While editorializing around Goldman’s shiny new price target for the S&P on Wednesday, I noted that according to the bank’s estimates, AI infrastructure investment’s likely to account for around half of this year’s blockbuster earnings expansion.

That’s hardly surprising, but it’s worth dwelling on for a beat or two given the sheer scope of what the firm now expects for aggregate index profit growth — a staggering 24% this year.

On several occasions in the lead-up to Q1 reporting season, Goldman suggested the Street was far too conservative regarding hyper-scaler capex projections. That was really saying something because… well, because those projections weren’t exactly muted. But the bank was right. Have a look:

Headed into earnings, consensus saw $673 billion, $790 billion and $815 billion in hyper-scaler outlays for this year, next and 2028, respectively. Fast forward two months and those forecasts are $754 billion, $905 billion and $918 billion.

If you ask Ben Snider, the Street’s still underestimating the magnitude of AI spending. “We believe there is upside risk to consensus capex estimates in 2027,” he said, adding that analysts “have been too conservative” for three years running, and hyper-scaler revenue backlogs “point to a continued supply/demand imbalance.”

All that spending’s good news for semis, which Snider described as “the primary direct earnings beneficiaries of the AI investment boom.” The chart on the left, below, illustrates the point made here at the outset — i.e., that more than half of this year’s expected earnings growth is concentrated among AI infrastructure firms.

The table on the right gives you a sense of how concentrated this is. Between them, Nvidia and Micron are expected to account for nearly third of this year’s S&P 500 EPS growth by themselves, and more than a quarter of next year’s growth.

Let that sink in: Just two companies are responsible for 32% of aggregate index profit growth. In the same note, Snider highlighted (again) the sharp inflection — some 30% — in forward earnings expectations for AI infrastructure stocks since the start of the Iran war.

That’s a critical point. As detached from geopolitical reality as the surge from the late-March SPX lows might be, it’s hand-in-hand with earnings optimism.

The figure on the left, below, shows you the upturn in forward EPS estimates for AI infrastructure firms beginning in mid-March, which is to say just prior to the war lows.

The figure on the right’s familiar: It shows the 90-degree inflection in out-year revisions breadth, which tracks very well historically with rolling three-month equity returns.

None of that’s to say the rally’s on solid footing. Indeed, you could argue it says precisely the opposite: That everything hangs on the AI (pipe?) dream. But it is to suggest markets haven’t completely lost their minds.

We may be living and rallying “on a wing and a prayer,” as the saying goes, but that’s not the same thing as the sort of totally baseless speculation that defined the dot-com boom and, episodically, the go-go days of 2021.

Speaking of 2000 and 2021, those are the only two years going back a century during which a Goldman composite metric of equity market valuations, concentration and recent returns was as overextended as it is currently.