Catch-up edition, Jul 8 → Aug 17.

Why it matters

Three implications. (1) The capacity price for the biggest US data center market is now pinned at whatever the cap is, indefinitely — there is no scenario in current supply/demand where it clears below. Hyperscaler 2028+ capex models should treat the cap as the floor, and treat cap-renegotiation politics (the collar expires after 2029/30) as the actual price-risk variable. (2) A price collar that produces a reliability shortfall is politically self-undermining: it delivered neither cheap power nor enough power, which is precisely what empowered the board to reach past the stakeholder process two weeks later. (3) The IMM's 'separate auction for data centers' concept — unthinkable two years ago — is now the moderate position in the debate. The radical position is PJM's own board letter (next item).

Why it matters

This is the single most consequential development of the window. The de-rate / firm-equivalent-MW framework this wiki has been building since April — nameplate MW vs firm MW vs curtailable MW as different products with different prices — is no longer an analytical lens; it is PJM's filed tariff design. Every PJM data center deal signed after June 2027 must now answer, in writing: what generation are you bringing, what curtailment are you accepting, and what does your firm-equivalent capacity actually cost? The same logic ERCOT wrote into SB 6 (co-located gen, curtailment hierarchies, PCLR elections) has now been adopted — independently, under different statutes — by the largest US market. Two consequences: (a) BTM generation stops being the exotic Texas option and becomes the default national architecture for new hyperscale load; (b) the arbitrage window between 'load that has secured supply' and 'load that hasn't' becomes the pricing spread that determines which projects live. Watch: FERC action on ER26-3380/ER26-3515 (comments Sept), the backstop auction clearing price in December, and whether the September governance ultimatum produces a PJM that states can steer.

Why it matters

The audit is the ERCOT load-realization thesis executed at governor scale. §25.370 already wrote 'actual historical realization rates' into rule (49.8% for data centers); Batch Zero already forced skin-in-the-game elections; now the state is manually marking 205 GW to market before studying any of it. Three reads. (1) For developers with real load, the audit is a gift: surviving it is the credential that separates you from 60%+ of the queue, and everything that proves load reality (contracted offtake, equipment orders, PCLR elections, financial security posted) is now underwriting collateral. (2) For the market, the published audit result — whatever fraction of 474 GW survives — becomes the single most important denominator in US power planning; every national forecast (BNEF's included) is currently built on unaudited queues. (3) The five-week arc from 'fastest process in the country' to 'stop everything and verify' is the same arc PJM ran in the same month, arrived at independently. Queue verification is now the national posture.

Why it matters

Texas and PJM converged on the same trade from opposite directions: PJM now punishes grid-served load that brings nothing (curtail-first), Texas now prices grid-served load honestly (12CP, pay-own-way) — and in both, the rational response for a hyperscaler with 2028 deadlines is the same: build your own generation. Amazon at 7.65 GW off-grid is the proof that the biggest, most conservatively-run buyer reached that conclusion. Watch two second-order effects: (1) gas turbine slots are the new interconnection queue — GE Vernova's equipment backlog hit 116 GW with capacity ramping to 20 GW/yr, Siemens is at a record €162B backlog, Mitsubishi at 35 GW delivering 2028–30; a 2026 slot reservation is now worth more than most queue positions. (2) The environmental-permitting fight follows the generation off-grid — EPA's July 16 memo exempting 'islanded' BTM plants from the Acid Rain Program (and the 60-day NOI environmental groups filed over San Antonio BTM plants six days later) is the opening exchange of the air-permit war this wiki's Layer 5 work anticipated.

Why it matters

The first statewide moratorium is a milestone the county-cluster analysis said was coming — but it arrived by executive order at 50 MW, not by statute at 20 MW, and that distinction is the practical takeaway. Executive moratoria are political instruments: reversible, negotiable, and calibrated to preserve the governor's optionality (Hochul can lift it for a project with the right labor agreement and ratepayer terms). Statutory moratoria are regime changes. Every developer's NY strategy now runs through a framework-development process Hochul controls — and every other Democratic governor now has a proven template that claims the political win without freezing the mid-market. Expect EO-style moratoria to propagate faster than statutes; watch whether A11560 gets signed under tour pressure (that would stack the statute on top of the EO and drop the threshold to 20 MW).

Why it matters

Virginia was the control case — the market where the industry was too entrenched, too tax-productive ($1B+ localities) to turn. It turned anyway, and the groundwater finding is the piece that can't be politicked away: east-of-I-95 water-cooled builds now face a state-agency finding that the water isn't there, which pushes new Virginia capacity toward air/liquid-cooled designs (higher power draw, worse economics) or out of state entirely. Meanwhile the by-right pivot (Amazon's Bridgefield) is the developer counter-move to watch nationally: where zoning already permits data centers, skipping the rezoning vote sidesteps the entire opposition apparatus — at the cost of inheriting whatever the by-right envelope allows. The QTS Digital Gateway public-notice-defect template plus Loudoun's moratorium track means the era of assembling 2,000-acre rezonings in NoVA is over; what remains is by-right infill, brownfield conversion, and points south and west.

Why it matters

Three structural reads. (1) The wave moving into top-15 cities kills the 'this is rural NIMBYism' frame — Indianapolis and Fort Worth are pro-growth, pro-industry jurisdictions in pro-data-center states, and Fort Worth doing it inside Texas during Abbott's own audit shows state-level enthusiasm no longer transfers down. The state-local divergence variable is now the dominant siting risk in every 'friendly' state. (2) The Calvert Cliffs abandonment is the first clean demonstration of the electoral kill chain — organize → flip commission → developer walks. With data centers running as an issue in 21 races this November, every project with a 2027+ entitlement date now carries election risk on its critical path, which is a fundamentally different risk type than a hearing loss (you can't out-lawyer an election). (3) The moratorium threshold convergence at ≥50 MW (Lakeland, Sarasota, NY EO 62, PJM's IRAS definition) — rather than the activists' 20 MW — is quietly the most important number in the wave: it leaves the sub-50 MW tier (enterprise, modular, edge) almost universally untouched. The small-modular thesis inherits an regulatory moat the big-campus world is building for it.

Why it matters

The geographic sorting the state-vs-federal misalignment predicted is now visible in the deal flow: hyperscale capacity is migrating to the two venues where local politics can't reach it — federal land (Paducah, Portsmouth, Savannah River) and off-grid ranchland (GW Ranch). Both venues share a profile: remote, gas-powered, transmission-adjacent, politically pre-cleared. The implication for the national map is a barbell — mega-campuses in politically-immune venues on one end, sub-50 MW infill on the other, and the traditional mid-size suburban grid-served campus getting squeezed out of existence in between. Note also what the federal government is NOT doing: it is not fixing transmission (NIETC cancellations), not preempting local zoning (no bill moving), and not resolving the RTO cost-allocation fight (show-cause likely extended into December). The federal play is additive capacity on federal terms, not a rescue of the conventional siting path.

Why it matters

The financing structure inverted the old constraint hierarchy. In 2024–25 the question was whether hyperscalers would keep spending; the Q2 prints and the private-credit machinery answer that spending no longer depends on their cash generation at all — Meta at $784M FCF is still guiding $130B+. What the machinery cannot do is conjure power, win a county election, or jump a turbine queue — so capital abundance plus physical scarcity means the binding constraints are now entirely on the atoms side, which is where all the pricing power migrates (power developers, turbine OEMs, entitled land, and anyone who can verify what's real). It also concentrates a new systemic risk: GPU-collateralized lending at $500B scale, funded by retirement money, exempted from ABS disclosure rules, priced on residual values Nvidia itself backstops — works if compute demand compounds; unwinds like telecom vendor-financing if it plateaus. The tell to watch is not capex guidance but depreciation schedules and the first secondary-market GPU repricing.

Why it matters

Hold both facts at once and the picture resolves: the pipeline numbers are inflated (Exelon, EIA, Abbott's audit) while the delivered-capacity market is the tightest on record (JLL, BNEF, absorption). The gap between the two is duplicate and speculative queue entries — the same phantom-load problem ERCOT quantified at 49.8% realization — finally being repriced by utilities, states, and forecasters simultaneously. Practical translations: (1) any project that can PROVE its load and its power is more valuable than it was in June, because verified projects inherit the scarcity premium the phantoms were diluting; (2) utility 'pipeline' disclosures bifurcate from here — contracted-with-collateral (AEP's $2B) versus letters-of-intent — and the market will learn to price the difference; (3) Southern's Camellia structure (3.2 GW with 1 GW flexible) is the demand-side template spreading fast: flexibility written into the ESA is becoming the price of interconnection, which is the commercial face of the same firm-vs-flexible MW distinction PJM just turned into tariff law.

Why it matters

Three reads for the long game. (1) The Oracle capitulation is the sleeper precedent of the window: the strongest-balance-sheet challenge to a state financial-security rule ended in voluntary dismissal, which tells every other PSC that collateral rules, minimum-take contracts, and credit thresholds are litigation-proof in practice. Expect the 24-state tariff architecture to harden and spread to all 50 — the cost side of every US data center pro forma just became a ratchet that only turns one way. (2) The midterm structure means the next legislative season (Jan–Jun 2027) is being campaigned on RIGHT NOW; whatever the November results, both parties will arrive with mandates to be tougher, and the 2027 sessions — Texas's three Abbott asks, New Mexico's statewide bill, NY's A11560 question — are where this window's politics become statute. (3) The litigation stack now covers the full facility lifecycle: siting (EIS challenges, Open Meetings suits), construction (nuisance-in-progress, Newsom), operation (noise torts), and even speech (the O'Leary defamation suit; Edgerton suing its own residents). 'Entitled' no longer means 'de-risked' at any stage — which is precisely why verified, litigation-screened, politically-cleared capacity is becoming its own asset class, priced above everything else.