Take PwC's own path and do the addition. Annual data center capex of about $800bn in 2026 rising to $1.8bn-times-a-thousand, or rather $1.8 trillion, in 2050 [7] is a compound rate of roughly 3.4% a year [3], and the twenty-five annual payments along that line sum to about $30.9 trillion [4], which is within touching distance of the $31.6 trillion headline [6]. The total simply describes the area under that first forecast rather than offering a second one. A 3.4% compounding series reads less like an acceleration and more like a boom PwC assumes has already happened by 2026, one that then drifts for a generation, which is a much more falsifiable claim than the round number suggests.
Geography concentrates it further. The US $15.1 trillion is 47.8% of the total and Asia Pacific's $8.2 trillion is 25.9% [8][1][2], leaving roughly $8.3 trillion, about a quarter, for everywhere else [2]. Spread across 2026 to 2050, the US line averages about $604bn a year [5], against an implied 2026 US figure near $382bn [6], so the forecast needs American annual spend to run about 1.6 times its opening level through the middle of the path [7].
The Beige Book channel is narrower than any of those numbers. What the districts actually reported was manufacturing orders tied to data centers and defense, and nonresidential construction increasingly pointed at data center work [4], which is concrete, steel, switchgear and site labour rather than the compute installed afterwards. The source material does not split PwC's dollars between silicon and shells, and that split is the entire transmission mechanism: capex can keep climbing while the construction content of each dollar falls, in which case the Chicago contact's construction recession [5] arrives with headline capex at a record.
There are a few ways to read this. Funding is one: if hyperscaler capex runs off operating cash flow rather than borrowing, and the material here does not say which, then a hike priced at roughly 65% [11] is close to irrelevant to the buildout and the rate leg of this story is decoration. Evidentiary caution is another: the Beige Book is qualitative reporting from twelve regional banks through August 24 [2], and one contact's sentence is not a national account. The reading I weight most sits inside Goldman Sachs Research's August 19 list, where improving server and model performance is filed alongside rising spending [10] even though better performance per chip is a reason to buy fewer of them.
This is probably wrong, but the fragility that ends this cycle looks more like power and permitting than the funds rate, and the tell will be a quarter in which announced capex holds while data-center construction starts flatten. It would prove me wrong if a 25 basis point move visibly changed project starts, which is testable within two quarters. One sourcing note, since it bears on the rate leg: the 65% hike pricing and the August 28 Jackson Hole line attributed to Fed Chair Kevin Warsh both reach us through a single publisher's account of Reuters [11][12], and anyone sizing exposure off that should read the primary text first.
