PJM / capacity market overhaul / NCBL

Why it matters

Five updates layered. (1) PJM is now openly de-prioritizing data center reliability versus residential load, which is the single biggest structural change to data-center underwriting in the PJM footprint in 20 years. Historically, PJM has maintained one reliability standard — 99.97% — for every customer class, with the implicit promise that data centers got the same uptime as hospitals. The NCBL proposal explicitly breaks that promise. The technical reliability degradation is non-trivial: 98% uptime equals ~175 hours/year of curtailment risk, which is materially worse than even on-site diesel-backed generation (which typically targets 99.9%+ uptime). For an AI training cluster running a multi-month training run, 175 hours of unplanned curtailment in the middle of training is catastrophic — it doesn't just delay the run, it can void months of compute work depending on checkpoint frequency. This is not a number any AI hyperscaler can accept on a flagship training site, which means Framework 2 effectively forces every PJM-territory hyperscale developer into Framework 2(c) BYOG — bringing their own generation. Cliff's BTM-generation product surface (Layer 4 in the competitive-landscape map) is now structurally favored by the PJM proposal in a way it was not 30 days ago. Update strategy/competitive-landscape-and-adjacencies.md with a 'PJM NCBL favorable' tag on the BTM SKU. (2) The 50 MW NCBL threshold matches the four-state ratepayer-protection-tariff cluster (Pennsylvania PUC model tariff 50 MW individual / 100 MW aggregate; Florida SB 484 50 MW; PPL Electric 50 MW individual / 75 MW aggregate within 10-mile radius; Wisconsin We Energies 100 MW). The convergence at 50 MW across multiple PUCs AND the ISO suggests an emergent regulatory consensus that 50 MW is the threshold above which a new customer is treated as 'systemically significant' and therefore subject to different rules than commercial / industrial. This is exactly the kind of convergence that the regulatory-knowledge-graph product is built to capture — the per-utility-per-jurisdiction tariff lookup needs a 'systemically-significant-large-load threshold' field, and 50 MW is now the default value in nearly every PJM and PJM-adjacent jurisdiction. (3) The BYOG pathway in Framework 2(c) is a clean validation of Cliff's BTM-generation product wedge. PJM is now telling data center developers explicitly: if you want to interconnect in the PJM footprint and you're above 50 MW, your default option is to bring new generation that energizes with the load. That is the exact economic question the BTM SKU underwrites — gas turbine vs. CCHP vs. solar+battery+hybrid vs. SMR (when SMR becomes commercially available 2030+), and the math depends on the local gas pipeline access, capacity-factor expectations, REC pricing, and tariff-class treatment of self-generation. The BTM-generation page (strategy/btm-generation-economics.md if it exists; if not, create it) needs a 'PJM NCBL Framework 2(c)' section that walks through the economics on a 200 MW hyperscale site under each of the three sub-frameworks (curtailment-only, diesel-backup, BYOG). (4) The FERC filing path is the procedural hook for Cliff's Layer-5 live-docket product. Once PJM files the NCBL tariff at FERC (target Q3-Q4 2026), it becomes a federal docket with formal stakeholder comment process, ALJ proceeding, and final order — exactly the kind of regulatory-text-ingestion-and-summarization workload Cliff's product is built for. The pjm-er26-1088 page should be expanded to cover the parallel NCBL docket; the two will run on different procedural tracks but will inform each other (NCBL provides the tariff structure for new large loads; ER26-1088 provides the co-location structure for new generation behind the load). The intersection is the AI-data-center-with-BTM-gas configuration that hyperscalers are now sizing for. (5) The press framing is the political ceiling, not the technical text. Modo Energy's careful read of the document confirms PJM does not explicitly prioritize residential over data center load — but the Daily Caller / yournews May 10 framing characterizes it that way, and that framing will dominate state-PUC reading and political reaction across the seven PJM states with concentrated data-center activity (VA, MD, PA, NJ, IL, OH, NC). The implication for Cliff's risk module: the political-economy ceiling on PJM-state data-center entitlement is now closing faster than the technical regulatory text alone would suggest. Sites in early-stage approval in any of these seven states should be flagged as 'press-cycle-vulnerable' through Q3 2026; sites with sub-50 MW initial buildouts but ≥50 MW total campus size should be flagged as 'aggregation-risk' under the 50 MW threshold convergence. Translation for investor-updates/2026-05-12.txt: the single biggest grid operator in the country just released the strategy document that tells data center developers they will pay more, get less reliable service, or bring their own generation. PJM did not say 'PA-style 50 MW tariff' or 'OR-style 1¢/kWh surcharge' explicitly, but the structural direction is identical to what four state PUCs separately proposed in the prior two weeks, which is the kind of cross-jurisdictional alignment that's only legible in real-time when you're tracking the regulatory text at LLM-ingestion cost.