649 --> 00:00:30.449 infrastructure. It's the new gold rush. Except 00:00:30.449 --> 00:00:33.469 instead of gold, we're mining zero G compute 00:00:33.469 --> 00:00:36.210 power. Welcome to the Deep Dive. Today, we're 00:00:36.210 --> 00:00:38.609 synthesizing a stack of sources that sits right 00:00:38.609 --> 00:00:40.729 at the intersection of, well, exponential AI 00:00:40.729 --> 00:00:43.170 growth and the very real physical limits of our 00:00:43.170 --> 00:00:45.670 planet. We're watching the boundaries of what's 00:00:45.670 --> 00:00:48.409 possible get pushed to two extremes. On one hand, 00:00:48.509 --> 00:00:50.710 massive centralization off -planet, and on the 00:00:50.710 --> 00:00:52.689 other, hyper -efficiency right on your own desktop. 00:00:53.009 --> 00:00:55.189 And we have a phenomenal roadmap to get you plugged 00:00:55.189 --> 00:00:57.450 into the high -value insights here. We're going 00:00:57.450 --> 00:01:00.310 to start with the absolutely massive scale of 00:01:00.310 --> 00:01:03.170 that SpaceX orbital AI vision. Then we'll get 00:01:03.170 --> 00:01:05.590 into a pretty crucial kind of vulnerable admission 00:01:05.590 --> 00:01:08.010 from the AI community about beginner education, 00:01:08.250 --> 00:01:11.030 which, by the way, includes an immediate security 00:01:11.030 --> 00:01:14.209 warning you absolutely need to act on. And finally, 00:01:14.310 --> 00:01:16.090 we'll look at the technical counterpunch from 00:01:16.090 --> 00:01:18.849 NVIDIA, proving that intelligence isn't always 00:01:18.849 --> 00:01:22.950 about brute size. Sometimes efficiency wins the 00:01:22.950 --> 00:01:24.810 day. Let's start with that colossal valuation. 00:01:25.519 --> 00:01:27.439 the numbers are just they're beyond anything 00:01:27.439 --> 00:01:29.719 we've really seen in financial history spacex 00:01:29.719 --> 00:01:33.280 is rumored to be prepping for a 2026 ipo at a 00:01:33.280 --> 00:01:37.040 valuation of 1 .5 trillion dollars 1 .5 trillion 00:01:37.040 --> 00:01:38.700 if that actually happens it would be the biggest 00:01:38.700 --> 00:01:41.680 stock market debut in history Period. And just 00:01:41.680 --> 00:01:43.500 to give you some context for that, that's not 00:01:43.500 --> 00:01:45.959 just bigger than Saudi Aramco's debut. It's way 00:01:45.959 --> 00:01:48.599 bigger than Meta at its peak valuation. In fact, 00:01:48.620 --> 00:01:51.140 it's larger than the entire U .S. defense industrial 00:01:51.140 --> 00:01:54.180 base combined. It's a total reevaluation of what 00:01:54.180 --> 00:01:56.319 one company, one tech infrastructure company 00:01:56.319 --> 00:01:58.459 could even be worth. And just to ground that 00:01:58.459 --> 00:02:00.530 a little bit. Their current private valuation 00:02:00.530 --> 00:02:03.790 is already sitting at around $800 billion. So 00:02:03.790 --> 00:02:05.469 the bet investors are making is that they nearly 00:02:05.469 --> 00:02:07.969 double that value in just two years. Right. And 00:02:07.969 --> 00:02:10.569 that soaring valuation is tied directly to this 00:02:10.569 --> 00:02:14.349 cloud in the clouds master plan. It's so much 00:02:14.349 --> 00:02:17.370 more ambitious than just basic Starlink Internet. 00:02:17.789 --> 00:02:21.550 Phase one is revolutionary. Dedicated AI data 00:02:21.550 --> 00:02:24.580 centers in orbit. They're upgrading their upcoming 00:02:24.580 --> 00:02:27.939 Starlink V3 satellites with specialized AI chips 00:02:27.939 --> 00:02:31.280 designed to process huge amounts of data right 00:02:31.280 --> 00:02:34.120 there in space. So the Q -shift is processing 00:02:34.120 --> 00:02:36.439 the data where it's actually generated in orbit, 00:02:36.560 --> 00:02:38.939 not beaming it down to Earth first just to send 00:02:38.939 --> 00:02:40.879 it back up again. That cuts out a massive amount 00:02:40.879 --> 00:02:42.919 of latency. Exactly. I mean, think of it like 00:02:42.919 --> 00:02:44.860 stacking Lego blocks at data centers, but you're 00:02:44.860 --> 00:02:46.680 assembling the whole thing in low Earth orbit. 00:02:47.069 --> 00:02:49.610 And phase two. Phase two is the ultimate sci 00:02:49.610 --> 00:02:52.069 -fi infrastructure play. Factories on the moon. 00:02:52.189 --> 00:02:54.469 The plan is to eventually build satellites and 00:02:54.469 --> 00:02:56.689 all this orbital gear using lunar materials, 00:02:56.909 --> 00:02:59.889 which cuts down on the astronomical cost of lifting 00:02:59.889 --> 00:03:01.789 every little component out of Earth's gravity. 00:03:02.090 --> 00:03:04.610 OK, so investors are clearly buying into this. 00:03:04.789 --> 00:03:07.710 They're paying 60 to 70 times forward revenue 00:03:07.710 --> 00:03:10.509 for this. This sci -fi meets infrastructure story. 00:03:11.030 --> 00:03:13.569 But why now? I thought the company was always 00:03:13.569 --> 00:03:15.629 clear they'd wait until Mars was a regular trip. 00:03:16.009 --> 00:03:19.069 Because AI changed the game. It changed the timeline, 00:03:19.189 --> 00:03:21.729 and it changed the economic necessity. The problem 00:03:21.729 --> 00:03:25.370 is energy and heat. Earth is, frankly, running 00:03:25.370 --> 00:03:28.590 out of physical capacity for compute. Cooling 00:03:28.590 --> 00:03:31.810 these massive, always -on AI data centers is 00:03:31.810 --> 00:03:34.569 becoming an existential crisis for the environment 00:03:34.569 --> 00:03:37.069 and for the balance sheets. That's the real shift 00:03:37.069 --> 00:03:39.189 in. Space gives you three things we're running 00:03:39.189 --> 00:03:41.289 out of down here. You get near -infinite solar 00:03:41.289 --> 00:03:44.009 power, a perfect vacuum for cooling, and zero 00:03:44.009 --> 00:03:46.460 land -use fights. Precisely. And they want to 00:03:46.460 --> 00:03:48.639 get there first, establish dominance. But they're 00:03:48.639 --> 00:03:50.479 not alone. We're seeing Bezos working on proposals 00:03:50.479 --> 00:03:52.800 for orbital gigawatt data centers. I mean, literal 00:03:52.800 --> 00:03:55.659 power plants in space. And Google is deep into 00:03:55.659 --> 00:03:58.000 testing radiation -hardened AI chips for these 00:03:58.000 --> 00:04:00.060 kinds of extreme environments. The competition 00:04:00.060 --> 00:04:03.099 is heating up and the stakes are incredibly high. 00:04:03.300 --> 00:04:05.479 But this valuation assumes they pull off moon 00:04:05.479 --> 00:04:08.659 factories and orbital data centers with just 00:04:08.659 --> 00:04:12.900 unprecedented reliability. Isn't that just deeply 00:04:12.900 --> 00:04:16.730 speculative? bubble territory? Or is there some 00:04:16.730 --> 00:04:19.709 hard metric we're missing? Investors are paying 00:04:19.709 --> 00:04:21.689 those high multiples because they see this as 00:04:21.689 --> 00:04:24.509 solving Earth's crippling compute and energy 00:04:24.509 --> 00:04:27.069 limits. That really sets the stage for the scale 00:04:27.069 --> 00:04:29.189 of this whole thing, doesn't it? But let's bring 00:04:29.189 --> 00:04:31.069 it back down to Earth for a moment. Let's focus 00:04:31.069 --> 00:04:32.910 on the people who are just trying to learn how 00:04:32.910 --> 00:04:35.810 to use these powerful new tools. Yeah, what's 00:04:35.810 --> 00:04:39.990 fascinating here is a really genuine, vulnerable 00:04:39.990 --> 00:04:44.050 admission. from a huge AI community, over 70 00:04:44.050 --> 00:04:46.589 ,000 subscribers. Yeah. And they realized they 00:04:46.589 --> 00:04:48.670 had just, well, fundamentally ignored beginners. 00:04:48.810 --> 00:04:50.310 They kind of assumed everyone was already at 00:04:50.310 --> 00:04:52.449 an intermediate level or had been building alongside 00:04:52.449 --> 00:04:54.769 them from day one. And that's the core struggle, 00:04:55.009 --> 00:04:56.649 isn't it? If you're trying to catch up in AI 00:04:56.649 --> 00:04:58.610 right now, it feels like drinking from a fire 00:04:58.610 --> 00:05:01.769 hose. You log on and five new major models have 00:05:01.769 --> 00:05:04.269 dropped while you were asleep. Totally. You know, 00:05:04.269 --> 00:05:06.689 I still wrestle with prompt drift myself sometimes. 00:05:07.209 --> 00:05:10.269 And for anyone new to that term, prompt drift 00:05:10.269 --> 00:05:12.990 is just when your AI model slowly starts to forget 00:05:12.990 --> 00:05:15.389 the original instructions you gave it. It's frustrating 00:05:15.389 --> 00:05:17.850 even for pros. So the course correction here 00:05:17.850 --> 00:05:19.829 is a real commitment to foundational knowledge, 00:05:19.949 --> 00:05:22.250 which I think is just essential. So they've developed 00:05:22.250 --> 00:05:24.610 something called the Beginner's Daily AI Plan. 00:05:24.829 --> 00:05:27.490 It's free. And it's structured around building 00:05:27.490 --> 00:05:31.759 simple... actionable skills the idea is a day 00:05:31.759 --> 00:05:34.899 -by -day habit plan just one core idea a day 00:05:34.899 --> 00:05:38.480 so you don't get paralyzed and crucially it focuses 00:05:38.480 --> 00:05:41.579 on beginner friendly tools no coding needed that's 00:05:41.579 --> 00:05:43.120 how you actually build sustainable confidence 00:05:43.120 --> 00:05:44.459 right instead of just collecting links you'll 00:05:44.459 --> 00:05:47.399 never get back to but speaking of tools and browsers 00:05:47.399 --> 00:05:50.139 yeah we have a critical and immediate security 00:05:50.139 --> 00:05:52.800 warning that really needs your attention right 00:05:52.800 --> 00:05:55.300 now what did the sources find Researchers caught 00:05:55.300 --> 00:05:57.600 popular Chrome extensions, and I'm talking some 00:05:57.600 --> 00:06:01.180 with up to 8 million users, like UrbanVPN Proxy. 00:06:01.319 --> 00:06:03.759 They were secretly copying and selling users 00:06:03.759 --> 00:06:07.620 private GPT or Gemini chats for a profit. These 00:06:07.620 --> 00:06:09.720 extensions needed access to your browser activity 00:06:09.720 --> 00:06:12.560 to work, and they were just monetizing your most 00:06:12.560 --> 00:06:15.220 sensitive data. your private AI conversations. 00:06:15.680 --> 00:06:18.959 That is a terrifying breach of trust. When we 00:06:18.959 --> 00:06:20.779 use these tools, we assume the data is secure. 00:06:21.079 --> 00:06:22.939 How did they manage to pull this off without 00:06:22.939 --> 00:06:25.639 anyone noticing? Well, they leveraged that trust. 00:06:25.879 --> 00:06:29.199 A VPN or a utility extension needs broad access 00:06:29.199 --> 00:06:31.860 to your browser data just to function, and users 00:06:31.860 --> 00:06:34.500 grant that permission without thinking. But once 00:06:34.500 --> 00:06:36.699 they have it, the extension could scrape everything 00:06:36.699 --> 00:06:38.459 you were sending to or getting back from the 00:06:38.459 --> 00:06:41.220 AI portals. The immediate takeaway is... check 00:06:41.220 --> 00:06:43.459 your extensions. Now, if you've granted read 00:06:43.459 --> 00:06:45.500 and change data on all websites permission to 00:06:45.500 --> 00:06:47.860 any utility you don't absolutely need, disable 00:06:47.860 --> 00:06:50.740 it or just remove it. It's so easy to focus on 00:06:50.740 --> 00:06:53.300 all the exciting new tools, but the very thing 00:06:53.300 --> 00:06:55.660 that makes them so powerful, their ubiquity, 00:06:55.660 --> 00:06:58.560 is why we need this constant vigilance. You have 00:06:58.560 --> 00:07:01.939 to maintain extreme skepticism, especially with 00:07:01.939 --> 00:07:04.220 browser extensions asking for access to your 00:07:04.220 --> 00:07:07.100 sensitive data. Okay, shifting gears a bit, let's 00:07:07.100 --> 00:07:10.680 look at the globalization and... the more formal 00:07:10.680 --> 00:07:12.660 adoption of these tools. This is driving both 00:07:12.660 --> 00:07:15.959 the need for that vigilance and the huge demand 00:07:15.959 --> 00:07:18.160 for talent. Accessibility is just increasing 00:07:18.160 --> 00:07:20.879 so rapidly. It really is. I mean, take Google 00:07:20.879 --> 00:07:23.759 Translate. It now offers near real time audio 00:07:23.759 --> 00:07:26.720 translation using just any regular pair of headphones. 00:07:26.920 --> 00:07:29.860 That just instantly breaks down language barriers 00:07:29.860 --> 00:07:33.360 for business, for travel, for anything. And then 00:07:33.360 --> 00:07:35.639 you have projects like Sam Altman's World App. 00:07:35.759 --> 00:07:37.740 It's basically a proof of human super app for 00:07:37.740 --> 00:07:40.439 digital identity. It's got secure chats, global 00:07:40.439 --> 00:07:42.500 payments, and it's already running in over 100 00:07:42.500 --> 00:07:45.220 countries. That's a massive global footprint 00:07:45.220 --> 00:07:47.680 for a system that kind of bypasses traditional 00:07:47.680 --> 00:07:50.860 banking. And all that global competence is translating 00:07:50.860 --> 00:07:53.600 into huge domestic demand for talent. The U .S. 00:07:53.600 --> 00:07:55.720 government is making a major play here. They're 00:07:55.720 --> 00:07:57.860 planning to hire a thousand top techies for a 00:07:57.860 --> 00:08:00.079 tech force. The goal is to drive specialized 00:08:00.079 --> 00:08:03.120 AI projects across different agencies. And we're 00:08:03.120 --> 00:08:05.600 not talking entry -level jobs. The salaries are 00:08:05.600 --> 00:08:09.300 between $150 ,000 and $200 ,000 a year. They're 00:08:09.300 --> 00:08:11.560 partnering with huge players like Microsoft, 00:08:11.899 --> 00:08:16.339 Adobe, Amazon, Meta, even XAI. to pull in that 00:08:16.339 --> 00:08:19.100 top tier talent. Which just shows you how critical 00:08:19.100 --> 00:08:21.980 AI competence is viewed now. It's not a niche 00:08:21.980 --> 00:08:24.019 coding skill anymore. It's considered foundational 00:08:24.019 --> 00:08:26.779 national infrastructure. And they are willing 00:08:26.779 --> 00:08:29.199 to pay top dollar to compete with the private 00:08:29.199 --> 00:08:31.839 sector for that expertise. And the investment 00:08:31.839 --> 00:08:34.980 world completely agrees. Lightspeed Venture Partners 00:08:34.980 --> 00:08:38.139 just raised a staggering $9 billion across six 00:08:38.139 --> 00:08:41.360 new funds, all targeting AI development. That 00:08:41.360 --> 00:08:43.379 boosts their total assets under management to 00:08:43.379 --> 00:08:46.779 over $40 billion. That kind of capital raise 00:08:46.779 --> 00:08:49.539 confirms that investors see AI not as a trend, 00:08:49.659 --> 00:08:52.120 but as the core utility for the next decade. 00:08:52.360 --> 00:08:55.100 When you see that kind of money pouring in, you 00:08:55.100 --> 00:08:57.259 know the future is already here. So what does 00:08:57.259 --> 00:08:59.500 the government's willingness to pay $200 ,000 00:08:59.500 --> 00:09:02.320 salaries signal about the competitive value of 00:09:02.320 --> 00:09:05.059 just fundamental AI skills right now? It signals 00:09:05.059 --> 00:09:08.159 AI competence is now highly valued infrastructure. 00:09:08.460 --> 00:09:11.080 It's attracting top tier pay because it determines 00:09:11.080 --> 00:09:15.100 national competitive advantage. We've covered 00:09:15.100 --> 00:09:17.220 the macro view trillion dollar bets in space, 00:09:17.480 --> 00:09:20.899 massive global talent acquisition. But the AI 00:09:20.899 --> 00:09:23.419 story isn't just about infinite scale. It's also 00:09:23.419 --> 00:09:26.629 about extreme efficiency. So now let's dive into 00:09:26.629 --> 00:09:28.389 the technical breakthroughs that might actually 00:09:28.389 --> 00:09:31.710 make powerful AI cheaper, more private and, you 00:09:31.710 --> 00:09:34.070 know, decentralized for the average user. Yeah, 00:09:34.129 --> 00:09:36.250 this segment is so important because for the 00:09:36.250 --> 00:09:38.269 last year, everyone's been watching the open 00:09:38.269 --> 00:09:40.809 source race, particularly China, which has been 00:09:40.809 --> 00:09:43.090 leading on sheer model scale. They're consistently 00:09:43.090 --> 00:09:45.789 at the top of the 2025 leaderboards with massive 00:09:45.789 --> 00:09:48.830 models like DeepSea, Quen and Kimi. The conventional 00:09:48.830 --> 00:09:51.750 wisdom was just bigger is smarter. More parameters 00:09:51.750 --> 00:09:54.110 means a better model. But that arms race for 00:09:54.110 --> 00:09:56.669 size relies on building bigger and bigger data 00:09:56.669 --> 00:09:59.429 centers, which brings us right back to that original 00:09:59.429 --> 00:10:01.809 problem of energy and capacity limits here on 00:10:01.809 --> 00:10:05.570 Earth. Exactly. But NVIDIA? NVIDIA took a completely 00:10:05.570 --> 00:10:08.250 different path. They decided not to compete on 00:10:08.250 --> 00:10:10.590 size, but on intelligence through efficiency. 00:10:10.909 --> 00:10:13.529 They released something called Nemotron 330B 00:10:13.529 --> 00:10:16.929 Nano. And this model is a fundamental redesign. 00:10:17.110 --> 00:10:19.690 It's a highly distilled model focused entirely 00:10:19.690 --> 00:10:22.850 on running powerful open models efficiently on 00:10:22.850 --> 00:10:25.190 consumer hardware. Okay, let's unpack that efficiency 00:10:25.190 --> 00:10:27.450 because the technical details here really matter. 00:10:27.509 --> 00:10:29.570 What makes it so much more efficient? The key 00:10:29.570 --> 00:10:32.289 metric is throughput. And throughput just means 00:10:32.289 --> 00:10:35.330 how much information the model can process and 00:10:35.330 --> 00:10:38.990 deliver quickly. Nimitron 3 Nano achieved 3 .3 00:10:38.990 --> 00:10:41.710 times higher throughput than previous open models 00:10:41.710 --> 00:10:44.289 of a similar size. And that translates directly 00:10:44.289 --> 00:10:48.070 into raw speed. It's generating text at 377 tokens 00:10:48.070 --> 00:10:51.850 per second. 377 tokens a second. To put that 00:10:51.850 --> 00:10:53.990 in human terms, that's like generating a sophisticated 00:10:53.990 --> 00:10:56.710 5 ,000 -word analysis paper while you wait for 00:10:56.710 --> 00:10:58.710 your coffee. It's a massive difference in speed. 00:10:58.929 --> 00:11:01.470 And it's not just fast. It's still smart. It 00:11:01.470 --> 00:11:03.909 managed to get a bronze medal on the... very 00:11:03.909 --> 00:11:06.470 difficult international math Olympiad benchmark. 00:11:06.909 --> 00:11:09.549 And it's built specifically for long context 00:11:09.549 --> 00:11:12.970 tasks and sophisticated agentic work. Could you 00:11:12.970 --> 00:11:16.389 quickly define agentic work for us? Sure. Agentic 00:11:16.389 --> 00:11:19.129 work just means the AI can take a complex, high 00:11:19.129 --> 00:11:21.269 -level goal, break it down into the necessary 00:11:21.269 --> 00:11:23.889 steps, and then execute those steps on its own, 00:11:23.950 --> 00:11:26.330 all without constant hand -holding from a human. 00:11:26.620 --> 00:11:28.059 This is where it gets really interesting for 00:11:28.059 --> 00:11:30.440 me. NVIDIA didn't just build this. They open 00:11:30.440 --> 00:11:33.580 sourced the entire stack. They released the three 00:11:33.580 --> 00:11:36.799 trillion tokens of pre -training data, the post 00:11:36.799 --> 00:11:38.419 -training data sets, the full infrastructure 00:11:38.419 --> 00:11:41.700 code. That level of transparency is pretty rare. 00:11:41.919 --> 00:11:43.720 And that democratization is the whole point. 00:11:43.899 --> 00:11:46.259 The crucial accessibility part is that you don't 00:11:46.259 --> 00:11:47.960 need to rent some giant cloud server anymore. 00:11:48.139 --> 00:11:50.980 You can run this model locally on your own machine. 00:11:51.320 --> 00:11:54.179 You just use tools like LM Studio and it only 00:11:54.179 --> 00:11:56.490 needs about 25 gigabytes of RAM. Right. And LM 00:11:56.490 --> 00:11:58.350 Studio is basically a user -friendly app that 00:11:58.350 --> 00:12:00.129 helps you download and run these open source 00:12:00.129 --> 00:12:02.610 models right on your own computer. Precisely. 00:12:02.610 --> 00:12:05.029 You keep your data private, you customize the 00:12:05.029 --> 00:12:07.409 model for your own needs, and you can run it 00:12:07.409 --> 00:12:11.070 forever. Whoa. I mean, just imagine scaling that 00:12:11.070 --> 00:12:13.870 efficiency to a billion queries without needing 00:12:13.870 --> 00:12:16.250 the massive energy drain of a giant data center. 00:12:16.370 --> 00:12:19.029 That's a total game changer for localized, customized 00:12:19.029 --> 00:12:22.330 AI. So if models become this cheap and efficient 00:12:22.330 --> 00:12:25.529 to run privately, will this shift towards smaller, 00:12:25.570 --> 00:12:28.769 more optimized models change how companies structure 00:12:28.769 --> 00:12:31.720 their AI adoption? Yes. Efficient models allow 00:12:31.720 --> 00:12:33.919 for cheaper, private, and permanent deployment 00:12:33.919 --> 00:12:37.759 of customized AI. It shifts power away from centralized 00:12:37.759 --> 00:12:39.860 cloud providers. So what does this all really 00:12:39.860 --> 00:12:42.139 mean for us? Looking across all the sources, 00:12:42.220 --> 00:12:44.220 it feels like the boundaries of computation are 00:12:44.220 --> 00:12:46.679 literally leaving the planet. The scale is being 00:12:46.679 --> 00:12:49.200 pushed off world, driven by Earth's capacity 00:12:49.200 --> 00:12:52.159 limits, and just the sheer speed of AI innovation. 00:12:52.559 --> 00:12:54.440 Right. And whether that computation is happening 00:12:54.440 --> 00:12:56.559 in low Earth orbit or right there on your desktop, 00:12:56.860 --> 00:13:00.200 the path forward requires efficient models, constant... 00:13:00.110 --> 00:13:03.429 learning and absolute security vigilance. You 00:13:03.429 --> 00:13:05.690 have to be proactive about what data you're sharing, 00:13:05.769 --> 00:13:08.149 especially with those browser extensions. That 00:13:08.149 --> 00:13:11.009 orbital data center concept really forces a critical 00:13:11.009 --> 00:13:14.129 reflection on governance, though. If AI processing 00:13:14.129 --> 00:13:17.149 moves off world outside the reach of existing 00:13:17.149 --> 00:13:20.629 national laws, what kind of ethical or regulatory 00:13:20.629 --> 00:13:24.149 framework do we even need for data centers that 00:13:24.149 --> 00:13:26.690 operate beyond any single country's jurisdiction? 00:13:27.629 --> 00:13:29.789 That's a deep question. The rulebook for space 00:13:29.789 --> 00:13:32.330 is still being written, and AI is moving way 00:13:32.330 --> 00:13:34.870 faster than any treaty can keep up with. A great 00:13:34.870 --> 00:13:36.690 question to take with you as you process the 00:13:36.690 --> 00:13:38.669 future of compute. Thank you for sharing your 00:13:38.669 --> 00:13:40.789 sources and diving deep with us today. Until 00:13:40.789 --> 00:13:41.230 next time.