When these massive AI models are running, whether you're training the next big thing or just serving billions of queries a day, there's this really simple question that every company has to ask itself. Are we trying to be the fastest? Or are we just trying to be the cheapest? Because right now, Google is... very clearly choosing to be the cheapest. And that whole strategy is their direct shot at Nvidia's dominance. And welcome
to the deep dive. Today, we're digging into the latest intelligence on these new AI infrastructure wars. And it's... It's a really fascinating fight because the goal isn't just about winning on performance benchmarks anymore. It's about making AI compute so cheap that it just becomes the default. Exactly. So first, we're going to unpack that whole cost per token battle. It's Google's very specialized TPUs versus NVIDIA's GPUs that are everywhere. It's a fight for the very foundation
of AI. Yeah, and after we break down that chip war, we've got a really fast -paced segment, some critical security alerts, a look at some surprising new consumer tech. I'm talking translation glasses. And we'll explore some new interactive learning tools that are coming online. And then finally, we're jumping into a huge medical breakthrough. This one is out of Harvard Medical School, a
new AI tool called PopEVE. And this system is solving these really complex genetic mysteries that, you know, even the big one, Alpha Missense, couldn't quite get right. We'll look at the data, and it's pretty clear this is a huge step forward. So let's start right there with the hardware. Looking at Google versus NVIDIA, I mean, the sources we saw make it really clear. Google has completely given up on winning the benchmark race. They're going all in on cost supremacy.
That's the pivot. That is the entire strategy. Google's goal is to make AI compute so cheap, so accessible, that other companies almost have no choice but to use Google Cloud. They want to win on price. And the key to their advantage, they own the whole stack. Right, the entire vertical stack, from the chip design itself, the TPUs, all the way to the data centers they run in, and then the software that sits on top of all of it. And that control just gives them this
incredible power over pricing. Precisely. I mean, if you look at Nvidia's market right now, a huge chunk of their revenue comes from the margins on their hardware. We hear estimates of, what, 70 % markups on those high -end GPUs? Meanwhile, Google can basically sell its own custom TPUs through Google Cloud at cost, or maybe even below cost. And they can do that because they just make the money back on all the other services that are tied into that cloud ecosystem, or just
on pure volume. And this is a total game -changer when you look at where the real spending is happening now. Absolutely. The market has shifted so dramatically, 90%, maybe even more, of all the AI spending today. It's not for training the models. That's a one -time cost, basically. Most of the money is for inference. It's running the model every single time a user asks a question. So the main thing that these big cloud customers care about
is no longer raw speed. It's not FLOPS, floating point operations per second, which is what training always focused on. Correct. The metric that truly matters now is the lowest possible cost per token, but at a massive scale. And Google has designed the TPU specifically for that one job. Drive the cost of inference down as low as it can possibly go. And this cost advantage is, well, it's causing real anxiety for NVIDIA. The buzz we're hearing is that huge players, think meta, anthropic.
they might shift billions of dollars in compute spend over to Google's TPUs. Even if they only move a little bit of that, it could shave, what, 10 % off NVIDIA's AI revenue. That's an earthquake. It is. Now, NVIDIA's defense is strong, but it's entirely about the platform. They immediately come out and say TPUs are locked in, narrow purpose, inflexible. And what's so fascinating is how they use CUA, their software ecosystem, as their shield. CUA is the bedrock of their whole argument.
It is. The combination of their GPUs and CUDA. Well, it works with almost any model. It handles training and inference, no problem. And it runs everywhere, any cloud or even on -prem. So NVIDIA is fighting a cost war with the platform argument. They're basically saying, we're the default. We're the flexible ecosystem you can bet your
whole company on. So if that lock -in argument is so powerful, you know, the fear of getting stuck with one vendor like Google, what's stopping these big players, the metas and anthropics, from... Just sticking with NVIDIA's flexibility for the long term. The massive cost savings. It just incentivizes adopting specialized hardware, even with those lock -in fears. Dollar stock. Especially when you're dealing with billions of user queries every day. It's a tough trade
-off. Okay, switching gears. Let's hit some rapid -fire updates from the world of AI. Starting with security and trust, which this just keeps coming up in all the sources we see. Yeah, we saw a pretty big security alert about chat GPT user data. And it wasn't OpenAI's main platform that got breached. It was a third -party partner with a sloppy configuration that exposed user data, emails, things like that. And this just brings up such a critical point about the whole
AI ecosystem. As we weave these tools deeper into our businesses, our security isn't just about the main company anymore. It's tied to the weakest link in the chain, the partners, the APIs, all the extensions they use. It's a distributed risk. Exactly. Even if you have perfect internal controls, that one API gateway on a partner system can be the point of failure. And honestly, it's something I still wrestle with, you know, trying to manage security across, what,
a dozen different APIs and platforms. It's just... It's incredibly difficult to keep a consistent security posture when you rely on that many outside integrations. That's a really important point. Meanwhile, this whole idea of trust is being tested by the tech itself. We all saw those Thanksgiving photos of Elon Musk and Mark Zuckerberg that just went everywhere. Oh, they were so convincing. But verified later as totally fake, made by a
tool called Nano Banana Pro. It just goes to show you how easy it's becoming to create really high quality, convincing misinformation. Deep fakes are here. On a more positive note, the tools for learning are getting way more interactive. I'm really excited about this. Gem and I just rolled out these new interactive diagrams. And this is where the tech goes beyond just being
like an audio textbook. If you're looking at a complex system, like a diagram of a cell or the digestive system, you can now tap on any specific part of it. And you instantly get a definition, a deep explanation, all the context just for that one little piece. It's like having a dynamic tutor for any complex image you see. And we're also seeing AI pop up in some surprising consumer hardware. Alibaba just dropped some
AI glasses that are surprisingly cheap. They look pretty much like normal glasses, but they can scan prices and translate speech in real time as you're walking around. A practical use case. Finally. And then on the pure ambition side of things, IBM is launching a $500 million. And it's focused specifically on AI and quantum breakthroughs. Their goal is not small. They're aiming for a fault -tolerant quantum computer by 2029. Wait, by 2029? Whoa. I mean, imagine
scaling that. In just four years. Fault -tolerant compute power at that level. That changes, well, it changes everything. Every industry from material science to finance. It changes what's even possible. Before we move on to genetics, we did see some really useful career advice from an ex -meta director about how to break into the AI field. So beyond the usual talk about getting a PhD, what was the most valuable takeaway you saw from
those tips? Practical experience. Solving real -world problems is way more critical than just having advanced degrees. That makes sense. Okay, so moving from compute and careers, let's jump into a genuine breakthrough in medicine. We're talking genetics. And the huge challenge of finding that one critical disease signal inside all of the harmless genetic noise. Right. So DeepMind's alpha missants made huge headlines for flagging potentially harmful DNA mutations, but just flagging
a mutation. That's only the first step. The really hard part is figuring out which of those mutations actually causes a disease and which ones are just, you know, common variations that don't do anything, the background noise. And that difference, the signal versus the noise, that's exactly where this new AI from Harvard Medical School, Poppy VE, is just proving to be incredible. PopBVE's accuracy really comes down to its method. It
doesn't just look at human DNA. First, it analyzes mutation patterns across hundreds of thousands of different species. That gives it this massive evolutionary context. It's like checking a variant against the master blueprint for all of life, not just the latest human version. So it's basically asking, how important has this gene been over millions of years? If it's essential for a frog and a fish, it's probably pretty important for
a human too. Precisely. And then it calibrates those evolutionary predictions against these huge databases of healthy human genomes. Just to double check if a variant is actually common in people who don't have the disease, the result is a much, much more reliable ranking system for doctors. And the numbers here are just dramatic. Pop EV cuts false positives by over 75 % compared to alpha -miscence. That is a huge improvement in clarity for a diagnosis. Think about what
that means for a patient. Alpha -miscence flagged 44 % of healthy people as having harmful genetic variants. That creates so much unnecessary anxiety and confusion. Pop EVD, after it gets rid of all those false alarms, only flags 11%. That reduction in noise is life -changing. And this is where the real -world impact is just stunning. Researchers... ran PopEvie on the data from 31 ,000 undiagnosed children, all with severe developmental issues. These were cases that had stumped doctors
for years. The results were immediate. I mean, just incredible. Poppy solved one out of every three cases that had previously been unexplained. And it wasn't just confirming things we already knew. It flagged over 120 new genes that had never been linked to a human disease before. And get this, at least 24 of those have already been independently verified by other research teams. That's massive validation. It proves this AI is moving from just being an identifier to
a real diagnostic discovery tool. It's actually accelerating research. So given how successful PPE has been with these rare developmental disorders, how quickly do you think this kind of method could be adopted for screening more common hereditary conditions? The high accuracy is going to rapidly accelerate its adoption for broad population screening and for personalized medicine. Wow. This has been a really comprehensive deep dive.
We've covered the infrastructure battles, the security landscape, and now the frontier of medicine. Can we quickly recap the big ideas we hit today? The big strategy battle in AI compute, it's not about raw speed anymore. It's about owning the stack and winning on cost. Google is making a massive, very deliberate play for inference supremacy.
And as these AI tools get deeper into our lives, our personal security is critically dependent on the reliability of all those third -party partners and APIs, not just the main company. And finally, AI, like POPEE, is moving beyond simple prediction. It's becoming a genuine, verifiable tool for diagnostic discovery. It's drastically cutting down that painful, crucial diagnostic time for families who are dealing with these
severe, unexplained illnesses. Which really brings us to our final thought for you to think about. If Harvard's AI can solve one in three previously unexplained genetic mysteries today, What percentage of major human illnesses, the common ones we all deal with, will be fully explained by these AI models in, say, the next five years? That's the question. And that's what's driving all this innovation forward. Thank you for sharing your
sources with us for this deep dive. We really encourage you to explore these ideas further. We will catch you on the next deep dive.