Begin with what a county board is actually being asked to approve. Virginia, the country's largest data-center market, gets one permanent job for every $54 million invested, against an economy-wide average of 17 jobs per $1 million of investment [12], which is about $59,000 per job and makes the data-center version roughly 918 times more capital-hungry per lasting position [1]. Indiana's arithmetic is friendlier and still nowhere close: the $10 billion Lebanon campus employs more than 4,000 during construction and keeps about 300 once the servers are running [10], or about $33 million per permanent job, some 567 times the broad average [2]. A typical facility supports fewer than 200 long-term local jobs, per the U.S. Chamber of Commerce [11], while drawing power comparable to a mid-size city [18]. A town that says no to that trade is not being irrational, and the jobs pitch is harder to make when nearly 200 economists warned in July of large-scale AI displacement [15] and Microsoft cut close to 5,000 roles in early July while still funding data centers [16].
The national ledger runs the other way, which is the whole tension. Compute infrastructure at roughly 1.4% of GDP in the first quarter of 2026, up from 0.7% [7], is doing about 28 times its weight if you set it beside the St. Louis Fed's finding that information-processing equipment accounted for 39% of GDP growth through the third quarter of 2025 [8][4] (two different series, so treat that multiple as a direction rather than a measurement). Data centers' dollar contribution to growth has now passed consumer spending, a category that is normally two-thirds of the economy [9]. That asymmetry, national benefit against local cost, is why the federal posture is enthusiastic and the local one is 71% opposed, with a partisan spread of only 12 points between Democrats and Republicans [3][6] and a fight that CNN found skips national ideology entirely because it is local [14].
For an underwriter, the useful question is what the opposition can actually stop. The observable evidence is thinner than the noise: one governor's moratorium, endorsed by his party's Senate nominee [4], plus candidate messaging in Wisconsin [5] four months out from midterms in which millions are being spent to get the issue right [17]. This is probably wrong, but I would price it as a duration and jurisdiction problem rather than a volume problem, because entitlement timelines stretch, the premium on land and interconnection in politically quiet counties rises, and the megawatts still get built somewhere. Or rather, the more interesting version of the risk is fiscal: if the states now each forgoing more than $100 million a year in abatements [13] start clawing them back, the hit is not local siting delay but a repricing of after-tax returns in every jurisdiction at once [5].
What would prove the siting thesis wrong is simple and testable. If the next GDP prints show compute's share climbing past 1.4% while the Gallup number sits at 71%, then local opposition is a headline rather than a cost of capital, and the discount I am arguing for is money left on the table.
