OK, so let's try to unpack this. We are looking at a future where the most valuable real estate isn't on the ground. It's, what, 300 miles above it. The sheer size of the financial bet here is just staggering. We're talking a potential $1 .5 trillion IPO. And it isn't just about launching rockets anymore. This is about building real functional infrastructure, literally a cloud in the clouds. Hmm. This is where high stakes sci -fi really meets, you know, global compute
infrastructure. It's the new gold rush. Except instead of gold, we're mining zero G compute power. Welcome to the Deep Dive. Today, we're synthesizing a stack of sources that sits right at the intersection of, well, exponential AI growth and the very real physical limits of our planet. We're watching the boundaries of what's possible get pushed to two extremes. On one hand, massive centralization off -planet, and on the other, hyper -efficiency right on your own desktop.
And we have a phenomenal roadmap to get you plugged into the high -value insights here. We're going to start with the absolutely massive scale of that SpaceX orbital AI vision. Then we'll get into a pretty crucial kind of vulnerable admission from the AI community about beginner education, which, by the way, includes an immediate security warning you absolutely need to act on. And finally, we'll look at the technical counterpunch from NVIDIA, proving that intelligence isn't always
about brute size. Sometimes efficiency wins the day. Let's start with that colossal valuation. the numbers are just they're beyond anything we've really seen in financial history spacex is rumored to be prepping for a 2026 ipo at a valuation of 1 .5 trillion dollars 1 .5 trillion if that actually happens it would be the biggest stock market debut in history Period. And just to give you some context for that, that's not just bigger than Saudi Aramco's debut. It's way
bigger than Meta at its peak valuation. In fact, it's larger than the entire U .S. defense industrial base combined. It's a total reevaluation of what one company, one tech infrastructure company could even be worth. And just to ground that a little bit. Their current private valuation is already sitting at around $800 billion. So the bet investors are making is that they nearly double that value in just two years. Right. And that soaring valuation is tied directly to this
cloud in the clouds master plan. It's so much more ambitious than just basic Starlink Internet. Phase one is revolutionary. Dedicated AI data centers in orbit. They're upgrading their upcoming Starlink V3 satellites with specialized AI chips designed to process huge amounts of data right there in space. So the Q -shift is processing the data where it's actually generated in orbit, not beaming it down to Earth first just to send it back up again. That cuts out a massive amount
of latency. Exactly. I mean, think of it like stacking Lego blocks at data centers, but you're assembling the whole thing in low Earth orbit. And phase two. Phase two is the ultimate sci -fi infrastructure play. Factories on the moon. The plan is to eventually build satellites and all this orbital gear using lunar materials, which cuts down on the astronomical cost of lifting every little component out of Earth's gravity. OK, so investors are clearly buying into this.
They're paying 60 to 70 times forward revenue for this. This sci -fi meets infrastructure story. But why now? I thought the company was always clear they'd wait until Mars was a regular trip. Because AI changed the game. It changed the timeline, and it changed the economic necessity. The problem is energy and heat. Earth is, frankly, running out of physical capacity for compute. Cooling these massive, always -on AI data centers is becoming an existential crisis for the environment
and for the balance sheets. That's the real shift in. Space gives you three things we're running out of down here. You get near -infinite solar power, a perfect vacuum for cooling, and zero land -use fights. Precisely. And they want to get there first, establish dominance. But they're not alone. We're seeing Bezos working on proposals for orbital gigawatt data centers. I mean, literal power plants in space. And Google is deep into testing radiation -hardened AI chips for these
kinds of extreme environments. The competition is heating up and the stakes are incredibly high. But this valuation assumes they pull off moon factories and orbital data centers with just unprecedented reliability. Isn't that just deeply speculative? bubble territory? Or is there some hard metric we're missing? Investors are paying those high multiples because they see this as solving Earth's crippling compute and energy limits. That really sets the stage for the scale
of this whole thing, doesn't it? But let's bring it back down to Earth for a moment. Let's focus on the people who are just trying to learn how to use these powerful new tools. Yeah, what's fascinating here is a really genuine, vulnerable admission. from a huge AI community, over 70 ,000 subscribers. Yeah. And they realized they had just, well, fundamentally ignored beginners. They kind of assumed everyone was already at an intermediate level or had been building alongside
them from day one. And that's the core struggle, isn't it? If you're trying to catch up in AI right now, it feels like drinking from a fire hose. You log on and five new major models have dropped while you were asleep. Totally. You know, I still wrestle with prompt drift myself sometimes. And for anyone new to that term, prompt drift is just when your AI model slowly starts to forget the original instructions you gave it. It's frustrating
even for pros. So the course correction here is a real commitment to foundational knowledge, which I think is just essential. So they've developed something called the Beginner's Daily AI Plan.
It's free. And it's structured around building simple... actionable skills the idea is a day -by -day habit plan just one core idea a day so you don't get paralyzed and crucially it focuses on beginner friendly tools no coding needed that's how you actually build sustainable confidence right instead of just collecting links you'll never get back to but speaking of tools and browsers yeah we have a critical and immediate security warning that really needs your attention right
now what did the sources find Researchers caught popular Chrome extensions, and I'm talking some with up to 8 million users, like UrbanVPN Proxy. They were secretly copying and selling users private GPT or Gemini chats for a profit. These extensions needed access to your browser activity to work, and they were just monetizing your most sensitive data. your private AI conversations. That is a terrifying breach of trust. When we use these tools, we assume the data is secure.
How did they manage to pull this off without anyone noticing? Well, they leveraged that trust. A VPN or a utility extension needs broad access to your browser data just to function, and users grant that permission without thinking. But once they have it, the extension could scrape everything you were sending to or getting back from the AI portals. The immediate takeaway is... check
your extensions. Now, if you've granted read and change data on all websites permission to any utility you don't absolutely need, disable it or just remove it. It's so easy to focus on all the exciting new tools, but the very thing that makes them so powerful, their ubiquity, is why we need this constant vigilance. You have to maintain extreme skepticism, especially with browser extensions asking for access to your
sensitive data. Okay, shifting gears a bit, let's look at the globalization and... the more formal adoption of these tools. This is driving both the need for that vigilance and the huge demand for talent. Accessibility is just increasing so rapidly. It really is. I mean, take Google Translate. It now offers near real time audio translation using just any regular pair of headphones. That just instantly breaks down language barriers
for business, for travel, for anything. And then you have projects like Sam Altman's World App. It's basically a proof of human super app for digital identity. It's got secure chats, global payments, and it's already running in over 100 countries. That's a massive global footprint for a system that kind of bypasses traditional banking. And all that global competence is translating into huge domestic demand for talent. The U .S.
government is making a major play here. They're planning to hire a thousand top techies for a tech force. The goal is to drive specialized AI projects across different agencies. And we're not talking entry -level jobs. The salaries are between $150 ,000 and $200 ,000 a year. They're partnering with huge players like Microsoft, Adobe, Amazon, Meta, even XAI. to pull in that top tier talent. Which just shows you how critical AI competence is viewed now. It's not a niche
coding skill anymore. It's considered foundational national infrastructure. And they are willing to pay top dollar to compete with the private sector for that expertise. And the investment world completely agrees. Lightspeed Venture Partners just raised a staggering $9 billion across six new funds, all targeting AI development. That boosts their total assets under management to
over $40 billion. That kind of capital raise confirms that investors see AI not as a trend, but as the core utility for the next decade. When you see that kind of money pouring in, you know the future is already here. So what does the government's willingness to pay $200 ,000 salaries signal about the competitive value of just fundamental AI skills right now? It signals AI competence is now highly valued infrastructure. It's attracting top tier pay because it determines
national competitive advantage. We've covered the macro view trillion dollar bets in space, massive global talent acquisition. But the AI story isn't just about infinite scale. It's also about extreme efficiency. So now let's dive into the technical breakthroughs that might actually make powerful AI cheaper, more private and, you
know, decentralized for the average user. Yeah, this segment is so important because for the last year, everyone's been watching the open source race, particularly China, which has been leading on sheer model scale. They're consistently at the top of the 2025 leaderboards with massive models like DeepSea, Quen and Kimi. The conventional wisdom was just bigger is smarter. More parameters
means a better model. But that arms race for size relies on building bigger and bigger data centers, which brings us right back to that original problem of energy and capacity limits here on Earth. Exactly. But NVIDIA? NVIDIA took a completely different path. They decided not to compete on size, but on intelligence through efficiency. They released something called Nemotron 330B Nano. And this model is a fundamental redesign.
It's a highly distilled model focused entirely on running powerful open models efficiently on consumer hardware. Okay, let's unpack that efficiency because the technical details here really matter. What makes it so much more efficient? The key metric is throughput. And throughput just means how much information the model can process and deliver quickly. Nimitron 3 Nano achieved 3 .3 times higher throughput than previous open models of a similar size. And that translates directly
into raw speed. It's generating text at 377 tokens per second. 377 tokens a second. To put that in human terms, that's like generating a sophisticated 5 ,000 -word analysis paper while you wait for your coffee. It's a massive difference in speed. And it's not just fast. It's still smart. It managed to get a bronze medal on the... very difficult international math Olympiad benchmark. And it's built specifically for long context tasks and sophisticated agentic work. Could you
quickly define agentic work for us? Sure. Agentic work just means the AI can take a complex, high -level goal, break it down into the necessary steps, and then execute those steps on its own, all without constant hand -holding from a human. This is where it gets really interesting for me. NVIDIA didn't just build this. They open sourced the entire stack. They released the three trillion tokens of pre -training data, the post -training data sets, the full infrastructure
code. That level of transparency is pretty rare. And that democratization is the whole point. The crucial accessibility part is that you don't need to rent some giant cloud server anymore. You can run this model locally on your own machine. You just use tools like LM Studio and it only needs about 25 gigabytes of RAM. Right. And LM Studio is basically a user -friendly app that helps you download and run these open source models right on your own computer. Precisely.
You keep your data private, you customize the model for your own needs, and you can run it forever. Whoa. I mean, just imagine scaling that efficiency to a billion queries without needing the massive energy drain of a giant data center. That's a total game changer for localized, customized AI. So if models become this cheap and efficient to run privately, will this shift towards smaller, more optimized models change how companies structure
their AI adoption? Yes. Efficient models allow for cheaper, private, and permanent deployment of customized AI. It shifts power away from centralized cloud providers. So what does this all really mean for us? Looking across all the sources, it feels like the boundaries of computation are literally leaving the planet. The scale is being pushed off world, driven by Earth's capacity limits, and just the sheer speed of AI innovation.
Right. And whether that computation is happening in low Earth orbit or right there on your desktop, the path forward requires efficient models, constant... learning and absolute security vigilance. You have to be proactive about what data you're sharing, especially with those browser extensions. That orbital data center concept really forces a critical
reflection on governance, though. If AI processing moves off world outside the reach of existing national laws, what kind of ethical or regulatory framework do we even need for data centers that operate beyond any single country's jurisdiction? That's a deep question. The rulebook for space is still being written, and AI is moving way faster than any treaty can keep up with. A great question to take with you as you process the future of compute. Thank you for sharing your
sources and diving deep with us today. Until next time.