D2D Cont'd: How Much is Left in the AI Trade?
How long can six companies hold up the entire economy? Maybe a while.
Week In Review
We began the week by downgrading AMD to neutral. We are big fans of the company, but we think their AI story is going to take longer to play out than we had originally expected. One of the big disconnects between the Street and the semiconductor market is the concept of time. It takes years to take a chip from design to first silicon, and then more time for customers to fully evaluate those chips. So it is possible for a company to design a compelling chip, but not see it actually hit revenue for four or six quarters. Purchase decisions can change over those timelines in unexpected ways.
The other big event for us was Broadcom’s results. The company reported generally in-line numbers, but on the call they revealed that they had signed on a fourth custom ASIC customer. This is big news, and as we discuss below, cements the credibility of the company’s opportunity in AI.
Increasingly, we see the battle for AI semiconductors are a contest between Nvidia and Broadcom. Any time Nvidia loses share today it is most likely to a hyperscaler’s internal silicon that Broadcom helped bring to market. The two companies offer divergent visions of the market. Nvidia offers a vertically integrated stack – they want customers to buy GPUs, networking, memory, cooling, cabling and entire racks from them. True, they let customers pick and choose, but they make it much easier to buy the whole stack. By contrast, Broadcom lets the customers pick and choose what elements they want. Broadcom offers much of the stack, but it is decomposed and customers get a much freer hand in choosing what parts to do themselves and what parts to buy from Broadcom. Broadcom’s approach is more complicated, and trickier to get right, but offers considerable advantages when it does go right.
Of course, the compute industry has been having this debate for decades. There are echoes here of the Mac vs Windows debate, or go even further back to the “Cathedral and the Bazaar” era moving from mainframes to PCs. Those contests ended up in very different places, and so it is not clear who will win this round. For the moment, the market is growing fast enough that there is room for both approaches.
How Much Life is Left in the AI Trade?
When we were listening to the Broadcom call last week, we experienced some very mixed emotions. For several months, we have held that the Street does not fully see Broadcom’s AI opportunity, and the call made that opportunity very clear. The company has begun work on their fourth custom ASIC customer (likely OpenAI). But as we listened and worked the numbers, we realized that next year around half of Broadcom’s revenue would come from these four customers. The company went from an afterthought in AI to a full-blown AI story over the course of an hour. As we wrote a few weeks back, everything is an AI trade now.
This gives us considerable pause. It is hard to not see AI as a bubble right now. It is the one growth vector holding up semis, and probably a lot of the rest of the economy as well. But all of that rests on the back of essentially six companies, who combined are going to spend over half a trillion dollars this year on AI-related capex. Push them too hard on their capex and it often feels like none of them are too clear on where a return on that investment will come from.
We have studied the early days of the AI boom, way back in 2022. The launch of ChatGPT showed major advances, OpenAI was able to break some fundamental barrier. Microsoft then jumped in, doubling down on their investment in OpenAI and ramping up their capex. Google soon followed. Amazon took a little longer but they got on the bandwagon soon too. Since then the Street has rewarded these companies for boldly investing to capture this “once in a generation shift in technology”. (And yes people talking about things like generations is a red flag. Maybe it’s a whole New Economy.) When we ask what happens when those companies stop investing, we have been told repeatedly that the Street will punish any company for not capturing the opportunity. But when we look back and ask what did Microsoft see that sparked this whole wave, it seems a lot like they imagined something bigger than what he have right now. The whole thing reminds us of the children’s book “A Fly Went By”, where a boy sitting by a pond sees a stampede of animals running away. He asks each one what they are running from, and they say they are running from the animal right behind them, until he gets to the last animal (spoiler alert) a cow that kicked over a can and spooked themself.
We are big believers in AI – the technology – we see how it can reshape much of software and the Internet and eventually the wider world. But no technology this big has ever gotten adopted in a straight line, there are inevitable bumps and pauses along the way. Put another way Webvan and Friendster had to die before we got Amazon and Facebook. It’s not that we are skeptical about AI, it’s just that we’re skeptical about six companies holding so much on their shoulders.
And it is not like we are alone in this concern. We hear this sentiment voiced all over the Street and the Valley. This is almost more concerning, as we know prominent industry analysts who openly talk about the Bubble and then go on to provide a full-throated recommendation for the purest of AI pure-plays. There is a lot of cognitive dissonance making the rounds right now.
This is not our first Bubble. We lived through the 90’s dot.com mania, the US housing market in the 00’s and a handful more in Asia. And one thing we learned through all of that is that the old adage holds true: “The Market can stay irrational longer than you can stay solvent.”
Ultimately, we see this playing out in a few different ways. First, someone could invent something incredible and widely adoptable. Maybe that is a sentient computer, or a fully generalized robot model, or some engrossing video generation app. Maybe OpenAI gets to train GPT 6 on Blackwell and breaks some other barrier. Something that causes everyone to rush right out to upgrade their computers and crushes the data centers and AI factories with usage. In which case this AI trade goes on indefinitely.
We actually think those apps are out there, but we do not think they are coming soon. Instead, we think one of the hyperscalers throws in the towel or greatly pares down it’s spending to go after a much narrower, better-defined application. Then the spiral reverses, as everyone wonders what spooked that company, and slows down their own spending. That could take a while, and until it happens the AI Trade could remain strong. Put simply, there is now a healthy dose of momentum driving the trade. Our best guess is that this can keep going for another year, beyond that it is too hard to predict. Stocks can continue to run in this environment, and we have our clear opinions as to who we think will do best in that.
If you like this content you should listen to our podcast.
