The AI Compute Trade Is at Risk - Here’s What Could Win Next

Vincent (0:00)

If there's any negative news around frontier AI growth for kind of a sustained period of time that is convincing investors that this is at risk, there will be volatility for sure. Even though we're moving into a world of AI agents that we'll use to compute ultimately. That's what I'm saying.

LG (0:15)

What's up, everybody? It's all GD Sett here, and welcome to Milk Road Stocks, the daily market show that refuses to let Meta's muse order me a new oven, because what if it doesn't have an air frying option? Today is September 25th, 2026, recording on September 24th. Listen, it's our last show of the week, and the theme this week has been very clear on our show. Short-term, we're bearish because the frontier models are in trouble, because everyone's going to be using agents instead, and those agents are coming from big companies, not only just making really good products, but have mass distribution, and are making something that's actually more functional for the everyday man, and that puts those frontier models and their massive spending, that also drives up the premium, the price of the AI trade in jeopardy. Today, one of our leading analysts, Vincent, is on the show to break this down for us, and tell us just how real those risks might be, and how that rotation may happen as he continues to express this trade in his Milk Road PRO portfolio, which you can check out for a dollar at the link below. Before we start, I also want to remind you to like and subscribe. Every single little bit helps, and if you like this episode, or you didn't like it, or you have some kind of take, leave it in a comment below. Otherwise, enjoy the show. Vincent, welcome back to the show. Listen, the big theme this week, and I'm keen to talk to you about, is agents, okay? Yesterday on the show, I told Melvin I would download Muse and I would start using it to like, I'd order food and get my plumbing fixed and shit. I downloaded it, but then I was a little bit overwhelmed at the questions it was asking me, so I haven't used it yet. But that's my weekend plan, okay? Is to use that and use Grokbot and do some kind of a valuation.

You're on the show today to allude to the bigger business story that's going on there, because I think that that's the thing that's really starting to happen now is that these agents are proliferating, and people are wondering where that leaves our dear old ChatGPT and Claude.

Vincent (2:00)

Yeah. I guess it's a perfect introduction for our call today, because investors are asking themselves, what does it mean that Anthropic and OpenAI are losing a lot of market share? When it comes to the token volumes, to open weight models and others. While at the same time, those agents come in. Me as a European, I'm obviously disadvantaged again, because I cannot even download Muse because it's only US and Canada.

But still, I think I have some good takes also on what it means for the broader compute trade. I think there are some risks short-term to it, but also some traits that can be made around the whole topic. I think it's a good one.

LG (2:44)

Are you using GrockBot though? Can you use GrockBot?

Vincent (2:46)

Yeah, we can use GrockBot, but I'm personally using Anthropic the most, like the Fable Tools and GrockBot.

LG (2:52)

Okay, so you haven't moved over yet. You haven't shifted over yet. Maybe Muse would convince you if they eventually launch it in Europe.

Vincent (3:00)

I hope they do.

LG (3:01)

Canada, lucky to be on the list this time. We're usually excluded from this kind of stuff, so I will use it in your honor. But I'll throw it over to you, man, because we want to learn a little bit more on the business side today.

Vincent (3:10)

Give me a bit of slack in terms of explaining actually what is going on, and then I think we have a good context to understand what this whole thing means. First, let's start off with the model layer itself.

It's pretty obvious that the token costs are coming down, and that was pushed mainly by all the open weight models from China in the beginning, but now more and more also in the US with Nvidia, Palantir, Nemotron model, but also others are providing those, because in the end, the frontier AI token costs for everyday tasks, especially in companies, which is way too expensive, and there was basically better product market fit in the end for this open weight models. But this is in the end where we are today.