d by the way, Stephanie. How is any of that real? Even Elon, when he commented about the person leading Mars named Elon, even Elon said how is this real? How could it possibly be real?

Now I'll tell you how it could be real. Think of all the books that have ever been written. All the books that have ever been written. Don't you think that if you could search all the books that have ever been written you would find a whole bunch of these? These meaning not just stuff about Tesla or Elon but a whole bunch of coincidences. Hey that book predicted this and that book predicted that. And the answer is this is like the Bible Code. Remember when there was a, I always use this example because it's so good, there was a claim that there were secret patterns in the Bible that if you just did things like look at the second letter of every sentence and you know stuff like that that there would be a secret message predicting the future. And there were a bunch of them, a whole bunch of them in the Bible. And then somebody had the idea to run the same algorithm against War and Peace, just a random book, and it was full of codes. So you could take any book that's a good size and you can find a whole bunch of coincidences that appear to predict the future. And I think this is one of those situations where instead of looking at all the sentences in a big book you're looking at all the books. And if you could selectively pick just the books you wanted to show people, probably this is one of thousands of different amazing coincidences you could artificially create. Or we're part of a simulation. Pick one.

All right, well the big news of course that's moving the markets is this Chinese AI called DeepSeek. Now some of you watching this are going to say I do not care about this technology but this one's really important. This isn't like just a nerdish story. This is civilization-altering kind of stuff. And I think there's still some mystery about how it was developed but the things we know is that they somehow, China got a hold of a bunch of NVIDIA chips that they weren't supposed to have access to because they would be denied to our adversarial countries. But somehow they got them. And but they didn't get many compared to how many the United States has for its big AI projects. But they had enough combined with some really clever engineering that they built something that's just as good as the big AIs.

Now the big part of the story is that they innovated way beyond what the experts were expecting and faster. It wasn't beyond what I expected because I literally predicted that this was going to happen. That there would be a super cheap alternative that would pop up and that it would hurt NVIDIA stock. Which is why I told you a while ago that I sold my NVIDIA stock is because I was expecting a fairly quickly a competitor.

Now you might say Scott how in the world did you guess this when seemingly nobody else was guessing it when I have no skills whatsoever in this domain? And the answer is just pattern recognition and economics. So if you have a background in economics it helps. But when you see this many dollars involved, the AI dollars are beyond anything we've seen really. So there's immense amount of money involved. But also if you don't get in the AI game soon enough and you're a big country it's an existential risk. So you're essentially engineering for your life. If you put people in the situation of engineering for their life to try to save their country you're going to do better than somebody who's just working for money.

In the United States we have great engineers obviously or we wouldn't have the AI that we have. But they're largely working for money and they're under the impression that they're leading in AI and maybe they are. That's a completely different incentive. And I would argue that wars have taught us that when you get in a war the innovation goes through the roof because you're trying to live. So when you're engineering yourself to try to survive suddenly you get really clever. And you've seen this a million times in a million different contexts.

So given the amount of money involved and that it's an existential risk to China and other countries I predicted that somebody would find a really clever engineering workaround and they would do it pretty quickly. And it happened. It happened almost exactly when I thought it would, you know, not long after I sold the stock.

And so what we know right now is, I want to tell you a little bit, this might be a little too nerdy for you but I'll try to make it interesting. I saw a post by Morgan Brown who was in the AI space. This was on X describing what they got right. So let me just run through this. And by the way again if you think this is a nerd story about some technology you're missing the big point. This is everything. If this thing is real and it can compete with the big expensive AIs we're in a lot of trouble. Yeah America just went from dominance to uh oh like just overnight.

So let me give you the rundown so you're educated on this. First of all in America at the moment if you were one of those regular AIs like OpenAI or Anthropic it might cost you a hundred million just on computer resources to train something. A hundred million. And they need massive data centers and thousands of GPUs that cost $40,000 a piece. But if you were this little, it's called DeepSeek, that's the name of the new AI, the cheap one, it only costs them $5 million. So five instead of 100 million. And it can match or in some ways I guess beat the current version of the OpenAI product and some others. A little better than some others.

And they rethought everything from the ground up. So here's some examples. Traditional AI likes to write everything with 32 decimal places. The cheap one said what if we just use eight decimal places and it'll be close enough. And it uses 75% less memory. So they gain 75% of memory just by saying huh we could just take a little off of this. Be a little less, fewer decimal places.

Then there's some kind of multi-token system. So normal AI reads like a first grader. It reads sort of in order. The cat sat. So it sees each word in order, one at a time apparently. This new one, the cheap one, reads whole phrases at once. So if it reads a whole phrase just like something else would read one word at a time. If it reads a whole phrase apparently it's two times faster and 90% as accurate. Now do you know what they copied? That's speed reading. They put speed reading into the model. I think I've described before because I learned speed reading when I was a kid. You don't look at words. You look at the sentence and your brain picks out the important words in the sentence. And instead of going ba ba ba each word you just look at it and you just know what it says. Now it takes practice but obviously the AI can get it a lot faster. So it's basically treating a sentence like a word.

And then it became much more proficient. Here's another trick. They built an expert system instead of one massive AI trying to know everything. So apparently it has lots of expertise built in but it only wakes up, to use the technical term, it only wakes up the expert it needs. So it's not always looking at every expert at everything. It just wakes up the part of the model that's relevant to the question I guess.

And traditional models have 1.8 trillion parameters active all the time whereas this cheap one has only 37 billion active at once. So that's a big difference.

All right so here's the results. So instead of costing 100 million to train it's 5 million. Instead of needing 100,000 GPUs it might need less than 2,000. The API costs are 95% cheaper. And it can run on a regular computer, a high-end computer but one that you could buy for gaming etc. So that's a big deal. It's all open source which is why stocks in the other companies are falling. And let's see and they don't need a billion dollar data center etc.

And apparently DeepSeek did it with a team of fewer than 200 people. And I read something else separately, this is not from Stephanie or Morgan, but separately I saw that they did something where they cleverly trained it with incentives in a way better than regular AI. But anyway those are things you need to know. Lots of clever, clever engineering. But when you see how clever the engineering was that's the difference between engineering for your life and just engineering. Because it seems to me that every one of these clever moves were available to all of our AI people weren't they? But I think our AI people were saying no we have to hit the maximum. So instead of saying well if I cut some corners it'll be 90% as good I think the American way is it's got to be the best one.

Now what are the odds that this cheap one will destroy the entire AI industry in the United States? I think low. And here's why. I think NVIDIA won't fall to zero. It'll take a hit and I think it'll just make it back eventually during the year. My guess is that NVIDIA is still a good stock. That's my guess. But you know I would just be guessing so don't buy it because I said so. There's definitely a bigger risk than there was last week. So if you think NVIDIA is the same stock as it was a week ago it's definitely not. Definitely not the same stock it was a week ago. But that doesn't mean it's in trouble.

Here's why. As the experts say the very next version of OpenAI and our other AIs will probably be better than this DeepSeek. Will DeepSeek be able to keep up? We don't know. We're going to have to see. It might be such a fast follower that nothing we can do gets out ahead of it but we'll have to find out.

The other thing I think is that, and I've been waiting for this, I think the government, I don't know if a Trump government would do it but I think the government is going to make it hard for other AIs to compete. You remember Marc Andreessen told us he was in a meeting in which some intelligence people said to the AI tech people don't bother funding more AIs because we're only going to let a few big ones survive in the United States so we can basically have control over it. Well I feel like there will be artificial barriers put in place that might even make it illegal to use this open source one. I mean it may be as simple as that.

Here's what I predict. In fairly short order somebody's going to find some code in that DeepSeek thing because remember it's open that looks suspicious or it has a back door somehow or it's got some kind of thing that isn't completely predictable. And then the rumor will start and it might not be true but the rumor will start that it's like a virus and you can't let it free in America. And it might not be true but the government out of caution will say all right we're going to ban it. You can't use this cheap one because you know we're not sure it's good. We're not sure it's safe. So I kind of expect that to happen.

So if you add the fact that the US is going to try to stay ahead and you know we don't know if DeepSeek can reach the next level of, we don't even know if it can reach the next level of AI which is AGI, you know more of a general intelligence. But we also don't know if the American companies can. They say they're going to be there by two years but I'm not so sure. So we'll see. So lots of questions about this but it doesn't mean the end of AI in America.

Now I make similar predictions. Now you probably wonder why do I keep talking about battery technology and the reason is it's the same as this AI. I'm expecting a huge breakthrough in battery technology because the stakes are so