PA + OR + WI + FL / 14-day cluster

Why it matters

Five updates, layered. (1) The four-state cluster is now structurally different from the moratorium tape and needs a separate Cliff product surface. The moratorium tape (covered all week — ~80 jurisdictions, 50 active bans, 4 permanent) *delays* approvals: it says 'no new applications for X months.' The ratepayer-protection-tariff cluster *changes the math* on approved sites by shifting cost-allocation. A site in PJM territory that was underwriting at 'PJM-standard transmission upgrade cost-shared across all customers' now has to model 'developer pays unrecovered investment if canceled, plus 50 MW threshold trigger, plus universal-service contribution.' The cost delta on a 200 MW site can be tens of millions of dollars per year in operating cost, which is materially larger than the typical site-selection arbitrage between Indiana and Ohio or between PJM and ERCOT. Cliff's Layer-2 (incentive applications) and Layer-5 (submission-readiness) products both need a 'who pays for the substation / transmission upgrade / new generation' input that flips by state — and that input now varies by *PUC-jurisdiction-and-utility*, not just by state. PA has a state-level model + a utility-level rate class (PPL only — the four other PA distribution utilities haven't filed yet); WI has a single utility implementation (We Energies); OR has a single utility (PGE); FL has the statute but no PSC tariff yet. Cliff needs a per-utility tariff-status field, not a per-state field. (2) The 14-day cluster cadence is the new pattern. Before April 24, the only state-level ratepayer-protection-tariff actions were Ohio's pending AEP rate case (still in litigation) and Maryland's earlier Public Service Commission filings. The April 24-May 8 window saw four separate states move in coordinated direction — that's a regulatory pattern, not a coincidence. The likely explanation: PJM's December 2025 FERC order on co-located load + the public attention on hyperscaler capex (Q1 earnings April 29 disclosing $725B combined 2026 spend) created the political-economy conditions for state PUCs to move in parallel without coordinating explicitly. The moratorium tape continues to grow but is increasingly a leading indicator of state-PUC action: counties that pass moratoria (Bulloch GA, Sedgwick KS, Lordstown OH) are jurisdictions whose state PUC is the next likely-mover. Watch GA, KS, OH state PUCs over the next 60 days — if any of them move on parallel tariffs, the cluster becomes a wave. (3) The thresholds tell the planning story. PA + FL + WI converged on 50-100 MW; OR's 20 MW is a separate planning regime that captures a much larger universe of mid-size colocation. The 50 MW threshold is a clear hyperscale-only target — most enterprise/colocation is sub-50 MW. The 100 MW aggregate trigger is the 'campus' target — multiple ≥50 MW buildings on the same site combine. The 75 MW within-10-mile aggregate trigger PPL adopted is the 'cluster' target — a hyperscaler can't avoid the threshold by splitting one campus into two adjacent sites. Cliff's submission validator needs a threshold-by-utility lookup with both the individual and aggregate tests — and the 10-mile-radius test is a geospatial query, not a flat lookup. The Layer-5 product needs to add 'aggregation distance' as an input. (4) The exit-fee provision (PA + OR + PPL) is the new risk channel for site abandonment. Historically, a developer who canceled mid-buildout left the utility holding the bag for sunk substation / transmission costs (which the utility recovered through general rate cases — i.e., from all ratepayers). The exit-fee provision shifts that risk to the developer. The implication for cancelation economics: hyperscaler cancellation of a sited project is now a real-dollars decision, not just a sunk-cost-fallacy decision. The Microsoft / Aberdeen Scotland pullback (covered earlier) and the Stargate UK pause (covered Wednesday) would carry materially higher exit costs under PA/OR exit-fee provisions; the same dynamics will apply to any 2026-2028 cancellation in the four-state set. Cliff's risk module should add a 'cancellation cost under exit-fee regime' calculator. (5) The translation for the investor audience (software VC) for the Tuesday May 12 update: the cost of building a data center *in the wrong state* now varies by tens of millions of dollars per year, and the rule changes happened in 14 days. The framing for investors is that regulatory-cost-allocation has become a real-time variable, which is exactly the kind of change-tracking-at-cost-of-LLM-tokens that Cliff's regulatory-knowledge graph compounds against. Pull onto investor-updates/2026-05-12.txt under 'state-PUC tape.'