Are the hyper-scalers cheap?
That might seem like a ridiculous question. It’s tempting to shout, “No. Obviously not!”
But the intuition that tells us mega-cap US tech perpetually trades at rich multiples isn’t always right.
Currently, for example, the four companies that claim membership in both of the market’s most exclusive clubs — i.e., the association we call “hyper-scalers” and the “Magnificent 7” — trade at just 24x on a forward multiple.
The figure above, from Goldman, gives you some context. (Mind the double y-axis. The hyper-scaler multiple’s plotted on the left.)
24x is near the cheap-end of the 10-year range. In fact, it’s consistent with where the group traded in 2022, when each of the four suffered egregious drawdowns from their November 2021 highs.
Of course, part and parcel of that’s a dramatic upswing in profit expectations, which is to say forecasted earnings growth. I’ll repeat my ad nauseam warning. Projections are just that: Projections.
With that caveat, the figure below shows you the extent to which the hyper-scaler rally hasn’t kept up with the most recent leg higher in forward earnings estimates.
By contrast, chip stocks have outrun the upturn in consensus forward EPS estimates for semis.
What explains hyper-scalers’ “failure” to rally commensurate with the scope of the inflection in analyst profit forecasts? One answer is ongoing free cash flow worries amid the AI capex binge.
“The persistent gap between earnings and cash flows for some of the largest tech stocks should continue to weigh on their earnings multiples,” Goldman’s Ben Snider remarked, noting that although the hyper-scalers’ forward P/E multiple looks quite reasonable, their FCF multiple’s daunting, where that means “greater than 150x.”
The figure above shows you the impact of hyper-scalers’ cash burn on the FCF multiple for the 10 largest S&P stocks. For three decades, P/E and FCF multiples for the biggest names moved virtually in lockstep. Now they’re detached completely.
“While last quarter’s reports appeared to give investors incremental confidence in the likely returns to hyper-scaler capex spending, the likelihood of continued capital intensity in coming years suggests uncertainty around long-term future free cash flows will remain elevated,” Snider added.
I also wonder if there is a limit to how big these companies can get. Obviously, there is some natural nominal growth, but where is the Eddington limit for companies that already make up a huge proportion of the global economy? How soon (or long) until we see the first $10 trillion company?
I really know little about the math of this question. However, my observations over time seem to indicate that there are natural limits to just how big a firm can get and still be manageable. I’m nearly 82 and still old school. Buffett published his take on size a while back when he pointed out that the largest company in the world was Berkshire Hathaway. Size in this case is not market cap — I consider that metric to be fake news. Berkshire is the only company with more than a trillion in owned assets, ie cash, buildings, inventory, etc. = actual real physical and monetary assets. Not silly market cap. What we see right now is that its very hard to own and manage a trillion dollars of the real stuff.
Great post, H. Dayjob, I’m not sure there is a limit if everyone who runs their software runs it on the cloud. AI is creating an explosion of code running on the cloud. I think it expands not unlike our universe.
I wonder if we will hit a physical limit at some point, or psychological, or a great filter of some sort, or if we can start generating energy and resources from space.
All this ‘code running’, what does it mean for us humans? Will it make things “better” for us or better for a few global behemoths.
On that point, query your favorite LLM about “on-premises AI”
SMBs will likely remain as cloud users for a while, but larger users appear to be slowly migrating back to using their own servers. IF that plays out, it would be a remarkable move back in time to the pre-cloud and only cloud days.
(It would also vindicate IBM’s hybrid cloud offerings which were widely ridiculed.)