Murray sees additional complexities beyond the AI buildout. He notes that China is no longer exporting deflation. At the same time, countries have become focused on energy security, resource security, and domestic manufacturing. All of that, he says, should support resource extraction and energy names, which could imply some greater home bias among Canadian investors. While all those factors contribute to this shifting theme, Murray emphasized an understanding of the AI buildout as a route to finding companies with pricing power.
How data centers dictate pricing power
Before the rise of AI, with its huge infrastructural and resource requirements, big tech companies were defined by their capital light businesses with a high degree of pricing power of their own. As that field, once called the FANGs, then the Magnificent Seven, winnowed down to the ‘hyperscalers’ of Microsoft, Google, Meta, Amazon, and Oracle the big tech names began to shift. In building AI infrastructure, these companies began allocating all of the massive free cash flows that once defined them, going so far as to issue debt to finance this buildout. Now, Murray explained, those hyperscalers don’t have pricing power, the companies they buy from do.
That dynamic has created a series of rolling bottlenecks. The first, and most prominent, was in Nvidia’s graphics processing units (GPUs), but that was followed by a shortage in memory that sent memory manufacturers’ stocks soaring. Now we’ve seen a focus on CPUs, which has powered Dell stock to new heights.
What Murray believes investors need to look for in this environment is the so-called “golden screw,” a crucial component made by only a few companies in scarce supply. They need to identify those golden screws before the rest of the market does so they can participate in an eventual pop. At T. Rowe Price, the team has set up an internal map of a data center to try and identify what will be the next big item in demand. They use that map to work out the components that could be in short supply soon and identify the companies and sectors that can benefit when that shortage occurs.
Murray acknowledges that there are risks to a focus on the AI buildout. The first is that AI infrastructure is eventually overbuilt, as we saw with the construction of railroads, electricity, and fibre optic networks. While he says there are concerns about AI infrastructure eventually exceeding demand, Murray says there are no signs we’re near that point yet. There are also political risks associated with the infrastructure buildout, as communities increasingly call for the utility usage and energy costs associated with this buildout to be borne by the companies building and operating these data centers. While that may add to the cost of AI, Murray doesn’t see that pushback slowing the trend down in the immediate future.
