Two things: Bad breadth and an anomalous revisions trajectory.
Early Monday, I spent some time documenting resilient forward profit forecasts for corporate America. And also explaining why looks are deceiving in that regard.
Long story short, the bottom-up (i.e., cumulative sum of company analyst estimates) outlook for S&P 500 EPS has moved higher over the course of the war, but that’s entirely a function of upward revisions for Energy and Tech shares.
In other words, although earnings revisions are strong in aggregate, EPS revision breadth is narrow. Not surprisingly, that’s reflected in a Goldman metric of overall market breadth.
As the figure shows, breadth’s rarely been worse since the dot-com bubble.
Out of 1200bps worth of upside for the index since the late-March lows, Tech and Comms Services accounted for more than 850bps, according to Goldman’s Ben Snider.
That’s nothing new, of course. Bears have complained about Tech (and “tech”) dominance for nearly a decade insisting, at various intervals, that market concentration can’t possibly get any more extreme nor breadth any worse, and yet here we are.
Coming back to revisions, it’s well known that the sum total of company analyst EPS estimates is generally revised lower over the course of any given year. That dynamic’s been a fixture of bottom-up consensus for the entirety of my adult life, and while people tell me I’m not old, I ceased to be young at least three years ago.
There are some exceptions, though. So far this year’s one of them, as shown below.
That gives you some context for how the above-mentioned upward revisions to Tech and Energy EPS are impacting the overall trajectory of bottom-up consensus, leaving a stark divergence with the historical experience.
If you’re inclined to suggest that disparity likely presages sharp downward revisions going forward I won’t necessarily argue. But I will point you to the green line: Although bottom-up consensus was revised lower throughout 2025, the scope of the revision wasn’t nearly as pronounced as it is during a typical year.
As Snider put it, “the idiosyncratic impact of AI that has driven positive revisions creates the risk that the divergence persists.”
“As the figure shows, breadth’s rarely been worse since the dot-com bubble.”
You’re right, that’s not entirely new, but the degree is worth noting.
And how much of the EPS revision is driven by circular deals? Deals needed to juice forward modeling to justify the massive ($100s of Bs to Ts) Capex needed to buildout AI? Is the tail wagging the dog?
I think the thing that is currently getting lost in the AI “hype” (scare quotes because to-date the hype has mostly been realized in actual reported earnings) is the systemic risk building as more and more conglomerates of company P&Ls are reliant on other company’s Bal Sheets and P&Ls. These types of deals are often the source of late stage accounting shenanigans.
Thats in addition to the obvious potential that the ROI on such massive investments will likely not meet the current expectations in the timeframe needed. See the internet, eventually the ROI was achieved, just not before the collapse of it’s respective bubble.
We’re in a bubble (I believe the first of 3 waves of AI bubbles) but bubbles don’t burst because logic says they should, as H has reminded us many times.
So what is the actionable investing takeaway from this? Continue concentrating into Tech and Energy? Buy the names responsible for the bulk of the consensus upward revisions? Do nothing? Broad-based Index and chill?