NVDA remains a strong holding for the next 3 years due to ecosystem moat and supply chain dominance despite competitive threats.

And aren't they doing all of these discrete activities to build a GPU for Nvidia all in different places? All at different times.

So, with all these changes and disruption, what I'm trying to get at here is you've got all these players, you got the Microns, the Nvidias, and the TSMCs, and the ASMLs, etc.

What happens if like last week we spoke about your best risk reward which is Nvidia. What happens if they get disrupted? Yeah, I think So one of the lessons that I've learned from watching how Nvidia has like what it is that Nvidia is accomplishing right now.

But it's not just about what can you do but what is the system that you can bring along with you and what's the ecosystem that you can create? And I would be shocked if Elon was able to have this thing that just completely eliminated the entire need for the ecosystem that Nvidia has created and is powering.

Um So I would think that this would be something that layers on top of more likely than just completely eliminates you know, companies like ASML, companies like Nvidia. Um that there's a lot of like I said in the last episode, a lot of area for collaboration between these companies and entities.

But they had a record breaking fast tape out of their new chip, jalapeno, which allegedly is better, cheaper, faster than an Nvidia GPU.

And it this is a great data point to show that using AI, a company can tape out a great chip that works for, you know, a specific use case that they have. But one of the challenges that a company like OpenAI is going to have relative to Nvidia or relative to hyperscalers is, you know, unless they're able to just grow the overall size of that computing workload, then they're always going to be a smaller company or that comp customer than a hyperscaler or an Nvidia.

And then especially for Nvidia because Nvidia's a general platform, you know, one of the reasons why jalapeno's able to outperform on OpenAI's one inference workload is because it is a chip that's designed especially for that one job.

And it doesn't have to have all the overhead that the Nvidia system does that supports all kinds of different jobs, you know, it supports that exact job for OpenAI and then a similar job for Anthropic and then a similar job for Grok over at SpaceX.

Oh, yeah, and a similar job for Gemini over at Google even though Google has their own TPUs. Oh, yeah, and a similar job over at Meta. And so that broad support for all of those different jobs is one of the things that makes it very doable to have a chip that performs better on a job than Nvidia chips will do on that job as long as you can get the the CUDA support underneath it.

So, Nvidia chips, the GB, the Vera Rubins, whatever, they'll be op- optimal for training, but 99% of the future is inference.

Nvidia still is really good at doing inference though when it comes to large quantities of inference in the cloud.

And so, Nvidia's exceptional in today's regime of artificial intelligence at delivering some of the lowest cost inference tokens even with these systems that were kind of initially designed for that training workload, they've been able to make their adaptations um to also excel in the inference cloud workload.

But are they going to be able to grow that allocation to a point where it competes with and take share away from Nvidia? I would say it's going to be pretty hard to do that since Jensen's already out there making, you know, billions and billions of dollars worth of deals with all of the suppliers who do control memory supply and all of these upstream bottlenecks like by the time that Sam and the OpenAI team discover what the bottleneck is going to be that caps jalapeno's ability to take share Jensen was there several steps ago and already snapped that up.

just to flip the thing on its head, you know, right now OpenAI is trying to compete with Nvidia. They're trying to and I don't know, maybe you don't even call it competing. Maybe just call it diversifying their supply chain, trying to to increase their amount of leverage that they have in the relationship with a company like Nvidia and all those things are things that you like you can't be a responsible CEO of a company and not explore all of those avenues.

But on the flip side, Jensen also can compete with his cut like if his customers want to compete with him, then that kind of gives him a license to compete with them and he can go produce open weight models that compete head-to-head with OpenAI models, with Anthropic models.

We've seen them can increasingly doing that over the recent years. And um I do expect for that effort to increase and for the demand for, you know, sovereign US-made frontier open-source AI models um to just grow and I think Nvidia's in the best place to manufacture and deliver those.

And it obviously makes sense because, you know, those workloads have to run on hardware still. And so if they can increase the total demand for tokens, then that increases the demand for their chips.

And so he's got all of the incentives in the world to do that and um and maybe maybe if you do that, you increase the demand for tokens so much that all of the Nvidia GPUs get sold and all of the jalapeno GPUs get sold and all of the Google TPUs get sold and all of the and that seems to be what's happening.

Now they kind of have left building out their own compute. Now maybe building out their chip is kind of a stepping into it. But the big question is after all this noise, you're not worried about your Nvidia position, you're not worried about Micron, you not worried about ASML at least for the next 3 years.

Yeah, and like if I was, you know, if we had expanded that 3-year time horizon to 5 years, I probably would have picked SpaceX over Nvidia. And I think SpaceX is going to do incredibly well inside of that time or that 3-year time frame, too.

Um but yeah, I think all of those things are are pretty pretty safe for the uh foreseeable future unless we have some, you know, magic physics breakthrough that no one is expecting yet.