AI data centers (AIDCs) are often described as electricity-hungry assets. This study treats AI DCs as gigawatt-scale capital builds and focuses on how a small set of REE-bound subcomponents enter the campus cost stack yet exert outsized leverage on delivery timelines. Generic criticality rankings and sector-level demand narratives rarely locate this dependence inside the campus cost stack, where procurement decisions and lead times are actually managed. An auditable bridge is developed from build economics to element-level exposure. A per-GW AI-campus CAPEX baseline is decomposed into a transparent taxonomy of buckets and subcomponents, then mapped to REE-bound components using documented physical mechanisms. Bounded cost-intensity parameters and a 0–2 schedule-at-risk score are used to discipline uncertainty and separate cost exposure from schedule vulnerability. These inputs are consolidated into a traceable Rare-earth Exposure (REX) metric and extended with forward-looking scenarios that jointly evolve CAPEX mix and intensity parameters under plausible architecture shifts and substitution pathways. In the baseline (mid) case, total mapped exposure is ≈ US$158.0M/GW, concentrated in IT equipment (≈ US$149.2M/GW; 94.4%) with a smaller cooling contribution (≈ US$8.8M/GW; 5.6%). Servers and accelerators account for ≈ US$138.8M/GW (87.8%), dominated by Nd (≈ US$83.3M/GW) and Pr (≈ US$34.7M/GW), with heavy-REE add-ons (Dy ≈ US$16.7M/GW; Tb ≈ US$4.2M/GW). Normalized, Nd exposure is 17.49 bp of total CAPEX. Cost share and delivery risk need not align: schedule-weighted exposure can be driven by long-lead, qualification-sensitive components. Scenario results show that design evolution reallocates exposure across elements and channels, revealing both mitigating and amplifying pathways that static assessments miss.

AI data-center (DC) growth is increasingly constrained by limited deliverable electricity, interconnection capacity, and cooling demand. This study develops a boundary-consistent screening framework for waste-to-energy (WtE)-coupled AI DC cooling, treating cooling as an energy service that can be supplied through grade matching rather than solely through electricity-driven mechanical chilling. The framework translates plant-side exportable heat into corridor-level planning objects by explicitly accounting for thermal attenuation, absorption-based conversion, and parasitic electricity associated with delivery and auxiliaries. Three results structure the analysis. First, a reference-case energy-service ledger shows how a representative regulated WtE plant with municipal solid-waste throughput of 1500 t/day and lower heating value of 10 MJ/kg yields ~78.1 MWth of exportable driving heat and, at a 20 km corridor, ~53.0 MWcool of delivered cooling and ~8.0 MWe of net avoided cooling electricity after parasitic debiting. Second, the coupled system is governed by operating regimes, not a single efficiency score. Under the baseline package, full thermal coverage is maintained up to ~20.9 km, the stricter quality-adjusted criterion remains positive to ~22.9 km, and the electricity–relief criterion remains positive to ~44.7 km. Third, deployment-scale translation for a 1 GW IT campus ($u=0.70$, $L=5 \mathrm{k}$m) implies a net grid relief of ~116.9–264.4 MW across scenario packages, while the required WtE footprint ranges from roughly three to 148 equivalent representative plants, or about 0.6–40 full-load-equivalent plants at a 25% displacement target. The contribution is a siting-ready planning framework that identifies when WtE-coupled cooling remains corridor-feasible, when it becomes hybrid and marginal, and when infrastructure scale rather than thermodynamic benefit becomes the binding constraint. It is intended as a screening tool for planning and comparison, not as a project-specific hydraulic or plant-cycle design.

The Paycheck Protection Program (PPP) was the largest targeted business support program in the United States, yet its firm-level effects remain contested. I link administrative PPP and SBA 7(a) records to a near-universe panel of U.S. employer firms from Dun and Bradstreet, covering roughly 30 million establishments, and evaluate short-run impacts on employment, financial stress, and commercial credit risk. To address non-random take-up, I combine propensity score matching with difference-in-differences on a balanced panel from March to September 2020 and exploit variation in loan holding duration. PPP receipt raises employment by about 0.07 percent on average but improves failure-risk and delinquency-risk percentile rankings by roughly 1.2 and 3.2 points, respectively, with longer loan duration strengthening all three margins. Heterogeneity analysis shows that small-to-medium firms and borrowers with intermediate pre-crisis risk experience the largest gains, while micro firms, very large firms, and highly stressed firms benefit less. Firms without prior 7(a) borrowing relationships realize particularly large credit-score gains. Overall, the evidence indicates that PPP functioned more as a balance-sheet and credit-risk backstop than as a powerful jobs program for the average treated firm. The results highlight how firm structure, pre-crisis financial health, and access to government-backed credit shape the effectiveness of large-scale emergency support.

This article examines how legacy lending relationships shape the allocation of emergency credit under severe information frictions. Using a novel dataset linking Small Business Administration (SBA) loan records with Dun and Bradstreet microdata for over 26 million U.S. firms, I investigate whether prior participation in the SBA 7(a) program acted as a gateway to the Paycheck Protection Program (PPP). Employing entropy balancing to construct a strictly comparable counterfactual group, I document a distinct dynamic evolution in credit rationing. In the program's initial "panic phase" in April 2020, banks relied heavily on legacy ties as a screening technology: firms with prior 7(a) relationships were approximately 29 percentage points more likely to receive funding than observationally identical non-7(a) firms. By June 2021, however, this insider advantage had largely vanished, suggesting that policy adjustments and extended timelines eventually mitigated the initial intermediation frictions. These findings highlight a fundamental trade-off between speed and equity in crisis response. While leveraging existing credit rails accelerates deployment, it systematically excludes informationally opaque borrowers. I discuss policy implications for designing future digital infrastructure to decouple verification from historical lending relationships.