NVIDIA, AI Infrastructure & The Next Big Tech Trade

The first half of 2026 delivered an explosion in AI cap, a rising token consumption as well as exploding cloud backlogs.

But behind the headlines, the tech trade is fractured.

Now physical infrastructure as well as some have dominated while software applications are struggling.

Joining us how do now to help us examine the first half wins as well as how we shape the tech portfolio playbook is on the key research analyst at Janice Henderson Shaon Baqui great to have you here.

Good morning and thank you so much for joining us.

So even this week we're seeing plenty of volatility.

Yeah, thank you so much for joining us.

So even this week we're seeing plenty of volatility when it comes to the AI trade, and we've been looking at this divide.

AI infrastructure is doing well and crushing it, but software applications not so much.

So do you think software can finally figure out how to monetize these AI tools in the second half of this year, or is this infrastructure only boom all we've got when it comes to the AI trade?

Sure, so, um, you know, we continue to look for the second half for, for, you know, signs of, um, you know, soft app software companies will be able to, you know, re-accelerate their business, um, via these AI tools, right?

So we're seeing some positive early signs with chip designers like Cadence, um, injecting AI into their EDA workflows, but I think it's really too early to call it a trend, and I think that's gonna be key for some of these applications software companies are changing the narrative, um.

I think on the other hand, you have the AI infrastructure group where fundamentals continue to strengthen here in the second half, even after sort of a red hot first half of the year.

So, uh, we sort of got a saying here, you know, you want to fish where the fish are, and right now the fish continue to be on the AI infrastructure side, so to speak.

And I do want to zoom in on a name in particular Nvidia.

So that company did manage to cross a mind-boggling $5 trillion valuation before paring back, but at the same time it has actually underperformed other chip names.

So is it just getting too big to easily move the needle, or are investors actively fleeing to hunt for the next bottlenecks?

Yeah, it's interesting.

Over the last, you know, 22 plus years, it's been really, it's really been the Nvidia show, right?

It's GPUs and have really taken center stage.

But I think, you know, as we moved in earlier this year that, you know, those bottlenecks shifted not from just GPUs, but, you know, folks shift looking for that next bottleneck, whether it's, you know, CPUs, memory, analog semis, and, you know, we were out in Taiwan a month ago and You know, these, these constraints have moved all the way down the supply chain to things like substrates and high density PCBs.

So, uh, you do have this element of bottlenecking where people are sort of searching for that next, uh, you know, choke point in the supply chain.

And interestingly, you know, there's that also comes with a sort of double whammy where you get pricing upside and gross margin expansion that you don't necessarily see with a more mature business like Nvidia.

So, Um, that's kind of what's happening on the surface, and you dig a little bit deeper.

There's, you know, the constant noise around, um, increasing competition, right?

So you have custom silicon from the likes of all the hyperscalers, right?

AMB's been more resurgent, right?

And to your point, you did bring up an interesting, um.

An interesting view that, you know, at greater than 5 trillion in cap, it's a little bit harder to move the needle, right?

So you add all that up and we've seen some relative underperformance there, but you know, ultimately I think this is a name in video that folks will ultimately revisit here.

I think the stock is trading at a discount to the market, sub 16 times consensus calendar 27 earnings for a business, frankly, that should steadily compound earnings over many years, right?

And you know, you just look here to the second half, they've got a really important product cycle with Ruben on the GPU side.

They're stacking incremental TAs with the Vera CPU.

They've got the Grok LPU, and they're actually taking more of an Apple-like approach to capital allocation.

So you add all that up, um, you know, I think over the fullness of time, you'll start to see, um, you know, the stock perform as, as, um, as earnings compound.

And Sean, I understand that as intellectual capital shifts away from the traditional chip leaders, you're also looking at very specific areas such as leading edge foundaries, objects, and memory.

So why do some of the sector titans out there, ASML, TSMC, and even Micron offer a better profile for gross margin upside right now?

Can you explain this in layman terms?

Sure.

So, we can, we can start with memory, right?

This is probably the biggest choke point now in the supply chain, perhaps even replacing uh leading edge wafers from TSMC, but, um, what you have is a situation here that, that, um, we see a really a structural supply-demand imbalance, right?

So, Demand is growing extremely fast rate, driven by AI, and this is not just the high bandwidth memory that sits next to your GPU.

We've seen this resurgence in traditional servers, right?

And these guys can consume a lot of traditional DR RAM, right?

So it's, so whether it's a high bandwidth memory, it's DDR-5 memory.

And then Nanflash, a market that has been historically very volatile, we're seeing increasing use cases there for key value cash offloading in a lot of these inferencing applications, right?

So you add it all up and bit demand growth is rising much faster than these companies can bring on capacity, right?

We hear a lot about these big capback announcements from the likes of Samsung, Heinix, and Micron, but frankly, when you, when you step back, it takes 2 to 3 years to build out a new fab.

You've got labor constraints, you've got construction lead times, things of that nature.

So we, frankly, we see that, um, you know, prices continue to rise for DRAM and NAN here for the foreseeable future.

Just given this, this structural supply demand imbalance, right?

And I think perhaps what's more, most important on this, on this memory piece is that unlike historically where you'd see boom-bust cycles, the industry is actually looking at this as a, as a strategic, sort of a strategically important business, right?

So we heard from Micron, they signed like 1616 different long-term agreements with their customers that extend out 3 to 5 years.

So what this does, it provides increasing visibility and hopefully, you know, dampens the cyclicality of the industry and, you know, we can start valuing these at a higher multiple than they've traded at historically, right?

And Sean, finally, before I let you go, I do want to round out the conversation by talking about some of the intense debates out there over the sustainability of CPAs as well as growing public backlash against massive data centers.

So when it comes to tokens, if open source models keep pushing the price of tokens down, how concerned are you about the reinvestment return of some of these multi-billion dollar facilities?

Certainly.

So, you know, I think in recent weeks, we've heard these increasing anecdotes on customers worried about overspending on tokens.

We heard from, you know, Uber and ServiceNow that they're consuming their entire, you know, year's budget of tokens.

So, I think it's natural you'd want to rationalize some of that spend.

And, you know, I think one of the ways these companies are doing that is to, um, rely, increasingly rely on, um, you know, sort of cheaper open-source models to route some of those.

Maybe less sophisticated back office workflows, I think your accounts receivable, your IT service desks, etc. to cheaper models and maybe based out of China like a DeepSeeker Kimmi and really reserving those highest value tokens for mission critical applications like coding or legal, etc.

Um, and frankly, I think it makes a lot of sense, right?

I mean, not all tokens are created equal.

And do you really want to be driving your Ferrari to the grocery store, right, with, by using a fable or, or, or one of these very top-end frontier models, right?

So, I think that's natural.

I think when you step back, even though we're seeing some rationalization of spend, we haven't really seen that token usage slow down at all.

In fact, um, you know, we step back.

We think that cheap tokens are a good thing.

Um, if anything, it, it creates incentive to, to adopt AI more and to use more tokens, and, you know, it's a, it's a term we got introduced a year to a year ago called Jevin's paradox, and we think that's going to play out over time.

And we did hear interestingly from RAM last week during, during a, um.

A podcast, from one of their, uh, from their executives, he indicated that, you know, even with this rationalization and spend, enterprises are still less than 1% penetrated with AI and that token spending actually increased 14% month on month for that top 1% of, uh, of AI spenders.

So, you know, even with some rationalization here at the margin, I think that the, the, the, the rate and pace of, uh, AI adoption and token consumption is still up and to the right, and we think this is just, you know, another example of the pie growing rather than shrinking.

Well Sean, we will have to leave it there, but I'll have to remember that driving the Ferrari to that supermarket analogy.

So thank you so much for joining us this morning and thank you so much for all of your insights.