So everyone seems to be shouting about an AI bubble these days. You hear it everywhere. People are dusting off the history books, right? Pointing straight at the 2001 .com crash. But looking through our sources today, it feels like we might be focused on the wrong kind of risk entirely this time. Yeah, it's interesting. That 2001 bust, it was fundamentally about infrastructure, wasn't it? Too much supply. People call it dark fiber. All that empty, unused network capacity
built on speculation. The paradox now, and it's fascinating, is what if we're seeing like maximized dark GPUs from NVIDIA? where every single chip, all the compute power is just firing on all cylinders the moment it's out the door. Welcome to the deep dive. That paradox, that's really our mission today. We're going to unpack these market dynamics that make the current capital environment feel
so different from past tech booms. We'll look at the incredible speed of consumer adoption and also, importantly, the rise of these really powerful small AI models. They could really shake things up. Absolutely. And just to set the stage, our goal here is, you know, impartial analysis for you listening. We're not trying to validate the hype necessarily, but really understand the actual. mechanisms, the money, the usage, the tech that are driving this, well, unprecedented
environment. Okay. Let's dive into the market dynamics first then. We're seeing just massive AI funding rounds. NVIDIA stock is hitting record highs almost constantly, it feels like. And the actual usage of the chips is described as maybe even bigger than the funding suggests. Right. And this is where comparing it to 2001 is... useful, but also maybe misleading. That telecom bust, it happened because companies poured billions
into infrastructure. That empty dark fiber waiting for web traffic that just didn't show up fast enough is pure speculation on future capacity. And today, the sources we have are stressing the complete opposite scenario. There are, and this is a quote, no dark GPUs. The big players buying the chips, they're putting them to work immediately. It's actually a supply construct problem, not a demand one. Exactly. Which brings us to the hyperscalers. That's the term for the
big cloud providers, right? Microsoft, Google, Meta, Amazon. They're spending just colossal amounts on data centers. Billions and billions. But, and this is key, they seem to be seeing a genuine, pretty immediate return on capital, or ROIC, from that huge spend. Because the capacity is getting soaked up instantly, either by their own projects or, you know, by their cloud customers. So it's not building empty stadiums, hoping the
crowds will come later. If the infrastructure is getting used right away and generating a return, what does that actually mean for this whole bubble conversation? Why is this maybe financially safer than past manias? Well, it shifts the risk profile, doesn't it? Instead of risking oversupply and empty capacity, maybe the risk is more about depending on a really constrained supply chain. This is what one source called a fiber -rich
pumpkin season. It implies it's harvest time now, not just planting seeds for some distant future. That's a powerful way to put it. So if every piece of this expensive compute hardware is being used to generate actual revenue now, the foundation of this room feels like it's built on present income streams, not just future hopes. That seems to be the core argument for why it's different. The growth is fueled by tangible, immediate use of infrastructure, not just betting
on future capacity needs. OK. Speaking of you, Sid, let's pivot a bit to the cultural side because adoption is moving incredibly fast and sometimes with some pretty hilarious results along the way, which always gives us something to think about. Oh, absolutely. The speed is just incredible. Fun fact number one, OpenAI's Sora, their video generation model, it's exploded. Apparently it hit hashtag one on the App Store and they just launched character cameos. For a limited time,
no invite code needed. You could literally put your cat. or I don't know, your kid's favorite toy, into a short video generated by AI. That's like immediate viral mainstream utility right there. That is a huge sign of accessibility hitting the consumer level. And then, of course, you have the famous fails, which kind of remind us these models are still pretty young, right? We saw that funny story about ChatGPT folding under almost zero pressure. That was amazing, yeah.
A parent used some really complex obscure password as an anti -cheat. on their kid's computer. The kid just described the password setup vaguely to ChatGPT and asked it to figure it out. And boom, the model just gave it up immediately. Soft chuckle. Zero real pressure, total exposure. It really shows how complexity can sometimes just crumble with the right kind of simple, maybe unexpected prompt. It's a great little example of the unintended power of these accessible language
models. And you see the platforms reacting to this kind of immediate access to open AI apparently had to stop chat GPT from giving out specific medical and legal advice, personalized advice. Anyway, they're clearly trying to manage the most obvious risks as people start using these tools for really serious. But the education side is trying to catch up fast as well. University of San Francisco, for instance, launched a free online AI prompt course aimed right at beginners.
No coding required, just focusing on practical everyday uses. Which is needed. I mean, I still wrestle with prompt drift myself sometimes. Getting the exact output you want, it feels more like an art than a science occasionally. It really does. In this mix of real utility and sometimes just pure entertainment, it's definitely driving the money side. The culture and the capital feel completely linked. Like this AI singer, Zanya Monet, became the first AI artist to actually
get on the Billboard Airplay charts. That's mass market consumption happening. And then the valuations just keep climbing. founders, all 22 years old, apparently beat Mark Zuckerberg's record. Youngest self -made billionaires. Their AI startup hit a $10 billion valuation after raising $350 million. That scale is just hard to grasp sometimes. And it's not just late stage money flooding in either. Sequoia Capital, the big VC firm, they're doubling down right at the beginning, launching two new
funds. $950 million total just for early stage AI startups around the world. They are betting big right from the ground floor. And then, of course, the rumor mill keeps churning. The biggest one lately, OpenAI potentially exploring an IPO maybe in 2026 with a valuation that could hit up to $1 trillion. So here's the question then. Why do we keep seeing these like... astronomical funding rounds and valuations right alongside these immediate, sometimes kind of silly consumer
fails. I think the massive investment reflects this huge belief in the long term potential, even with the very obvious, sometimes amusing short term limits we're seeing right now. Sponsor. Let's shift gears now to efficiency. Our sources suggest this might be the next really big frontier. The race isn't only about making models bigger and bigger anymore. It's also about packing powerful features into smaller, more efficient packages. Which brings us to IBM's Granite 4 .0 Nano, their
smallest AI models released so far. Yeah, the Nano part is the real kicker here. And it speaks volumes about decentralization. First off, these models are open source. Apache 2 .0 license, which is pretty permissive. Developers can look inside, build on top of them freely. And crucially, they're optimized for efficient runtimes like this popular library called Llama .cpp. And for listeners, why should we care about something
called Llama .cpp? Well, because it essentially lets these complex AI models run really fast, even on hardware that isn't specialized AI chips. Think your laptop, maybe even a cheap device at the edge. you know, without needing access to a massive server farm, it really democratizes who can run powerful AI. Okay. And importantly, they didn't seem to sacrifice training quality just to shrink the size. These nano models were
trained on, what, over 15 trillion tokens? That's just a mind -boggling amount of data, like stacking Lego blocks of really high -quality information. And they use the same quality pipeline as IBM's much bigger models. Right. And maybe we should quickly define parameters here for clarity. When we talk about parameters in an AI model, we basically mean the size of its internal knowledge structure. Think of it like the number of connections or
variables it uses to process information. Usually, more parameters mean better performance, but also way higher compute costs and energy use. But Granite Nano seems to be pushing back against that core assumption. It's showing you don't necessarily need 70 billion plus parameters to get really useful, impactful results. In benchmark tests, it actually came out on top against other models in its specific size class, including ones from Alibaba and Google like Quinn and Gemma.
And what's really interesting is how well it did on what the sources call agenty stuff. So that means things like complex reasoning, generating code, tasks that need multiple steps to complete. It's not just spitting back facts it memorized. Whoa. OK, imagine staling that kind of capability, running it potentially billions of times on small local devices, even offline. That could totally change the economics for businesses wanting to
deploy AI everywhere. IBM even certified it for responsible AI use, which is a huge deal for companies needing audit trails and, you know, some level of certainty. That efficiency is just critical. If you can do complex tasks, tasks that used to need these giant models on a small local chip, you slash the cost per query, the energy use, everything. So thinking about the market again, what is the success of these smaller
open source models like Granite Nano mean? for the dominance of the big hyperscale players, the ones who currently own all that expensive, specialized compute power. It means competition is heating up fast, especially at the smaller, more efficient edge of the AI model spectrum. It forces the big guys to innovate beyond just sheer scale. Okay, let's try to bring this all together. Looking back at our sources today, we've really hit on three major dynamics shaping
AI right now. First, the huge investment wave seems arguably justified by this heavily utilized, maxed out hardware. We're seeing real capital return, which feels different from pure speculation. The risk profile seems shifted. Second, consumer adoption is just incredibly fast. Sometimes funny, yeah, but undeniably happening. And that's driving the need for more investment, more hardware. And third, powerful AI is actually shrinking.
Models like Granite Nano are proving that serious AI capability can run pretty much anywhere now and can compete effectively within specific size categories. And these three themes, they seem tied together by what people are calling this virtuous cycle that's driving the whole AI frenzy, right? More usage demands more chips, which justifies more spending on infrastructure, which leads to better models, which then drives more usage.
And the cycle starts again. Yeah, exactly. And we also saw some quick hits showing the real -time risks and global spread, like Google having to pull their Gemma modder pretty quickly from their AI studio after some decimation claims popped up. Shows how volatile and legally tricky this still is. And meanwhile, you see Anthropic opening its first office in Asia Pacific and Tokyo. It underlines that this whole movement
is definitely global now. So maybe here's the provocative thought to leave you with today. If immediate capital return, driven by maxing out all the current hardware, is what's validating this AI boom right now, what happens to the market structure and maybe those huge valuations when the core technology driving that return, like these granite nano models, becomes cheap, open source, and capable of running almost anywhere?
That definitely raises a big question about the long -term value proposition for the hyperscalers, doesn't it? What's their unique edge when the underlying tech becomes potentially commoditized? Well, thank you for providing these sources and for diving into this material with us today. Until next time, out to real music.