The hyperscaler firms are spending obscene sums on artificial intelligence (AI)-related capital expenditures, and the big question is whether the trajectory can keep pace in the coming years.
Indeed, it’s hard to tell where the AI spend from mega-cap tech will go from here, especially as large language models (LLMs) become more viewed as a commodity, rather than something to pay top dollar for with a monthly subscription. With the rise of open-weight and open-source models, perhaps the commoditization of AI is already well underway.
Either way, frontier labs and hyperscalers are going to keep investing in the cause, and while it’s hard to tell how this massive AI-driven run-up ends (especially when it comes to the heated AI chip stocks), I think that the “picks-and-shovels” plays remain the ways to benefit in the earlier innings of a AI infrastructure and data centre buildout that could last some number of years.
Source: Getty Images
The great AI buildout is expensive. Betting on the hardware plays has been the move
Now, it’s hard to project how long the boom will last and which companies will get paid first (and the most), but, in my view, playing the obvious “bottlenecks” isn’t a sure way to score big gains over the near term.
Instead, I think a proven firm that’s continued to innovate could be the way to do well over the long haul, especially as the AI infrastructure boom turns into an “intelligence explosion” bull case or something similar.
Indeed, much of the “picks and shovels” (think the chips, semiconductor equipment makers, foundries, and tool providers) is already going for top dollar, and there’s lingering concern over their cyclicality. Indeed, it’s easy to ride the boom to its peak, but it’s pretty tough to get out of the way when things reverse course, and the bust sets in as investors race to the exits on any given day or week.
Celestica
One innovator in Celestica (TSX:CLS), I think, could continue to do well, even as a part of the AI trade starts to show signs of going bust. Indeed, it’s hard to tell if this is the end of a boom or just another correction that will soon become a blip. In my view, Celestica is a firm that’s collecting as most others opt to spend (maybe overspend) on infrastructure.
At the time of this writing, the shares trade at a very reasonable 35.7 times trailing price-to-earnings (P/E). Looking into the next year, the shares look even cheaper at 26.4 times forward P/E. Undoubtedly, Celestica might be at risk of over-earning right now, as firms give the AI buildout their all. But that doesn’t mean the firm can’t keep over-earning.
As Celestica works hard on the next generation of hardware (think switches and servers) that go well beyond chips, I think that there are more big earnings years to come.
Of course, it all hinges on the state of the buildout, which may very well be in for a pause if energy bottlenecks don’t resolve themselves. Like it or not, Celestica is doing its part to alleviate network-side bottlenecks. As next-generation data centres look to go higher on bandwidth and lower on latency, it’s hard to imagine growth suddenly grinding to a halt, especially if you believe the AI buildout is still just getting started.
