June 5.

Why it matters

Two product implications. (1) The Form W Part A + modeling-data package every Texas developer must produce by July 24 is now a hard, dated artifact — same structure for every applicant, same modeling-year inputs, same minimum-firm-load disclosure. That maps directly onto the live-docket wedge: a 'Batch Zero filing assistant' that ingests project specs (location, peak MW, ramp curve, expected firm minimum), produces a draft Form W Part A, and flags missing inputs is the cleanest possible application of Cliff's value prop, with a 7-week window of urgency. (2) Pair this with the PUCT financial-security rule (16 TAC §25.194 — $50k/MW non-refundable) so a developer sees the ERCOT timing gate AND the PUCT cost gate in one underwriting view. The minimum-firm-load number a developer puts in their Declaration of Intent becomes the binding constraint on operations for years; under-disclosing leaves capacity on the table, over-disclosing locks in obligations that don't match real workload shape. That's exactly the kind of probabilistic-tradeoff judgment Cliff's approval-outcomes corpus should price.

Why it matters

Oregon and Wisconsin's tariff orders, plus Pennsylvania's model tariff (see next section), are now the three reference precedents every other state's next data-center tariff docket will cite. The variables that move between them are the same three each time — cost-allocation %, MW threshold, contract length — plus Oregon's new 90% take-or-pay floor. For Cliff's underwriting product this means every parcel-level economics view needs a 'tariff exposure' field showing the local utility's data-center tariff status (pending / approved / absent) and the implied LCOE delta against a baseline. The 30-year contract on big sites is the under-rated piece: it changes what a 'site' even is from a financial-modeling standpoint — closer to a perpetual ground lease than a development project. Sites that can be permitted in 24-36 months but require a 30-year operating commitment to the utility are now a fundamentally different asset class than a colo lease, and the underwriting view should highlight that explicitly. Likely next states to copy Oregon's emissions-linked interconnection language: Washington, Colorado, New Mexico, Minnesota.

Why it matters

The 'model tariff' format is the key insight to track because it is administratively faster — every PA utility now has to file a conforming tariff in a short window, which means PA's data-center tariff picture will be fully baked nationwide-fastest. Two implications. (1) Pennsylvania becomes the first state where a Cliff site-readiness view can show, with high confidence, what the tariff economics will be for any utility territory in the state — because they're all converging to the same template. That makes PA the cleanest state to build the 'tariff exposure' field against first and use as the canonical reference implementation. (2) Other states' commissions will copy the model-tariff PROCESS (not just the parameters), because it gets to a final answer in months instead of years. The likely next adopters of the model-tariff process: Ohio (PUCO), Indiana (IURC), Maryland (PSC), Illinois (ICC). Specifically watch for 'model tariff' language in any 2026 H2 generic docket in those states. Once the process spreads, the underwriting product can shift from per-utility-tariff lookups to per-state-template lookups across most of PJM and MISO, which is materially less manual work to maintain.

Why it matters

This is the single most consequential pending decision in U.S. data-center site readiness right now — bigger than any individual state action because it would set the floor that every state has to build on top of. Two product implications, conditional on the ruling shape. (1) If FERC asserts jurisdiction at 20 MW and standardizes the cluster-study process: the per-state tariff complexity Cliff has been mapping doesn't go away (retail rates stay state-level), but the interconnection-process complexity collapses to a roughly uniform federal procedure. That makes interconnection-timeline modeling a federal-data problem instead of a 50-state problem — much faster to build and maintain. (2) If FERC declines to assert jurisdiction or punts back to state-by-state: the patchwork wins, every new state docket is independent precedent, and Cliff's per-state coverage moat is locked in for the next 18-24 months. Either outcome is product-shaping, so watch the docket daily through end of June. The colocation piece is independently important regardless of the broader ruling — most behind-the-meter strategies in PJM territory (the recent Three Mile Island / Amazon, Susquehanna / Talen-Amazon arrangements) live or die on the colocation framework, and a clean FERC standard would unlock a wave of BTM deals that are currently sitting on hold.

Why it matters

Two implications for the wedge. (1) The state-by-state coverage problem just got more urgent: six new state-level frameworks in 30 days is faster than any per-state-by-hand mapping can keep up with. That argues for a generic 'framework digester' that ingests new state orders and extracts the same canonical fields (threshold MW, cost-recovery %, take-or-pay %, contract length, water/aquifer flags, local-authority-preserved-or-preempted flag, effective date) — the kind of structured extraction that Opus 4.7 handles well and was the whole rationale for productionizing LLM ingestion of regulatory text. Build that extractor now against the 6 new orders as the validation set. (2) The 'preserves local authority' clause in Florida and the parallel patchwork at the local level mean even with state-level rules, the parcel-level question 'will the county or city say yes' stays primary. The state framework sets the financial floor; the local government still has the kill switch. For Cliff's product the implication is that the underwriting view needs BOTH a state-tariff layer AND a local-political layer for every parcel — and the local layer is the one that's harder to systematize. NJ's union-labor requirement is also worth flagging in the buyer-archetype model: it disproportionately excludes financial-sponsor buyers (PE-backed projects underwrite to non-union construction cost) and disproportionately favors traditional hyperscalers (Microsoft, Google, Meta all have building-trades agreements already). Expect site-selection by NJ to skew hyperscale-only as a result.

Why it matters

Three implications for the jurisdictions data model. (1) Add two more enums to moratorium_template: regulate_instead_of_ban (Little Rock template — no pause, new ordinance issued) and county_lacks_authority (Brown County template — symbolic county-level action, real zoning at the municipality below). Both have very different developer implications than the original three templates. (2) Track 'retroactive effective date' as a separate field — Scarborough's retroactive-to-April-1 clause is the kind of detail that determines whether in-flight applications are caught or grandfathered, and missing it materially changes a project's go/no-go. (3) The Monterey Park 86%-yes voter ban (cited in the Governing piece) is the new high-water mark for ballot-driven bans — and California's initiative process makes that template portable in a way that council-driven templates aren't. For the longer-term jurisdictions roadmap, flag any city with an active anti-data-center grassroots organization in a ballot-initiative state (CA, WA, OR, MI, OH, MA) as a higher-probability ballot-ban risk than its current council posture would suggest. The Little Rock 'ordinance over ban' template is good news structurally for developers — it suggests cities CAN settle on durable rules under political pressure, which is the long-run outcome the industry needs.

Why it matters

For the prospect list this means three concrete adds. (1) Vantage Data Centers — they are the canonical contracted-builder for this new buyer archetype; they need the same risk-quantification Cliff sells to PE buyers because their delivery commitments to OpenAI are contractually-binding milestones with real penalty exposure. Right contact tier: site-selection / land-acquisition leadership. (2) IREN — Australian-listed but US-active, materially smaller than Vantage but moving faster on global expansion. Their next 3-5 US sites will be public in the next quarter; getting in front of the site-selection team before commitments is the highest-leverage timing. (3) The Stargate ecosystem more broadly — every named participant (Oracle, SoftBank, OpenAI, MGX, Vantage, Related Companies, Crusoe on some sites) is a node in the same decision graph. Outreach should go through 2-3 of those nodes in parallel rather than picking one. Pair with the Blackstone-Google JV outreach flagged May 24 (Benjamin Treynor Sloss / Sean Klimczak / Eli Nagler). The structural insight to internalize: the next 12-24 months of US data-center site selection will be dominated by this small set of multi-party JVs rather than by individual hyperscalers acting alone, which means winning ONE relationship in a JV cascades into participation in every site under that JV's program.