Cliff Daily — May 18.

Why it matters

Three updates. (1) The 30-day cluster from earlier this month has now settled into a clean snapshot: PA (regulator-issued model tariff May 8), FL (statute signed May 7 with PSC implementing tariff), WI (We Energies VLC tariff PSC-approved April 24), OR (PGE Schedule 96 effective June 10), and OK (statute signed May 11 effective July 1) all have ratepayer-protection regimes in force or imminent. Colorado opted out by killing both bills (left to PUC). Virginia layered a statutory SCC directive on top of the existing Dominion GS-5 tariff process. The site-readiness map for any 2027+ in-service data center now needs an 'in-state ratepayer-protection tariff' field with values {statutory + tariff (FL, OK), regulator-issued tariff (PA, WI, OR), statutory directive on top of tariff (VA), no statutory action / PUC discretion (CO, most other states)}. (2) The Colorado death is the most actionable data point for Cliff's investor / GTM narrative this week. The takeaway is not 'ratepayer protection is becoming universal' — it's 'ratepayer protection is becoming heterogeneous across states, with structurally different statutory + tariff layers in each.' That heterogeneity is exactly the per-state-LLM-lookup workload that Cliff's regulatory knowledge graph is built to handle. Update strategy/competitive-landscape-and-adjacencies.md to reflect that Colorado is now PUC-only for the foreseeable future, and that the legislative-vs-tariff split is becoming the dominant structural dimension. (3) Virginia's signing is the most operationally relevant for any PJM-zone site being underwritten today. The SCC directive language is enforceable by the SCC immediately, layered on top of the in-motion GS-5 tariff process, and Dominion has separately published a $50.1B five-year capex plan with $17B explicitly identified as data-center-driven. Any northern Virginia siting analysis that doesn't price in the GS-5 + SCC cost-allocation mandate is underestimating the developer's likely tariff burden. Add to the Virginia row of the de-rate calculator: 'GS-5 tariff in motion + statutory SCC directive from HB 1393 + SB 253 (effective on signature).'

Why it matters

Two updates. (1) Reno is the largest single-meeting public-engagement event for a moratorium fight this year. A 7-hour special meeting with 100+ speakers and a packed chamber, ending 6-1 in favor of immediate pause, is the kind of political signal that other Mountain West and West Coast city councils will read as 'this is a winnable political fight.' Cheyenne WY (Public Services Committee today May 18, second reading May 26), Denver CO (final reading today May 18), Seattle WA (introduced April 30), Sparks NV (the second-largest Nevada city, adjacent to Reno, not yet on the agenda), and Boulder CO (discussions) all become higher-probability moratorium-adoption candidates after the Reno result. The Mountain West cluster that Cheyenne started two weeks ago has now extended to: Cheyenne (in progress), Denver (today), Reno (first Nevada). The site-readiness map should now treat the Mountain West as a fully-formed third regional cluster alongside the Midwest (MI / WI / OH / IN / IL) and Sun Belt (GA / NC / FL / AL) clusters. (2) The 'pending moratorium' procedural form is structurally faster than the two-reading approach used in most Midwest cities. A pending moratorium takes effect at introduction (no waiting for second reading), which means hyperscalers cannot rush in applications during the procedural window between first and second reading. Other states may pick up this procedural pattern, which would shrink the application-rush window across the country. For Cliff's regulatory tracking: add a 'procedural moratorium type' field (values: pending-immediate-effect, two-reading, MPC-180-day administrative pause) to the moratorium tracker, because the application-rush window is a real operational variable for any developer trying to get in ahead of a moratorium. The Nevada 'pending moratorium' should be a separate value because of its immediate-effect mechanics. NV Energy's 10 GW Reno-Sparks-Tahoe data-center pipeline is the largest single-territory utility disclosure in the Mountain West — the Reno moratorium materially affects that forecast, and the June 1 final vote is the inflection point.

Why it matters

Three updates. (1) Today is the densest single-day moratorium-decision-day on the 2026 calendar so far. By tomorrow morning's press, four city councils will have publicly declared themselves on the data-center question, and a regional ISO compliance filing will have landed at FERC. For Cliff's regulatory knowledge graph, this is the kind of single-day burst that justifies an automated 'next-day refresh' pull on every active local-moratorium tracking row — the manual journal-style pattern will not keep up at this density. Recommend productionizing a daily local-moratorium-scraper SDK script (Bisnow + DCD + state-level capitols + the Good Jobs First moratorium tracker as input sources) and running it daily before the morning's investor-update digest. The cost-per-token math for that kind of scraper at Opus 4.7 is on the order of ~$5-10/day at current volume, which is well inside the budget for site-readiness underwriting. (2) The Elyria-Swansea framing in Denver is the first major-metro environmental-justice-driven moratorium in the 2026 tape. Most prior moratoria have been rural-defensive (small county, no nearby air-quality history, framed around water and rural character). Denver is different: urban, Latino-majority neighborhood, documented air-quality hot zone, adjacent to the National Western Center and I-70. If Denver passes today, the EJ-framing becomes a replicable template for moratorium organizing in other major-metro under-served neighborhoods near existing data-center sites (Atlanta, Memphis, Houston, Phoenix, etc.). For Cliff's site-readiness underwriting, add an 'EJ-vulnerability score' to the parcel-evaluation pipeline — proximity to an environmental-justice-designated census tract (per EPA EJScreen) should now be a material risk factor for any urban or peri-urban data-center site. (3) The PJM compliance filing today is the procedurally important one for the BTM (behind-the-meter) generation product line. FERC's April 16 order specifically allowed three new pathways: provisional interconnection for co-located load, service-below-nameplate capacity (which lets a hyperscaler use less than the full data-center load without forfeiting the interconnection), and accelerated interconnection process. Today's PJM filing operationalizes those three pathways into tariff language. Once it's filed, the 30-60 day comment window starts — and stakeholder comments will set the contour of how Cliff's BTM-economics SKU prices these pathway tradeoffs for any PJM-zone site. Pull the PJM filing from FERC eLibrary tomorrow morning and tag the docket number in the wiki — this is going to be a recurring cite for the BTM product.

Why it matters

Three updates. (1) The TAC tabling is the first procedural setback inside ERCOT's Batch Zero governance ladder. Until May 13, the Batch Zero package was on a glide path: PRS clear, ROS clear, TAC vote May 13-15, Board June 2-3. A TAC table compresses the remaining schedule and creates real risk that the June Board meeting will see either a still-contested package or no package at all. For Cliff's ERCOT-zone site-readiness underwriting, any Texas site evaluation that assumed Batch Zero in force by mid-2026 needs to slip its assumed-energization date by 1-3 months. This is the kind of inside-baseball protocol delay that doesn't show up in mainstream coverage but materially shifts the underwriting math — exactly the wedge for Cliff's regulatory knowledge graph. Pull the TAC meeting minutes tomorrow and add the specific stakeholder objection that caused the tabling to the knowledge graph (likely candidates: cost-allocation methodology for stranded-cost recovery, BYOG / co-located generation netting rules, or the interaction with SB 6's PUCT-level rule tracks). (2) The 410 GW interconnection queue is the single most important macro number for any US data-center site decision in 2026. Almost no analyst piece has correctly contextualized this: 410 GW of pending large-load requests against ~85-90 GW current peak demand means that the vast majority of those requests will physically not interconnect on the developer's planned timeline. The market structure ERCOT is converging on (Batch Zero + SB 6's five PUCT rule tracks) is fundamentally a triage mechanism — who gets to interconnect first, who pays for the upgrades, who is forced to provide co-located generation or BTM, and who gets curtailed. Cliff's de-rate calculator SKU should price ERCOT siting differently from PJM and MISO precisely because the queue ratio is so extreme: the 'expected interconnection timeline given queue position' is the dominant variable. (3) The ERCOT tabling cross-references the broader PJM compliance filing due to FERC today (different ISO, similar procedural dynamic). Both are large-load-interconnection rule fights happening at ISO/RTO level, both are moving on overlapping timelines, and both will produce final rule text that materially affects underwriting math for any new hyperscale site in their territory. Add a cross-ISO comparison row to strategy/competitive-landscape-and-adjacencies.md: 'ERCOT Batch Zero (PGRR145/NPRR1325, TAC tabled 5/13)' vs 'PJM colocation compliance (FERC April 16 order, filing due 5/18)' vs 'MISO 2026 IRP-driven large-load tariff (status TBD)' — the three are converging on different solutions to the same problem.

Why it matters

Two updates. (1) The Hyperion expansion confirms that the dominant hyperscaler power-procurement model for 2026-2030 is 'hyperscaler-funded utility-operated gas-fleet' — not nuclear (too slow), not behind-the-meter solar+storage (too expensive at scale), not full retail PPAs (too transparent on cost). The structural reason: gas plants take 2-3 years to permit and build vs. 10+ years for new nuclear, the hyperscaler can pre-finance the CapEx to bypass utility rate-base recovery delays, and the utility operates the plants (preserving regulatory familiarity and avoiding the BTM tax/permitting complexity). For Cliff's BTM-generation SKU, this means the addressable opportunity is structurally smaller than the 'all hyperscale data centers need BTM gas plants' framing suggests — the hyperscalers with deep balance sheets (Meta, MSFT, AMZN, GOOG, ORCL) will fund gas plants at the utility (not BTM); only the smaller / faster-moving developers (the AI startups, the colocation providers) will need true BTM. The BTM economics page should reflect this split: 'hyperscaler-funded utility-operated' vs 'true BTM (developer-owned, behind-the-meter)' are two different procurement archetypes with different economics, different cost structures, and different addressable customer bases. Cliff's BTM SKU is primarily aimed at the second archetype. (2) The 30% Louisiana grid expansion from a single project is the most extreme datapoint in the macro 'data centers are reshaping state-level grids' story. For comparison, ERCOT's 410 GW pending interconnection queue is enormous in aggregate but spread across thousands of projects; Hyperion is a single project doing 30% of one state's generation. The state-level political economy of single-project megafacility siting will be very different from the diffuse hyperscale-corridor pattern that defined 2022-2024 (Northern Virginia, Quincy WA, Loudoun County). Add to strategy/competitive-landscape-and-adjacencies.md: 'megafacility procurement model' as a separate analytical axis from 'hyperscale corridor' — Hyperion is the cleanest US example, and Tennessee TVA / Mississippi Entergy / Alabama Power service territories are the next-likely sites for the same pattern. Cross-reference: the Louisiana state-level preemption of local zoning is the structural reason this pattern works there and not in Wyoming or Michigan; states with similar preemption regimes (TX, OK, AL, MS, LA, parts of GA) are the geography for megafacility expansion, states without are the geography for moratorium fights.