Welcome back to Market Movers, the opening bell.
While AI heavyweights like NVIDIA, Dell, and SEMTECH continue to smash earnings estimates and also beat elevated whisper numbers, macro pressure is weighing heavily on markets.
The U.S.
30-year Treasury yields hover near 5.3% and crude oil, WTI, topping $90 a barrel.
At the same time, we're keeping an eye on Fed rate expectations.
Underneath the surface, agentic AI token demand is also exploding, and this is driving infrastructure capacity to its limits.
And joining us live to break down this divergence between macro headwinds as well as real-world AI demand is David Fetherstonhaugh, principal at Dvx Ventures.
David, great to have you here.
Thank you so much for joining me.
Yeah, great to be here.
So you're here as the U.S. jobs report was released, so I do want to take a look at the macro.
There are a lot of expectations in terms of what will happen with inflation here in the U.S., and in turn, what that means for interest rates.
So how is this affecting some AI names, especially as Dell, NVIDIA, Semtec smash earnings expectations?
Yeah, it's funny.
One niche reference that keeps coming to mind as to how I reflect on the market right now is the 2011 Canadian Grand Prix.
It was the longest F1 race in history.
There were six safety cars, but it was the fastest iteration of the cars they've ever been.
And what I think is interesting is every time there was a green flag, there was a safety car.
So it just prolonged this very long race.
They never got a clean track.
And when I look at the market right now, you can look at the earnings season.
No one ever got a clean track.
Smash journeys, beat expectations, beat whisper expectations, and still something around the macro would kind of bring that multiple compression down.
And when we kind of think about areas where we want to look at in the AI trade, you slice it based on the first half of the year was very much so around anyone who was AI exposed did well.
The second half of the year is really who can translate that into durable earnings power.
And we've seen that through the earnings.
But a lot of the noise that has happened on the macro level has created a lot of volatility and compression in some of those names.
Yeah.
And I think it goes without saying that 2026 has been quite the volatile year across all sectors, all asset classes.
But here we are counting down the final months to the end of 2026.
And we have a better sense of where CapEx is taking place and expectations in terms of return on investment, but there is this narrative out there when it comes to building or overbuilding for AI, so how are you separating the signal from the noise?
Yeah, I think one of the interesting ways that you can just look at it, a lot of the references for the overbuild on historical precedents are looking at how much investment to GDP, how those cycles are already overbuilding the investment that's going into them.
But a lot of those arguments actually don't look necessarily at the demand aspect of things.
I mean, token consumption is growing exponentially.
Token per watt can't keep up and we can't put enough watts to actually power a lot of the token consumption that we're anticipating.
So when you look at how the market is reacting to that.
You're starting to see Anthropic IPO as well coming out.
That's going to be a very big lens into the demand side of things, which we haven't necessarily seen or have a lot of clarity in for a long time.
You can even point to old hardware, H100s that are four and a half years old now, and there's that whole concern around depreciation curves and GPUs that only last four years, suddenly those are being rented out for more than they were last fall.
So it kind of shows you, on the demand side of things, token consumption has gone up 14x in six months.
It's something crazy, like seven trillion tokens consumed.
And you look at everything from the hardware side, everyone's paying more for whatever they can get.
Yeah, and I think there are different data points that we can look at to see how all of this is playing out when it comes to the AI build-out as well as trade.
But should we be looking at enterprise ROI or should we be looking at ROIC?
What is your take, especially as we count down to some key IPOs, as you mentioned, Anthropic as well as OpenAI?
Yeah, I think what's interesting about the Anthropic S1 that people are very eager to read is it's the first read that we can get with more transparency on the demand side of things from the enterprise, from the consumption standpoint.
People have been alluding to the fact of there's a lot of power law going on with some of their customer concentration, and that's something that we're going to be keeping our eye on, is really understanding how are those tokens being consumed, where is their revenue really coming from, how diverse is that?
That's going to give you a little bit more of a coupling to, you can see some of the hyperscalers get some ROI back on their cloud spend.
Can you also see that on the enterprise side with token consumption?
Are people actually driving more tokens?
Who's actually driving more tokens, and how can you evaluate the health of that concentration mix?
Yeah, and David, while I have you here, I do want to get your perspective on Semtec.
So our viewers out there might be familiar with NVIDIA, Dell, and all the announcements as well as acquisition news, partnerships that have been coming through for NVIDIA as well as Dell.
But when it comes to Semtec, give us your take on this name.
Yeah, you hear some of the news of memory being de-specced on some of the new Rubin Ultra accelerators from NVIDIA.
And I mean, that's more of a supply constraint than anything else.
Rubin Ultra wants to make sure that they can get their accelerators to their end customers, and they're dealing with a limited amount of bits.
So they're just spreading the same amount of bits over more accelerators.
When you have less memory, you can't store it, you have to move it.
And when you move it, you want to move it as fast as possible.
Because all these thousands of GPU clusters are communicating simultaneously, and it can only go as fast as the slowest link.
And that's why we really like Semtech.
Semtech is all about when you move that data at super high speeds, can you keep the signal integrity high.
It's like as if you're having a phone call and you want to make sure that you can actually hear the other person on the other side.
That's all what SEMTECH is about and what's great about them is a lot of the discussions in AI is always about architecture.
Who wins, who loses?
It doesn't matter who wins or loses for SEMTECH because whether it's copper or whether it's lasers, they have a place in both architectures.
So that's why we like it.
When we're thinking about durable earnings power, they kind of fit into that architecture where they're a long-term earnings compounder.
Yeah, and David, finally, before I let you go, I do want to get your take on where the opportunities are when it comes to artificial intelligence, because this year has been quite the volatile year when it comes to all areas of the ecosystem, given some of the uncertainty out there.
And we've coined some new terms, including SaaSpocalypse, but now here we are heading into the final months of 2026.
So where are you seeing opportunity and what should we be paying attention to?
Yeah, I think one of the things that the AI trade tends to fall into quite a bit is this winner and loser mentality, when that's not necessarily really the case.
You can even look at memory as an example.
A lot of it gets traded off if memory would come down a little bit on the D-spec.
Everyone will kind of put into networking.
It's not actually true that one has to win and one has to lose.
There could be both winners.
And it's really about finding the pockets within both of those categories that have durable earnings power.
For memory, we still like high bandwidth memory as an example because there's some yield optimization that actually happens for a lot of these memory providers when they're stacking less memory.
They actually yield more, so it's actually a net benefit for them.
When there's a limited amount of bits, they can actually produce more bits.
So it's areas like that where you're constantly looking for who actually has a differentiated edge, who can actually continue to keep their moat, who has durable earnings power.
And if you're a long-term investor, you can use some of these opportunities and volatility and multiple compression to get your entry points.
Well, David, we will have to leave it there for today.
But thank you so much for breaking all of this down and simplifying it for our viewers out there today.
Thank you.
