NCI: Are Nvidia and Tesla on a collision course?

Speaker1: [00:00:00] Anything that I should 

Speaker2: retweet, or? We're live now, so yeah, if you go there, you can hit the retweet, and that should work. Oh, good. Okay. Please welcome to Lumida Non Consensus Investing, my friend, Michael Preck. We've had Michael on several times. I think you might have the record, Michael. 

Speaker1: Oh, I'm honored.

Speaker2: Congratulations. When I first met Michael, it was at a Howard's event. We're having dinner. And then we had a one hour chat, became a two hour chat, then a five hour chat. Somewhere in the middle, there's a crowd of 20 people trying to make sense of the future and unpack Michael's brain. And then we talked again for another two hours after everyone left.

So we saw, you could see the entire S curve from little interest to the hot ball of liquidity. Max FOMO, and then on the other side of it, and so we've stayed in touch since then. It's always a lot of fun. It was a 

Speaker1: wonderful conversation, very mutual, and an awesome discussion, so 

Speaker2: For those that [00:01:00] don't know, Michael, was leading Goldman Sachs Internet Analyst in the late 90s.

So I've seen this cycle. So he knows the players and the companies that are involved in this transformation of AI. He writes a daily sub stack that really covers the category well. He's listening to the, CEO's of, say, Google Cloud or Amazon are going a cut deeper than the rest, looking at policy, looking at energy, and, at Lumida we're generalists.

We're trying to find the right themes and stay on top of the themes. We believe AI and data center is a secular trend, and we try to identify the right subject matter experts in those themes. You can only focus your attention on a finite number of areas and directions. Michael is one of those areas and directions.

Michael, welcome. Happy New Year. How have you been? 

Speaker1: Good, and Happy New Year to you and everybody listening. And it's the year is off with a bang, as far as AI is concerned. It is. [00:02:00] 

Speaker2: I'm relieved that we've had three back to back weeks with one day off in the middle, just so markets can cool off.

You and I were talking ahead of the show how there's so much velocity of news and information. And we've got quite a few topics. That we're going to talk through. I'm going to just identify the topics for the audience and we'll get right into it. One is we're going to talk about Jensen's talk at the CES conference this Monday.

They're investing both in not just digital AI, but physical AI. We'll talk about Elon Musk and what he's doing there with, not just driverless cars and autonomy, but also Neuralink, Google's bringing Gemini AI to TV sets. Meta has replaced 40, 000 truth checkers. US and China are in a AI space race.

We've got the cloud wars, we've got custom silicon led by Broadcom that looks like they're going after some NVIDIA's market share. We've got some policy conversations, what David Sachs new AI and cryptos are, what are the implications, data center and [00:03:00] energy, all right, Microsoft said they want to spend 80 billion dollars on data centers this year, who are the beneficiaries, winners and non obvious winners and yeah, there's a lot, we'll see if we can cover all that ground, but those are some of the topics.

And so Michael let's start off, what are some of the highlights from Jensen's presentation at the CES conference? 

Speaker1: I've been watching Jensen's keynotes for, as you have for a very long time. And I think this one at CES 2025 was At least three keynotes smooshed into one. There was so much that they unleashed.

And I think it's worth even for the casual observer to at least watch many of the no, at least a compilation video, 12, 15 minutes that's available on YouTube to get a sense of it. There's a lot of technology being, just throwing out there, but the broad theme He is, at a three and a half [00:04:00] trillion dollar valuation, and I don't, I'm not suggesting that he's doing this for his valuation, but for the investors who obviously, you know, people like you and me who've been long time investors, et cetera and, we understand the long term roadmap that he's executing, but for a lot of investors with a 200 percent performance over the last year, two years.

The question is, is everything that's ahead is baked into the stock or not? And certainly they have 80 90 percent market share in the large language model AI market for data centers. Anyone who does anything with AI needs their GPU chips. And the GPU chips are moving very rapidly every year in terms of surpassing Moore's law, in terms of performance, instead of 2x, it's 3, 4, 5, 10x, depending on whether you're doing Training or inference or synthetic data or de distillation, chain of thought.

All of these buzzwords that, we'll go into a little bit later, but these are very rapidly [00:05:00] inno innovation, improving innovations of what you do with these huge clusters of GPUs. And there Jensen has a relative lock, not just with this hard. But with CUDA software, which is not just a monolithic piece of software, it's dozens of libraries for every vertical that literally millions of developers building all kinds of applications up into the sixth box of my AI tech stack that I talk about a lot.

We can, it's on my sub stack a lot and we can put it up there later, but, and by the way, I recently updated it for some of these things. So the point being that. He has 70, 80, 90 percent share, depending on which category we're talking about on that. What he did at CES was identify a new segment, which he's been talking about, but he's calling it the physical AI market.

In other words, think about he's basically saying, look we have this great position [00:06:00] and moat in digital AI, but we have a big opportunity in what's called the physical AI, which is basically taking all these. Language model generative AI techniques, technologies, and applying it to the physical world, and he identified two big buckets, automobiles, self driving cars, and robots.

Coincidentally, those are the two areas that Elon Musk has been talking about for the last two or three years. Robotaxis. He did a big show. Elon did a big show last fall. He also had a big party where he had half a dozen Optimus robots come out, all in, operated by humans in the back and so on.

And so the core thing that he and announced, which were, which was very different than what Tesla did and has done to date, is he identified specific piece pieces of Blackwell based chips. For each of the segments and a lot of the software libraries that anyone, you and I can start a [00:07:00] startup in automobiles or in robotics and literally use that open source stuff or buy his chips, which are going into production.

They're not fictional. They're going into production in volume that we can get Thor for automobiles. Cosmos for foundation model that has physics built in, so you can create huge amounts of synthetic data. I'm getting into the weeds here. But the point is that over the next two to three years, he can rightfully point that these two big markets are going to be worth tens of billions and then eventually hundreds of billions of dollars, meaning, and I'm talking about a timeframe 10 years.

So this is not like quantum computing. Which is 20 30 years. This is relatively real. And the reason it's important to highlight is that even though Jensen is very closely allied with Elon and Tesla, and they're very close frenemies, this CES basically [00:08:00] officially made both of them frenemies.

You follow? And what I'm saying here is just like NVIDIA competes directly with Google and Meta and sorry, less Meta, but Microsoft and others who are building their own chips, Tesla is doing the same. And yet Tesla is also the biggest, one of his biggest customers for NVIDIA GPUs, the Blackwells, the and And Elon has gone out of his way to praise Elon's capability to build 100, 000 plus GPU clusters in Tennessee.

And this was very meaningful. The last thing I'll say that Jensen announced that was very important. I've been talking a lot about local AI inference based technologies. There are a lot of the processing for AI in the next three to five years are going to be done not only on the big data centers, But locally on your local machines and so on.

And so in that context, he announced some very exciting things like Digits, which is a Mac mini style computer [00:09:00] that literally is a supercomputer that you're going to be able to buy starting in May for about 3, 000 that has the entire NVIDIA technology hardware and software stack with Blackwell capabilities that is compatible with their cloud stuff, their auto stuff, etc.

It basically, you can create applications that work across that. Those are very specific things that are very exciting for millions, if not tens of millions of developers, and almost every vertical will create entirely different categories of products and services. With AI that I think relatively cements NVIDIA's position with the AI infrastructure piece beyond just data center applications, which is what we've been thinking about it for the last couple of years.

I think I may have lost Ram [00:10:00] or I may have been off. Hang on. Hey,

‍