Cliff Daily — May 6.

Why it matters

Three separate items worth adding to the wiki: (1) Stratos is the proof point that the off-grid / fully-BTM model can clear local approval at 9 GW scale when paired with an existing pipeline tap. Until this vote, the largest fully-BTM gas-fueled campuses were the Fleet Storey (350 MW) and Joule Utah (4 GW phased) templates already tracked in scenario-all-btm-product-surface.md and byog-byong-state-patchwork.md. Stratos goes 2.25x larger than Joule's full buildout and skips utility interconnection entirely. For Cliff's de-rate calculator and submission-readiness products, this is now the upper-bound case — meaning underwriting a hyperscale site no longer requires modeling utility interconnection at all if the pipeline-tap path is open. The submission-validator product surface has to learn to recognize 'pipeline-tap available' as a regulatory-risk-eliminating site characteristic, not just a regulatory-risk-mitigating one. (2) The Ruby Pipeline tap is the load-bearing infrastructure detail. Ruby was built in 2011 to move Wyoming gas west to California / Pacific Northwest markets; demand never materialized, leaving substantial unused capacity. The Stratos approval is the first observable case of a hyperscaler tapping orphaned interstate-pipeline capacity for behind-the-meter power generation. Identifying which other interstate gas pipelines have similar unused capacity (Rockies Express, Bison, Pony Express on the liquids side, Mountain Valley) is now a discrete data-set to build — that map would be a defensible product input for the ERCOT / PJM / MISO siting workflow and a strong fit for the regulatory-knowledge-graph moat described in strategy/competitive-landscape-and-adjacencies.md. (3) The MIDA tax structure is the specific incentive being banned by NC HB 1004 (yesterday's lead). Utah is moving in the exact opposite policy direction — granting MIDA-mediated tax abatements at 80% rebate for the state's largest-ever single project — while NC is pushing to forbid local incentives outright for any ≥40 MW data-center load. The state-by-state divergence is now wide enough to model as a real geographic-arbitrage axis in the calculator: 'pro-incentive states' (UT, MS, IA, MN, OK) versus 'anti-incentive states' (NC if HB 1004 passes, possibly WI, MI). For Cliff's GTM, the divergence creates a different sales motion in each cluster — UT-cluster customers want speed-to-shovel; NC-cluster customers want litigation-defense and statute-validation.

Why it matters

Two items for the financing layer of competitive-landscape-by-layer.md and the green-loan-data-center-financing.md page suggested yesterday in the EdgeCore $1.5B / 114 MW writeup. (1) Per-MW debt anchor: $13B / 1 GW = $13M/MW for a single-tenant Meta-occupied AI campus. That's almost identical to the $13.16M/MW EdgeCore Sterling NoVa anchor from yesterday — meaning the two largest hyperscaler-site debt deals in the same week converge on roughly the same per-MW number for fully-leased / fully-occupied AI campuses. Cliff's de-rate calculator should now treat $13M/MW as the empirical anchor for senior-secured debt on AI-targeted sites with a creditworthy tenant. The implication for marginal sites (lower-quality grid interconnection, regulatory-overhang counties, no firm offtake) is that they should price at substantial discounts to that anchor; the firm-equivalent-mw-underwrite.md framing now has a concrete reference point. (2) Single-site $13B is a structural shift in how AI data-center financing scales. Pre-2026, debt for individual sites topped out around $4-5B; the EdgeCore (two sites, $1.5B aggregate) and the JLL data-center loan-syndication market both sized below $10B for any single asset. A $13B single-site package means hyperscalers are now using individual data centers as financeable infrastructure assets in the same way utility-scale generation projects were structured in the 2010s power-build-out. The Sopaipilla deal also matters for ERCOT specifically — adding 1 GW of confirmed-financed AI load to West Texas tightens the LMP / load-realization landscape in the ERCOT West load zone and feeds directly into ercot-load-realization-scoring.md and the ercot-loi-math-onepager.md base case. Watch for whether the debt offering closes at the rumored ~$13B level or tightens — if the syndicate cuts the deal, that signals AI-debt market saturation; if the deal is oversubscribed, that signals the spigot is still wide open.

Why it matters

Three updates for byog-byong-state-patchwork.md NC section and a new entity page on Durham/Charlotte/Mecklenburg if the cluster keeps growing. (1) The 60-day-cap-then-extend-via-planning-commission mechanic is procedurally distinct from the typical 6-12 month council-issued moratorium. Cliff's de-rate calculator regulatory-risk module currently treats moratorium duration as the input — but the Durham mechanic shows that the underlying procedural pathway (council unilateral vs. council + planning commission joint) determines the practical ceiling, not the headline duration. NC HB 1004 (yesterday's lead, on-site generation mandate) layered on top of the Durham 60-day-then-12-month pattern means a single NC site can face: (a) the 60-day council moratorium (immediate freeze), (b) extension to 12 months via joint commission (medium-term freeze), (c) statutory on-site-generation mandate at 25% if HB 1004 passes (permanent design constraint). That's three discrete regulatory-risk inputs for the same site, each layered chronologically — far more than the single 'moratorium yes/no' input the calculator currently models. (2) Charlotte mayoral-veto contrast: a 6-5 mayor-deciding-vote against even holding a public hearing is a cleaner test of where local political leadership lands than a typical moratorium vote, because it isolates the mayor's view from the council majority. Mayor Lyles holds the line for now, but adding data centers to the May 11 agenda creates a subsequent decision point. If Charlotte's east-side application then clears May 18 council, Charlotte becomes the rare NC city refusing to moratorium even as Durham and Wake County pause — a real-time intra-state regulatory-arbitrage trade-off Cliff customers will need to model. (3) The compounding implication of Durham + Camdenton MO + Lonsdale MN + St. Charles MO + Sheboygan WI all moving in the same 48 hours is that the moratorium pattern is now spreading at roughly 1 city-per-day, well above the ~1 city-per-week pace observed in March. If that acceleration holds, by mid-June there will be 50+ active municipal data-center moratoria nationwide. The byog-byong-state-patchwork.md changelog cadence needs to drop from weekly to daily, and the de-rate-calculator-gtm.md should treat 'check moratorium-cluster status' as a built-in workflow step, not a one-off diligence task.

Why it matters

Today's PRS vote is the last stakeholder vote before the May 19-20 TAC, which is the de-facto industry-blessing step before ERCOT board approval — meaning whatever clears PRS today is structurally the rule that governs how ~119 GW of pending Texas large-load applications get triaged for the next 5+ years. This is the live wedge already covered in ercot-pclr-batch-zero.md and ercot-pclr-batch-zero-timeline.md, and Cliff's submission-validator product is designed against exactly this rule set. Three things to watch on today's vote: (1) Does PCLR get adopted as filed (LPC-firm + curtailable above LPC) or modified by stakeholder amendments? Crusoe and TEBA have filed substantive comments suggesting the LPC threshold and curtailment-economics math need revision. The submission-validator code path needs to know what the final LPC math is by close of business today — anyone who runs validation on July 23 (the day before the Declaration of Intent deadline) using the wrong LPC math will produce wrong scores. (2) Does the Year-6 deliverability deferral hold? The PGRR145 base case defers full-load deliverability to future transmission upgrades after the Year-6 mark, which lets ERCOT approve a higher headline GW number now while pushing the firm-MW reckoning down the road. If stakeholders strip the deferral language out at PRS, the headline 119 GW gets a haircut today and the Cliff calculator's firm-equivalent-MW-underwrite.md base case needs to retake its delta. (3) The PRS vote outcome is also the immediate read-through to ercot-puc-58317-sb6-implementation-workshop.md (Texas SB 6) and ercot-puc-58777-load-realization-docket.md — both PUC dockets reference Batch Zero by name, and any PRS-to-PUC inconsistency that emerges from today's vote becomes a discrete docket-arbitrage opportunity for sites that filed pre-Batch-Zero. Pull the meeting recording tonight; the ROS vote tomorrow (May 7) is the second confirmation point.

Why it matters

Two items to update on yesterday's Hermantown writeup. (1) The disclosed project numbers ($650M / $2B total / 1,500-2,000 jobs) anchor the negotiated-tax-abatement math. The proposed $80M abatement against $2B total spend is a 4.0% effective abatement rate — at the lower end of typical hyperscaler / state-incentive packages (Henrico VA averages 6-8%, Mount Pleasant WI was ~12% on the original Foxconn / now Microsoft package). If NC HB 1004 passes (yesterday's lead) banning local tax incentives entirely for ≥40 MW loads, sites in NC would face an effective +4-12% capital-cost penalty against MN / VA / WI peers — a real geographic-arbitrage delta worth modeling in firm-equivalent-mw-underwrite.md. (2) Three phases over 8-10 years is the disclosure that shifts the lawsuit math. If Google is contractually committed to a 3-phase / 8-10 year buildout, the 'sue-then-table' play (now demonstrated in MN / OH / IL within 30 days) carries a new cost: each year of litigation delay pushes Phase 3 commissioning past Google's AI-capacity-commitment window, which (per Google's own April 2026 capex guidance) is structured around a 2030 baseline. A 2-year delay collapses Phase 3 entirely. That's a real operational risk Google will price into its alternative-site portfolio — and worth tracking whether Google quietly accelerates a backup MN / WI / IA site as a hedge over the next 30-60 days.