China's AI ambitions raise tech gap concerns

Transcript

All right, let's get right into it. Forget what you think you know about global power, because the real contest, the one that's going to define our entire century, isn't being fought with tanks or entreaty rooms. Nope, it's being fought with algorithms and silicon. We're talking about the AI race between the US and China, and believe me, it is heating up fast. Now, I know bifurcation point sounds like some academic jargon, right? But what it means is a moment where everything splits, where the future can go down to totally different paths. And this quote from way back in 2020 was spot on. It tells us this isn't just another competition. We're at a genuine turning point, and the stakes could not be higher. And that brings us to these central mystery of our story today. Just a little while ago, a Chinese startup you probably never heard of called Deep Sikh dropped an AI model that was, well, it was world class. It wasn't just competing with the giants in Silicon Valley on some tests, it was actually beating them. And this happened after years of tough U.S. sanctions. That were designed to stop this exact thing from happening. So that leads us to the big question, the one we're going to try and answer here. How on earth did they do it? Have all EU export controls, this whole chip war just failed? Or is there something else going on that we're missing? We're going to play detective and follow the clues. OK, our first clue is all about the most basic building block of AI compute. We're talking raw processing power. It's what you need to build and run these incredible models. And you know, this is where the battle lines were really drawn first. So the US strategy for a while now has been to create a choke point. It started back in 2018, locking China from getting the Super advanced equipment they need to make their own high end chips. Then in 2022, they just banned the export of the top AI chips altogether. But then plot twist. In 2024, a massive breach allowed a Taiwanese manufacturer to, well, inadvertently help Huawei get its hands on millions of advanced chip dyes. It gave China a big boost, even if it wasn't the absolute cutting edge. China's answer to all this? Go big. I mean, go huge on infrastructure. They launched this colossal national project, Eastern Data, Western Computing. The idea is pretty simple, actually. Build these enormous data centers out in the remote Western deserts where land is cheap and there's tons of renewable energy. And then just pipe all that computing power to the tech hubs back east. And boy, have they been building. We're seeing these gleaming futuristic buildings just rising out of the desert. But here's the paradox. The weird part. Despite this absolute construction frenzy, reports from inside China and you're seeing people online talking about it are telling a very different story. A story of empty server racks and silence. So just how empty are we talking? Well, local Chinese media outlets are reporting that as much as 80% of all this new capacity. Is just sitting there completely unused. 80%. It's an astonishing number. They've built the farms but it seems like they forgot the servers. And this brings us to a really crucial distinction in how AI works. You see, there's training. That's the Super heavy lifting, creating a model from scratch. It takes thousands of the most powerful, most expensive chips, the exact ones EU restricted. But then there's inference. That's the part where you actually use the model. And that part, which is like 80 to 90% of all AI work can run on older, cheaper chips. So here's the thing. Maybe China isn't trying to win the training race. Maybe they're preparing to completely dominate the inference battlefield. All right, let's move on to our second clue. And this one really pushes back on an idea that's been around for a long time, that China has its unbeatable advantage when it comes to data, the fuel that makes all AI go. I mean, on paper, the advantage looks massive, right? You've got 1.4 billion people by next year, China set to generate almost 49 zettabytes of data. That makes them the world's biggest data producer. And the logic seems so simple. More people means more data, which means better AI. But it's just not that straightforward because it's not really about the quantity of data. It's about the quality and how easy it is to access. Just look at this breakdown. This is from a huge web data set used to train AI models. English makes up. Almost 44% of it and Chinese a little over 5. See, a ton of China's data is locked away inside, closed as like WeChat or it's just not digitized. Finding high quality Chinese language data to train these models is surprisingly hard. And this really gets to the heart of the paradox. The Chinese government is basically trying to do 2 opposite things at the same time. On one hand, yeah, they wanted unlock all this data to fuel innovation. But on the other hand, they have to control information to maintain security and, you know, ideological purity. And that constant tension between being open and being in control creates a huge bottleneck for their AI developers. So for our final clue, we need to look at the unique way Chinas AI industry is actually structured. It's, well, it's very different from what we see in the West. It's a really dynamic space, super crowded. You've got the established giants, you know your badus and $0.10. Then there's a group of well funded startups that people are calling the AI Tigers. And of course you've got these total wild cards like DeepSeek that just seem to OU out of nowhere. But they're all playing in a very, very specific system. And that system is defined by one key idea, military civil fusion. It's a national strategy and it deliberately blurs the lines between a private tech company, a university lab, and the People's Liberation Army. The goal is written right there in black and white. Make sure any tech breakthrough in a commercial. AB can be quickly handed over for military use. So the government isn't just the referee here, it's the biggest, most important customer by a long shot. State owned companies are driving the vast majority of AI projects. What this does is it creates A guaranteed state controlled market that rewards Chinese companies for building AI that serves the country's strategic goals. OK, we've looked at the clues, the compute puzzle, the data quality gap, and the state driven ecosystem. So now it's time to put it all together. It's time for the revelation what the real battlefield in this AI race actually is. So think about it, if China can produce a world class model like DeepSeek even without access to the best chips, and if that massive data advantage we all talked about is kind of a mirage, then what is America's real lasting edge? What's the one thing we should actually be paying attention to? It comes down to 3 words. It's not about the quality of one single model, it's about the total nationwide ability to run AI systems at a massive scale. Total compute capacity. The Rand Corporation put it perfectly. They said America's real advantage. It's true. Moat isn't just having the best lab experiment. It's about having the industrial capacity to roll out the benefits of that experiment to everyone. See, China can focus all its limited resources to create one amazing showpiece, but it doesn't have the overall capacity to deploy AI across its entire economy. And the difference in scale we're talking about here is just it's staggering. the US has roughly 10 times more total compute capacity. In China, so think of it like this for everyone AI worker that China can put to work and its economy. the US can deploy 10. That's not a small gap. That is a massive compounding economic advantage. Now does this mean the race is over? Not at all. Far from it. China is playing a long strategic game here. It's a lot like the ancient board game Go. It's patiently building infrastructure, controlling territory on the board and just waiting for the right moment to make its big. Move This is a marathon, not a Sprint. And that leaves us with one last kind of sobering thought. We are so obsessed with asking who's going to win the AI race. But maybe, just maybe, that's the wrong question. As both of these countries rush to weave these incredibly powerful, not fully understood technologies into every part of their societies, from their economies to their militaries, maybe the more important question is this What happens to all of us if one of them, or both, makes a catastrophic mistake?