Big tech is investing millions in data centers and saving a quick tax buck while doing it—leaving some states collecting revenue loss

The market should treat this as a margin-transfer story, not an AI demand story. Hyperscalers and the infrastructure vendors that sit one step removed from them keep the upside, while the real losers are state budgets and local ratepayers absorbing the externalities; the second-order effect is a higher political cost of permitting in dense markets, which raises execution risk for marginal projects but does little to stop the largest balance sheets. Smaller colo operators are more exposed than the megacaps because they have less siting flexibility and less leverage to negotiate incentives.

The near-term catalyst is not earnings, it is policy drift: disclosure of foregone tax revenue, municipal backlash, and election-cycle scrutiny can turn into tighter eligibility rules over the next 1-3 quarters. If that happens, project timelines slip before capital commitments do, and the first-order hit shows up in delayed equipment orders rather than lower hyperscaler capex guidance. Falsifiers are simple: if states continue expanding abatements or if hyperscaler build plans reaccelerate despite narrower incentives, the thesis fades.

Structurally, higher AI power density makes electricity and grid interconnect costs a larger share of total project economics, which favors vendors that reduce watts-per-compute and improve power delivery. That argues for a pick-up in spend on electrical gear, cooling, and backup systems even if local tax subsidies get trimmed. The contrarian miss is that the subsidy debate may overstate the public-policy risk for mega-cap cloud, but understate the concentration risk in a handful of states where the next round of builds can still be slowed by permitting and utility constraints.