All five weekday entries (Monday August 24 through Friday August 28) are present in the research log. One item is deduplicated across weeks rather than within this one: the Insurance Journal / Bloomberg reporting on AI load volatility (same August 12 URL) was already reported in full in last week’s digest, so it appears here only as a two-sentence carry for the one genuinely new data point in the August 28 entry — Gartner’s projection that power constraints will restrict 40% of AI data centers by 2027 — and is not re-reported as a new item. Two other cross-week overlaps are handled as movement rather than repetition: the NERC Level-3 alert on computational load was covered August 21 and reappears here only for the newly-specified FERC directive and its December 31 filing deadline; and the “1–2% data-center peak cut yields a 0.5–2.8% rate reduction” figure this series has carried as the Duke Nicholas Institute number is traced this week to its Columbia commentary treatment and paired with the underlying Brattle grid-utilization analysis. Several items carry pre-week publication dates (CAISO’s sub-LAP framework from August 5, the Columbia/Brattle commentary from June 23, CalChoice’s DERMS from March 18, the New Jersey BPU straw proposal from July 27, the DOE Section 202(c) orders signed June 30); they are reported on the day they entered the log with original dates shown. Two watch items this series has carried resolve this week; both are noted where they land.
Last week’s digest argued that demand-side flexibility had stopped being advocated into the resource stack and started being written into the instruments that govern it. This week supplied the sequel, and it is a harder one: the same megawatt of large-load flexibility is now available by two entirely different routes, and both were exercised inside the same seven days.
The voluntary route is Google’s. The company disclosed it has now integrated a total of 1 GW of demand-response capability into long-term contracts with U.S. utilities — new agreements with Entergy Arkansas, Minnesota Power, and DTE Energy, on top of earlier deals with Indiana Michigan Power and TVA — under which it limits or shifts machine-learning workloads to bridge the gap between near-term load growth and the multi-year timeline to build generation. That is the largest single corporate electricity buyer in the country arriving at the table with curtailable capacity as its opening offer, compensated, contracted, and bundled with 600 MW of Entergy solar and 300 MW of wind plus 400 MW of storage tied to the Minnesota Power arrangement.
The involuntary route is the Department of Energy’s. Energy Secretary Chris Wright signed Section 202(c) Orders No. 202-26-32 and 202-26-33 authorizing PJM to direct transmission owners to switch any data center or large load of at least 50 MW onto its own backup generators within 15 minutes of an emergency signal, as a last resort before rolling blackouts — triggered by a July 2 demand forecast of 166,147 MW, above PJM’s all-time record. PJM has not implemented the order and no facility has been moved. The value, as with any capacity product, is in the option rather than the exercise.
Set those side by side and the strategic picture is unambiguous. Large-load flexibility is going to happen either way. The only variable left is whether a utility obtains it through a negotiated, metered, compensated instrument that also builds a settlement record — or waits and receives it as a federal emergency directive that pays nothing, proves nothing, and produces no reusable capability. Google’s 1 GW and DOE’s 15-minute order are the same physical megawatt approached from opposite ends of the willingness spectrum.
Underneath both, the pricing rails went in. SEPA’s large-load tariff database now tracks 77 tariffs and service rules across 60 utilities — 51 approved, 26 pending, spanning 36 states, with 29 approved by state regulators in 2025 alone. New York’s PSC overhauled its Dynamic Load Management suite. Puget Sound Energy took Washington’s 10%-of-peak flexibility mandate to market with a live RFP to grow from 129 MW to roughly 500 MW. And a PowerLines analysis of 51 investor-owned utilities put planned capital expenditure through 2030 at at least $1.4 trillion, up more than 21% in a single year, with the argument that DER integration at the meter is “no longer a research project” but the next mandatory infrastructure layer.
The through-line: last week flexibility acquired legal standing. This week it acquired a price, a procurement channel, and a federally-backed alternative for anyone who declines to negotiate.
🔋 Energy Storage
Duke Energy’s 2026 Carolinas Resource Plan, filed August 17 with the Public Service Commission of South Carolina, proposes a 15-year portfolio of 18.5 GW of solar, 13 GW of energy storage, 14 GW of gas, and roughly 4.5 GW of new nuclear against a base-case winter peak projected to rise more than 10 GW — about 30% cumulative growth — and it names grid-edge demand-side programs a “core execution tool” rather than a sensitivity. The gas figure breaks down as 8.2 GW of combined cycle plus 5.8 GW of combustion turbines, which puts storage within a megawatt-scale rounding error of the entire firm-gas build — the third regional or single-utility plan in two weeks to scope storage as a first-order capacity resource rather than an energy-shifting supplement. Duke’s own language is the quotable part for a DSM proceeding: the company calls its “grid edge programs a core execution tool to reduce, shift, and shape demand through energy efficiency, demand-side management, load curtailment, customer programs, and storage demand response,” and retains a minimum 1%-of-load annual energy-efficiency savings target in the forecast while crediting rising demand-response capability. The execution signals are concrete rather than aspirational — an RFP for 400 MW of standalone South Carolina storage, secured battery equipment, GE Vernova turbine supply agreements, an approved 1.4 GW combined-cycle plant in Anderson County, and pursuit of 80-year nuclear license extensions. The tell to watch through the SC PSC’s expected April 2027 hearing and June 2027 order is whether the 13 GW of storage and the grid-edge DSM/DR portfolio are credited with capacity value that actually displaces some of the 14 GW gas build, or whether firm gas remains the default that the demand side merely supplements. The avoided-cost benchmark this series tracks applies directly: energy efficiency at roughly $20.70/MWh against $45–108/MWh for new gas combined cycle. (Source: Utility Dive, August 20 — August 24 entry)
PJM is reopening its Surplus Interconnection Service rules after the first reform produced eight applications and two approvals — against MISO’s ~14.8 GW and SPP’s ~14.3 GW of surplus requests, the majority of them battery storage. Surplus Interconnection Service lets a new generator or battery connect at an existing facility’s interconnection point using that asset’s unused capacity interconnection rights, skipping costly network upgrades, and it is described as the fastest available way to add megawatts to a constrained grid. PJM is reaching for it because it missed reserve-margin targets in its last two capacity auctions, with the shortfall growing to roughly 6.8 GW for 2028/29 from about 6.5 GW for 2027/28 — the same adequacy gap that large-load growth is widening and that DR, storage, and non-wires alternatives exist to help close. The specific fix on the table is a market-modeling change that matters for how co-located storage gets valued: PJM staff floated letting hybrid resources that clear the capacity market as a single unit participate in energy and ancillary-services markets as separate resources, untangling the single-market-ID problem that today blocks a battery added to a solar farm from reaching the capacity interconnection rights it needs. The headroom is substantial — UC Berkeley researchers estimate PJM’s existing thermal and renewable sites hold surplus interconnection capacity able to host roughly 150 GW of solar, wind, and storage, and many PJM solar projects use only 40–60% of nameplate capacity interconnection rights. Indiana and Virginia both passed 2026 laws directing utilities to study surplus interconnection, which makes behind-the-fence additions at existing plants a recognized regulatory tool rather than a vendor pitch. For utilities, unlocking underused CIRs is a materially lower-cost route to dispatchable capacity than greenfield interconnection, and it belongs in the same avoided-cost comparison as a new peaker. (Source: Utility Dive, August 21 — August 24 entry)
CAISO’s proposed sub-LAP “accounting change” could pull upwards of 2 GW of behind-the-meter batteries into the wholesale resource-adequacy market by assigning wholesale value to DER aggregations that reduce load within any of the ISO’s 20-plus sub-load-aggregation points. Advanced Energy United’s Brian Turner calls it “a small accounting change [that] could significantly change the battery market in California,” and the framing is accurate — the reform is administrative in mechanism and structural in effect. Today behind-the-meter batteries are effectively locked out of CAISO’s RA market because exported power’s deliverability is unproven, so they monetize on-site load offset during peaks but earn only retail net-metering or net-billing credit for anything past the meter, stranding capacity value the grid could otherwise bank. The design discipline worth copying is what CAISO declines to do: it preserves the fundamental definition of demand response as load curtailment, so aggregations wanting to become net exporters must still enter the generation interconnection queue. That sidesteps the measurement-and-deliverability disputes a full export-generation reclassification would trigger, and grows DER participation without relitigating what a demand resource is. The parallel CPUC demand-response rulemaking sharpens the stakes: it is weighing “bridge year” funding to keep the IOUs’ existing DR programs alive through 2028–2029, plus four longer-horizon questions on DR valuation methodology, CAISO market integration, RA valuation, and cost-effectiveness — targeted for Q4 2026 on the less-urgent items with authority running to February 2028. For DERMS procurement in the nation’s largest behind-the-meter battery market, the specification consequence is precise: an orchestration platform must map heterogeneous aggregations to sub-LAP geography and settle them against RA-grade performance rules, not merely enroll devices and report an aggregate. (Source: Utility Dive, August 5 — August 24 entry)
⚡ Virtual Power Plants (VPP) & Demand Flexibility
Google has reached 1 GW of total demand-response capability integrated into long-term U.S. utility contracts, with new agreements at Entergy Arkansas, Minnesota Power, and DTE Energy adding to earlier deals with Indiana Michigan Power and TVA — the clearest evidence yet that hyperscale load can be operated as a dispatchable grid resource rather than a rigid one. The mechanism is squarely on this series’ thesis: Google limits or shifts a portion of the machine-learning workloads running in its data centers, allowing demand response to bridge the gap between near-term load growth and the multi-year timelines required to build new generation and storage. That reframes the largest and fastest-growing load class from an adequacy threat into a flexibility asset — a facility that can shed or shift hundreds of megawatts on signal is precisely the curtailable capacity that defers peakers and, increasingly, accelerates interconnection, because utilities are writing flexibility into the interconnection agreement as the price of connecting faster. Read this against last week’s Southern Company item, where 1 GW of flexible demand response was written into a 3.2 GW OpenAI campus contract near Savannah, and a benchmark starts to firm up: two of the largest AI load commitments in the country now carry codified flexibility provisions, and a gigawatt is the unit both landed on. The bundled investments matter for how these deals get structured — a $25 million Energy Impact Fund, a 600 MW solar project with Entergy, and 300 MW of wind plus 400 MW of storage tied to the Minnesota Power arrangement — because they show flexibility being negotiated as one term inside a supply package rather than as a standalone concession. For DERMS and ADMS procurement, the operative lesson is that the orchestration and telemetry layer that verifies large-load curtailment, settling performance the way a DR aggregator would, is what converts these commitments from corporate goodwill into bankable accredited capacity. A gigawatt that cannot be measured to settlement standard is a press release. (Source: Data Center Dynamics — August 27 entry)
Puget Sound Energy’s live 2026 Demand Response RFP takes Washington’s statutory 10%-of-peak flexibility mandate directly to market, seeking turn-key programs to grow from roughly 129 MW of DR today to about 500 MW available year-round by 2030. Washington’s Demand and Load-Curtailment Utilities (DALCU) requirement obligates PSE to build a load-flexibility portfolio equal to 10% of historical peak in both winter and summer — or the maximum the WUTC deems technically and commercially feasible — and PSE’s preliminary math puts that at approximately 500 MW, roughly 10% of projected 2030 peak. That is a ~370 MW build-out in four years, and it is the cleanest regulated instance of this series’ avoided-cost framework in action: a dual-peaking utility procuring turn-key demand response at scale to displace both winter and summer capacity, the same deferred-generation-and-T&D logic under which a 70 MW DR portfolio can defer roughly $150M in capital at about $20M in program cost, and which prices DR capacity near $66/kW-year. Two things make this the item to watch in the Pacific Northwest. First, the RFP’s turn-key, at-scale framing is a DERMS-procurement tell in disguise — a 370 MW build-out over four years is impractical without an orchestration layer that can enroll heterogeneous load and settle performance-based incentives, which means the DR procurement decision and the DERMS platform decision are effectively one decision made twice. Second, PSE is a live test of whether a hard percentage-of-peak statutory mandate produces bankable dispatchable megawatts or merely an enrollment count. This series has argued for months that statutory peak-reduction hooks are what drive DR to scale — Washington’s DALCU, Virginia’s H.B. 429, Illinois’ S.B. 25 and CRGAA — and PSE is the first of those to reach the procurement stage where the claim gets tested against delivered MW. (Source: Puget Sound Energy 2026 Demand Response RFP — August 26 entry)
The New York Public Service Commission approved enhancements across the state’s Dynamic Load Management umbrella — CSRP, DLRP, Direct Load Control, Term-DLM, and Auto-DLM — aimed at expanding customer participation, improving operational efficiency, and better aligning demand-response dispatch with evolving system conditions. New York remains the reference case for a mature, tariff-embedded DR portfolio, and the design details are the tell for how bankable the megawatts are. The Distribution Load Relief Program dispatches on network contingency — when the next contingency would push Con Edison toward a Condition Yellow — which means the program is explicitly sized against deferred T&D and contingency capacity costs rather than a flat energy price. That is the avoided-cost comparison in its most defensible form, because the deferred asset is identifiable and locational rather than a system-average abstraction. Two structural parameters carry beyond New York: a 50 kW minimum load-relief threshold per participant or aggregator keeps the programs open to aggregated commercial load rather than restricting them to large industrials, and the state’s earlier move to a minimum three-year program term is what makes battery storage economic to enroll, converting incentive-driven signups into durable dispatchable capacity. That three-year term is the underappreciated lever — a one-year program cannot underwrite a capital asset, so program duration is a de facto storage-procurement instrument. For IRP and DERMS work, the point New York has been demonstrating longest is that the orchestration layer which enrolls heterogeneous load and settles performance-based incentives is inseparable from the program itself; you do not buy the tariff and then shop for the platform. (Source: New York Department of Public Service — August 28 entry)
Hawaii’s Public Utilities Commission opened Docket 2026-0084 to design a virtual-power-plant grid-services program across Hawaiian Electric’s territories, against an installed base of more than 1 GW of customer rooftop solar and roughly 410 MW of DER-attached batteries that the Commission’s own premise says are largely not optimized to support reliability or displace fossil generation. This is the orchestration-layer-as-prerequisite argument stated by a regulator rather than a vendor, and it is the most useful cautionary case available to mainland utilities. Hawaii has among the highest DER penetration in the nation. The physical resources are already deployed at scale and paid for. And the value now lives entirely in the platform and program design that convert an unmanaged installed base into dispatchable grid services — which is to say, the expensive part was never the batteries. The design lever worth tracking is compensation and participation structure: whether the program pays performance-based dispatch that makes the fleet a bankable capacity resource, or merely refines the existing scheduled-dispatch grid-services programs at the margin. That is the identical fork the GridLab California blueprint identified last week (compensation on verified grid performance, not enrollment) and the identical fork the New Jersey BPU faces in its straw proposal below. Three jurisdictions, one question. For DERMS procurement elsewhere, Hawaii is the leading indicator rather than an island curiosity: a jurisdiction that hit high DER saturation first is now retrofitting the coordination layer after the fact, which is materially more expensive than specifying revenue-grade telemetry, standardized participation, and per-device compensation up front. Mainland utilities approaching high DER penetration should price that lesson into the current DERMS spec, not the next one. (Source: Hawaii PUC Docket 2026-0084 — August 26 entry)
The New Jersey Board of Public Utilities released a Virtual Power Plant Straw Proposal under Docket QO26030099, opening the design process for a program that would turn home batteries, smart thermostats, EV chargers, and heat pumps into a dispatchable fleet ahead of a 2027 launch. The proposal responds to Governor Sherrill’s Executive Order No. 2 of January 20, 2026, which declared a statewide energy emergency and directed the Board to build a VPP program, and it sits inside a one-year transition plan preparing for New Jersey’s first VPP. The design lever this series tracks is compensation: the state intends to direct utility rebates to customers with grid-connected devices, and how those rebates are structured — flat enrollment payment versus performance-paid dispatch — will determine whether the resulting fleet is a bankable, dispatchable capacity resource or an enrollment statistic with a marketing budget. The item is a straw proposal in stakeholder process rather than an order, which is exactly why it belongs in a digest: this is the stage at which the avoided-cost value stack either gets built into a program’s architecture or gets left out of it permanently. For DERMS and ADMS procurement, a state building a VPP program from scratch is an unusually clean opening to specify revenue-grade telemetry, standardized participation rules, and per-device compensation before the tariff exists — rather than retrofitting them onto a program already carrying tens of thousands of enrolled customers. New Jersey, Hawaii, and California are all standing at the same design decision this month, and only one of them still has a blank page. (Source: New Jersey BPU, July 27 — August 25 entry)
🔌 DERMS & Grid Integration Technology
A PowerLines analysis of 51 U.S. investor-owned utilities finds planned capital expenditure of at least $1.4 trillion through 2030 — more than a 21% jump over the $1.1 trillion the same utilities planned a year earlier — with more than 30 of them naming data centers as a top growth driver, and the argument attached to it is that DER integration at the meter is “no longer a research project” but the next mandatory infrastructure layer. That is the capital backdrop against which every DERMS and ADMS procurement decision in this series now sits, and the framing matters more than the headline number. A 21% single-year increase in planned capex is not a spending trend; it is a re-baselining of what utilities believe they are obligated to build. The specific claim worth carrying into an RFP is that DER volume is outpacing existing integration toolsets — meaning the coordination layer is failing not because it was poorly chosen but because it was scoped against a smaller grid edge than the one that arrived. That reframes the orchestration platform as table stakes rather than an innovation line item, consistent with the procurement-maturity shift this series tracked last week, and it ties the spend directly to load growth: the same data-center-driven capital surge building generation and wires is what forces utilities to make the grid edge visible and controllable. The caution embedded in the surge is the part to act on. With utilities simultaneously pouring capital into utility-scale wind, solar, gas, and nuclear, DER can slip to a lower near-term priority even as the grid-edge integration debt compounds — and integration debt, unlike a deferred generation project, gets more expensive the longer DER penetration runs ahead of visibility (see Hawaii, above, retrofitting coordination onto 1 GW of already-installed solar). The procurement discipline that follows: specify the DERMS and coordination layer as a funded line inside the load-growth capital plan, not as a discretionary add-on competing against supply-side builds it will lose to. (Source: MarketScale — August 28 entry)
DERMS is moving out of the innovation budget and into the capital plan, positioned alongside ADMS expansions, substation automation, and AMI deployments as standing operating infrastructure — with the global market projected to grow at roughly 16% CAGR from about $557 million in 2025 to $1.36 billion by 2031. The architectural point in this piece is the one worth putting in an RFP verbatim, because it is the distinction most vendor conversations blur: ADMS is built for grid-wide distribution operations; DERMS provides grid-edge DER visibility, forecasting, and control; DERMS extends ADMS by sharing data and workflows rather than operating as a standalone silo. A utility that procures DERMS as a parallel system rather than an integrated extension buys a second source of truth about its own grid, which is worse than buying nothing. The operational rationale is straightforward — as rooftop solar, battery storage, and EV charging proliferate at the grid edge, utilities need live visibility into DER performance, forecasts of system impact, and the ability to adjust output, none of which ADMS natively provides. For this series’ thesis, the piece reinforces that the orchestration layer is the prerequisite to monetizing distributed flexibility: the same platform that gives operators grid-edge controllability is what converts an installed base of customer assets into dispatchable capacity for DR programs, VPPs, and Order 2222 market participation. The market trajectory frames the spend as real budget rather than pilot money. For vendor evaluation, the specification consequence is to require integration with ADMS/OT — GIS, OMS, SCADA — with revenue-grade telemetry specified up front, rather than retrofitting coordination after DER penetration outruns visibility. Note how consistently that requirement has now appeared: four proceedings converged on revenue-grade measurement last week, and both of this week’s DERMS items independently arrive at the same line. (Source: T&D World — August 27 entry)
California Choice Energy Authority is standing up a Lunar Energy DERMS to coordinate member communities’ smart thermostats, EV chargers, and battery storage as a virtual power plant, with full launch targeted for the end of August 2026 — meaning it should be live as this digest publishes. The notable design choice is sequence, and interim CEO Rob Johnson states it plainly: build “the foundation of the grid and the DERMS framework first,” finalizing contracts and technical roles before going to market so that enrollment is seamless, then roll out in stages beginning with integration of the already-deployed battery-storage base into a single unified platform. That is foundation-before-devices, and it is the opposite of the Hawaii pattern in this same digest — one jurisdiction specifying the coordination layer before enrollment, another retrofitting it onto 1 GW of installed solar. The two items read together are the cleanest natural comparison this series has been able to offer on sequencing cost. CalChoice is working with Lunar Energy under a February 2026 board authorization, and is tailoring participation rules and financial rewards per member community — a heterogeneity requirement that is precisely what makes the orchestration layer, rather than the devices, the hard part. The program is framed as a tool to shift load when power is most expensive, cutting costly energy purchases during heatwaves and high-demand periods: standard avoided-cost and non-wires framing, which is the right framing for a resource that has to justify itself to a board rather than a sustainability report. For DR business cases, the structural signal is that a CCA joint-powers agency is building its own DERMS — the aggregator layer is consolidating below the incumbent utility, and DERMS procurement now spans load-serving entities well beyond the IOUs. Utilities in CCA-heavy territories should assume the orchestration decision is being made without them. (Source: CalCCA, March 18 — August 25 entry)
Watch-item resolution — CAISO’s August 19 double. This series has carried an open verification item on CAISO’s August 19 activity for two weeks. The large-loads half is now confirmed: CAISO posted a technical-requirements straw proposal and held its stakeholder meeting on August 19, proposing two flexible interconnection services (FILI and FLIP) and retaining a 50 MW single-site threshold for a Large Load — the same threshold PJM adopted in ER26-3515 and DOE used in the 202(c) orders, which makes 50 MW an emerging de facto national definition worth aligning internal customer screens against. Written comments were due September 2, with the FERC compliance filing due November 16. The demand-response half of that day — the revised framework expected out of the Demand and Distributed Energy Market Integration working group — remains without public coverage and stays open into a third week. (Verified via targeted search of CAISO’s stakeholder notices and mgrid.org’s August 18 analysis; not sourced from the daily research log.)
🏗️ Data Centers & Large Load Growth
The Department of Energy has handed PJM emergency authority to switch hyperscale loads onto backup power: Section 202(c) Orders No. 202-26-32 and 202-26-33, signed June 30, 2026 on a June 27 PJM application, authorize PJM to direct transmission owners to move any data center or large load of at least 50 MW of peak demand onto its own backup generators within 15 minutes of an emergency signal, as a last resort before rolling blackouts. The trigger was a July 2 demand forecast of 166,147 MW — above PJM’s all-time record of 165,563 MW set in 2006 — which is exactly the tail-risk peak that demand response exists to shave. This is the sharpest federal signal to date that large-load curtailment has moved from a market nicety to a resource-adequacy backstop, and it deserves careful reading rather than alarm. PJM has not implemented the order and no facility has been moved to backup. The value is in the option, not the exercise — which is precisely how a DR capacity product is priced, and precisely the argument this series makes when a commission asks why a utility should pay for megawatts it rarely calls. For the avoided-cost case, the order is unintentionally excellent evidence: capacity that can be shed on a 15-minute signal is the same ~$66/kW-year DR value that defers peakers, only here it is being obtained by federal fiat rather than a voluntary compensated program. That is the uncomfortable part. A utility that has not built a negotiated flexibility instrument does not thereby avoid large-load curtailment; it simply receives the version that pays nothing, builds no settlement record, and leaves no reusable capability behind. For DERMS and ADMS procurement, the 15-minute, delivery-point-specific curtailment requirement is a telemetry-and-controls specification in disguise — certifying that a 50 MW-plus load can transition on signal requires the same measurement-and-verification capability a DERMS must supply to make voluntary large-load flexibility bankable. Build it once, use it for both. (Source: Utility Dive — August 28 entry)
Texas has roughly 300 large-load data-center interconnections frozen under Governor Abbott’s August 3 moratorium while ERCOT races to audit a 474 GW queue — about 90% of it data centers, more than five times the grid’s record peak — targeting a December 10 filing so interconnections can resume. At the August 20 PUCT open meeting, ERCOT officials committed to auditing the roughly 300 data-center proposals of 75 MW or larger now held in the “Batch Zero” process, with request-for-information templates going to provisionally qualified large loads from late August through September and further rounds into October and November. This closes a watch item this series has carried since the moratorium was announced: the audit has a scope, an instrument, and a date. The numbers are the argument. A 474 GW queue against a record peak near 90 GW is the phantom-load problem in its purest form, and experts flag much of it as speculative or duplicative — which is exactly why cost-causation regimes like PJM’s BYONC and Pennsylvania’s consent-order mechanism, and demand-flexibility mandates like Texas’ own SB6, have to exist before a utility commits capital against unverifiable demand. The verification discipline is now a formal state process rather than an intervenor’s request: a community-impact review reaching down to 25 MW crypto and data-center facilities, a refreshed baseline of about 157 medium sites representing ~8,800 MW, and 17 large loads totaling ~6.6 GW of peak sitting at the final energization gate. The second-order effect is the one that matters for planning everywhere else: officials warned the pause could push back ERCOT’s long-term load forecast and reliability assessment, and that the Batch Zero study will slip past its April 9, 2027 deadline. A grid operator that cannot publish a credible load forecast cannot run a credible resource-adequacy process — and every DR, storage, and non-wires business case downstream inherits that uncertainty. Texas’ SB6, requiring loads above 75 MW to accept demand-response and curtailment terms, and this audit together convert large-load flexibility from a courtesy into a gating condition for interconnection. (Source: Utility Dive, August 21 — August 24 entry)
Microsoft and PowerHouse Hillwood took their data-center service agreements to FERC in back-to-back filings, and the two disputes together show that FERC’s mid-June large-load show-cause orders are now being litigated agreement by agreement rather than settled in the abstract. In an August 21 filing, Microsoft told FERC that four amended Large Load Project Commitment Agreements and a Minimum Transmission Charge Agreement — negotiated by American Transmission Co. and affiliate Wisconsin Electric Power Co. without Microsoft’s input to serve its Mount Pleasant, Wisconsin campus — “predictably contain significant deficiencies” and lack any mechanism to keep WEPCo retail customers from paying for the facilities. Microsoft warned that the large load “could pay twice,” objected to an early-termination “windfall” and to a construction-work-in-progress cost-recovery mechanism, and asked for settlement-judge proceedings. Note who is making the argument: Microsoft is a signatory to the White House Ratepayer Protection Pledge, and the Wisconsin PSC separately told FERC the ATC agreements are “far from fully responsive” to the show-cause order to MISO. When the hyperscaler and the state commission both say the ratepayer-protection mechanism is missing, that is not a negotiating posture. Separately, PowerHouse Hillwood on August 14 urged FERC to reject ComEd’s “notice of cancellation” of a transmission security agreement tied to a planned 1.8 GW, $20 billion data center in Joliet, Illinois, arguing that Exelon’s non-pro-forma security agreements “reflect the utilities’ monopoly power” and that — unlike generator interconnection — no anti-competitive guardrails yet exist for large-load interconnection. The dispute, over security-deposit timing, has already produced a lawsuit and the cancellation of a retail service agreement pending at the Illinois Commerce Commission. The synthesis for utilities, DERMS planners, and IRP practitioners: transmission security agreements have become the primary instrument for weeding speculative load and allocating interconnection cost — the same mechanism Exelon credits for a “high probability” 40% haircut in its data-center pipeline, to roughly 11 GW — and with FERC’s RTO/ISO show-cause response deadlines extended to mid-November, the rules governing that instrument are still unsettled. Every large-load forecast built on unsecured agreements is a forecast built on a contract term currently in litigation. (Source: Utility Dive, August 24 — August 25 entry)
Regulators are pulling AI campuses into the same reliability-standard framework that governs generators, and the enforceable layer now has a date: FERC directed NERC in July to develop computational-load reliability standards and registration criteria, with filings due December 31, 2026. The reference event remains the July 22, 2026 Northern Virginia transmission disturbance in which roughly 3 GW of data-center demand vanished at once on a normally-cleared fault. NERC’s May 2026 Level-3 alert — its highest tier — called for improved computational-load modeling, instrumentation, commissioning, protection, and operational coordination, but it is guidance rather than an enforceable Reliability Standard; the FERC directive is what converts it. If finalized, large AI campuses would fall inside the same compliance framework as generators and transmission owners, with modeling, reporting, and coordination obligations and penalty exposure for noncompliance. The lineage is worth naming because it tells you what the requirements will look like: the ride-through and controllability expectations that FERC Orders 841 and 2222 and the NERC IBR reliability standards (full implementation January 1, 2030) impose on inverter-based generation are now being extended to large loads. For DERMS and ADMS procurement and for IRP work, the conclusion is that revenue-grade telemetry, ride-through settings, and dispatchable curtailment are becoming table stakes for interconnecting a large load at all — and that a DR or VPP resource with certified controllability is worth more in a world where uncontrolled load loss is itself the reliability threat. A brief carry from last week’s fuller treatment, for the one new data point: the Insurance Journal reporting on AI load volatility damaging data centers’ own batteries, generators, and cooling adds Gartner’s projection that power constraints could restrict 40% of AI data centers by 2027. That converts load-smoothing from a grid favor into a shared interest — the hyperscaler now has its own hardware-protection reason to flatten its demand profile, and the on-site storage and controls it installs to do so are the same assets a utility can dispatch. (Sources: Utility Dive — August 26 entry; Insurance Journal, August 12 — August 28 entry, previously covered August 21)
📋 Regulatory & Policy
SEPA’s large-load tariff database now tracks 77 approved and proposed tariffs and service rules across 60 different utilities — 51 approved, 26 pending — spanning 36 states, with roughly 24 states having approved at least one and about six more with proposals pending. This is the aggregate view behind every individual data-center rate action this series has tracked one at a time, and it establishes that the trend has crossed into the mainstream: 29 large-load tariffs were approved by state regulators in 2025 alone. Cost-causation discipline for data centers is now the default rather than the exception. The structural feature these tariffs share is what makes the database relevant to demand-side work — minimum-take provisions, long contract terms, exit fees, and firm capacity commitments — because that is exactly the mechanism that converts an unpriced adequacy burden into a per-customer charge. Once a large load faces a firm capacity charge, curtailable operation and enrolled demand response stop being a corporate-responsibility gesture and become a rate-management tool, which sharpens the avoided-cost comparison in the customer’s favor as well as the utility’s. The orchestration platform that certifies that flexibility then appears in the customer’s cost case, not just the utility’s — a materially better place for a DERMS business case to sit. The landscape also carries a warning worth stating: 60 utilities writing bespoke terms is a compliance and modeling burden, and fragmentation at that scale is precisely what standardized, telemetry-backed flexibility products could rationalize. The parallel to the 3,000-distribution-utility interoperability wall this series flagged in the OEM V2G context is exact — the same fragmentation problem, one layer up. (Source: SEPA — August 27 entry)
The Tennessee Valley Authority’s board approved a dedicated wholesale rate class for data centers on August 20, making the nation’s largest public-power supplier one of the first big U.S. utilities to carve AI-driven load out of its general industrial tariff. Data centers made up about 18% of TVA’s industrial power use last year, previously billed under the manufacturing service rate. Existing data-center loads face a roughly 10% base-rate increase effective October 1, phased over three years, while new or expanding loads above 5 MW fall under a “Capacity Commitment Charge” — also effective October 1, 2026 — explicitly structured to recover the incremental capacity costs those customers impose rather than socialize them onto residential and other business ratepayers. The Capacity Commitment Charge is the instrument to watch for DSM and DR business cases, because attaching a firm capacity obligation to a large load is what makes flexibility economically rational for the customer: it converts an unpriced adequacy burden into a per-customer charge, and curtailable or flexible operation becomes the lever a data center pulls to reduce its capacity-charge exposure. Note the threshold. TVA’s is 5 MW — an order of magnitude below the 50 MW definition converging across PJM, CAISO, and the DOE orders, and below NERC’s proposed 20 MW computational-load criterion. Any utility building a reconciled large-load customer list now has at least four thresholds to screen against, and the lowest one determines the size of the list. For a public-power system that also just advanced gas generation to serve this load, the rate class signals that firm supply and demand-side commitment will be priced together. Across the Southeast, TVA is the template regulators elsewhere will test against their own large-load tariffs. (Source: Chattanooga Times Free Press, August 20 — August 25 entry)
ISO-New England’s FERC Order 2222 markets go live in weeks: FERC’s May 29, 2026 approval of ISO-NE’s updated market rules cleared the last structural blocker, and DER aggregations can participate in energy and ancillary-services markets from November 1, 2026, with the capacity market following February 1, 2027 for the 2028/29 capacity year. The specific fix was restoring a process for DERs to establish Network Resource Capability and Capacity Network Resource Capability — the qualifications a resource needs to participate — which had been inadvertently dropped during Order No. 2023 compliance. That detail is worth noting for what it says about implementation risk generally: an Order 2222 program can be defeated by a qualification pathway deleted in an unrelated compliance filing, which is an argument for reading the plumbing rather than the policy. For DERMS business cases this is the wholesale revenue layer that a VPP is built to monetize, and its arrival turns Order 2222 from a compliance abstraction into a bankable value stream across New England. The friction that determines whether aggregations actually clear is entirely on the distribution side: ISO-NE distribution utilities are still standing up the processes to review DER eligibility for aggregations and to accommodate third-party, non-host-utility meter readers — which are, exactly, the telemetry and settlement functions a DERMS must supply. A utility in ISO-NE that has not scoped third-party meter-reader accommodation into its DER data architecture will discover it as a market-participation blocker in roughly nine weeks. On sequencing, ISO-NE now leads MISO (Phase 1 by June 2027, full Phase 2 by June 2029) into the live-market column, which gives Northeast utilities a concrete wholesale price signal to stack against the avoided-cost value of the same distributed assets. The DERMS decision and the market-participation strategy are one decision. (Source: Troutman Pepper / Washington Energy Report, June 2026 — August 27 entry)
A Columbia commentary quantifies the ratepayer case for demand-side resources against surging data-center load, and the elasticity is the number to carry into a proceeding: a 1% to 2% reduction in data-center peak demand can lower electricity rates by 0.5% to 2.8% while protecting reliability. A modest amount of large-load flexibility produces an outsized rate-suppression benefit that accrues to all customers, not just the participating load — which is the argument that turns a DR program from a customer amenity into a system-level affordability instrument a commission can defend. This series has carried the 1–2% / 0.5–2.8% figure as the Duke Nicholas Institute finding; the June 23 Columbia treatment is where it enters the log with the supporting frame attached. That frame is the Brattle analysis, and it is the broader case: better utilization of the existing grid — through distributed energy resources, storage, grid-enhancing technologies, demand flexibility, energy efficiency, improved rate design, flexible interconnection, and enhanced planning — could save U.S. consumers more than $100 billion over the next decade and accelerate large new loads’ interconnection by several years. The interconnection-acceleration half of that claim is the one most often left out of DR business cases and is arguably the more persuasive with a load-growth-constrained utility: the value is not only deferred peakers but years of interconnection headroom while transmission catches up. Set against this series’ standing benchmark — energy efficiency at roughly $20.70/MWh versus $45–108/MWh for new gas combined cycle — the report supplies the avoided-cost narrative that justifies funding the orchestration layer before the capacity is needed rather than after the queue seizes. (Source: Utility Dive / Columbia commentary, June 23 — August 26 entry)
🔬 EPRI Research Spotlight
A direct check confirms no new major EPRI publication entered the research log this week — the seventh consecutive dry week. At seven weeks this has stopped being a gap in coverage and become a finding about EPRI’s publication cadence, and this series will report it as such rather than continuing to flag it as an omission. The FlexMosaic framework and the DCFlex expansion remain the most recent substantive releases.
What the direct check reconfirms is the item worth foregrounding, because it is the only near-term source of measured large-load flexibility data in a year when every RTO in this digest is writing requirements against behavior nobody has systematically measured. The 96-MW Aurora AI Factory in Manassas, Virginia — a coalition of NVIDIA, Emerald AI, EPRI, Digital Realty, and PJM Interconnection — is billed as the world’s first power-flexible AI factory and is intended to serve as the implementation of a reference design and certification standard for power-flexible AI infrastructure, with demonstration testing conducted through DCFlex. The test protocol is the part that matters: precise, real-time responses to simulated grid stress events, including mimicked demand spikes during summer heatwaves and sudden drops in renewable generation. That is a controlled experiment in exactly the response characteristics PJM’s ER26-3515 filing will have to define qualification standards against. The coalition’s estimate is that the reference design, if adopted nationwide, could unlock roughly 100 GW of capacity on the existing system.
The location remains the reason to watch it closely. Manassas sits inside the Dominion Northern Virginia zone where roughly 3 GW of data-center load tripped off a normally-cleared fault on July 22 — the reference event now cited in the NERC alert, the FERC directive, and every large-load ride-through argument in this digest. A purpose-built flexible facility coming online in the same load pocket that produced the largest correlated trip in PJM history is the cleanest natural experiment the sector is going to get, and its results land directly into the computational-load standards proceeding due at FERC by December 31. One caution on timing: the facility was originally slated for the first half of 2026 and is now tracked for late 2026. Commissioning data may not arrive before the standards are filed, which means the proceeding will likely be decided on modeling rather than measurement — an argument, if anything, for conservative interconnection terms in the interim rather than for waiting.
DCFlex now spans nine demonstration sites (six added in February 2026) with a coalition including Compass Datacenters, Constellation, Emerald AI, Google, National Grid, Nebius, NVIDIA, Oracle, and PADO AI, and has expanded into Europe. Read against this week’s Google item, the overlap is not coincidental: Google is a DCFlex participant and the counterparty on 1 GW of contracted demand response. The demonstration program and the commercial commitments are being built by the same parties, which is a reason to trust the direction of travel and a reason to insist on independent verification of the numbers.
The FlexMosaic five-class taxonomy remains the valuation grammar to expect in binding flexibility agreements — Class A (infrequent extreme stress) through Class E (frequency stabilization), keyed to notification time, duration, frequency, depth, and speed of response, with Classes D and E identified as unlocking the most system value. This week gives the taxonomy a second concrete job. DOE’s Section 202(c) order specifies a 15-minute transition to backup generation for loads at or above 50 MW; PJM’s ER26-3515 must define which “demand resources and distributed energy resource aggregations” qualify as capacity. Both are questions about response characteristics, and FlexMosaic is the only published vocabulary precise enough to answer them. A 15-minute federally-mandated transition and a Class-D fast response are not the same product and should not be priced as though they were.
Verification item — NERC Computational Load Entity revised draft: still not confirmed posted, now a third week open. NERC indicated it expected to issue proposed standards and draft registry criteria for public comment in August; a targeted search this week found the anticipation restated but no confirmation of an actual posting. The initial draft went out April 1 with comments closing May 15, and the criteria as proposed remain ≥20 MW aggregate at a single point of interconnection at ≥60 kV, hosting ≥1 MW of computational load. Board approval is targeted for December 5 and the FERC filing for December 31. Three weeks open on a draft that has to clear a board vote in fourteen weeks is worth noting as schedule risk, not just as a missing document.
Standing watch: the NERC CLE revised draft; Aurora AI Factory commissioning data; and any FlexMosaic uptake in the ER26-3515 comment record. Check EPRI.com directly again next run.
🚩 Utility-Sector Relevance Flags
⚑ The Same Megawatt Is Now Available Two Ways — Negotiated or Commandeered
Topic: Large-Load Flexibility Strategy / Federal Emergency Authority / Contract Design
Relevance: In one week, Google contracted 1 GW of data-center demand response across five utilities, and DOE’s Section 202(c) Orders 202-26-32/33 authorized PJM to force any 50 MW-plus load onto backup generation within 15 minutes. Same physical capability, opposite instruments. The voluntary version pays the customer, produces a settlement record, and builds a reusable telemetry capability. The involuntary version pays nothing, proves nothing, and leaves no capability behind — and it applies whether or not the utility has done anything. A utility that declines to build a negotiated flexibility instrument does not avoid large-load curtailment; it forfeits the version that has value.
Action Signal: Implement — Frame the flexibility conversation with large customers around this fork explicitly, and put the 202(c) orders in the deck. The argument is no longer “would you consider flexibility”; it is “flexibility is coming to your facility by one of two paths, and only one of them compensates you.” Google’s 1 GW and Southern’s 1-in-3.2 ratio are the two public comparables to anchor terms against.
⚑ 50 MW Has Become the De Facto National Large-Load Threshold — and Your Screen Has Four Numbers in It
Topic: Customer Segmentation / Compliance Screening / Tariff Alignment
Relevance: PJM’s ER26-3515 defines a New Large Load at ≥50 MW cumulative peak at a single site. CAISO’s August 19 straw proposal retains a 50 MW single-site threshold. DOE’s 202(c) orders apply at ≥50 MW. But TVA’s new Capacity Commitment Charge attaches at 5 MW, NERC’s proposed computational-load criterion is ≥20 MW at ≥60 kV with ≥1 MW IT load, ERCOT’s Batch Zero audit covers ≥75 MW with community-impact review down to 25 MW, and Texas SB6 curtailment obligations bite at 75 MW. Convergence at the RTO layer, divergence everywhere else.
Action Signal: Implement — Build one reconciled large-load customer list screened against all thresholds simultaneously, and let the lowest applicable threshold size it. A list built to 50 MW will miss the customers a 5 MW capacity charge or a 20 MW registration criterion captures, and those are the customers who will call first when the rate changes on October 1.
⚑ A Statutory Percentage-of-Peak Mandate Just Reached Procurement
Topic: DR Procurement / Statutory Mandates / DERMS Coupling
Relevance: Puget Sound Energy’s live 2026 DR RFP operationalizes Washington’s DALCU requirement — a load-flexibility portfolio equal to 10% of historical peak in both winter and summer — as a build from ~129 MW to ~500 MW year-round by 2030. This series has argued that statutory peak-reduction hooks, not incentives, are what drive DR to scale; PSE is the first such mandate to reach the stage where the claim gets tested against delivered, dispatchable megawatts rather than enrollment counts. A ~370 MW build in four years is also not executable without an orchestration layer, which makes the DR procurement and the DERMS decision the same decision.
Action Signal: Engage — Utilities and vendors in the Pacific Northwest should treat this RFP as the regional benchmark for turn-key DR pricing and M&V terms. Everyone else should watch what PSE requires of respondents on measurement and settlement, because a dual-peak, percentage-of-peak mandate forces harder verification language than a voluntary program ever will — and that language will be copied.
⚑ Transmission Security Agreements Are Being Litigated as the Load Filter
Topic: Large-Load Forecasting / Contract Risk / IRP Denominators
Relevance: Microsoft’s August 21 FERC filing challenges four Large Load Project Commitment Agreements and a Minimum Transmission Charge Agreement for lacking any mechanism to keep retail customers from paying — with the Wisconsin PSC agreeing the agreements are “far from fully responsive.” PowerHouse Hillwood’s August 14 filing argues Exelon’s non-pro-forma security agreements “reflect monopoly power” and that no anti-competitive guardrails exist for large-load interconnection, in a dispute over a 1.8 GW, $20B Joliet campus. Exelon credits these same agreements for a ~40% pipeline haircut to ~11 GW. The instrument doing the filtering is itself unsettled, and FERC’s show-cause response deadlines run to mid-November.
Action Signal: Watch — Any large-load forecast, avoided-cost calculation, or capital plan resting on security-agreement-filtered load is resting on a contract term currently in active litigation. Track the ER26-67 through EL26-72 record and the settlement-judge request; if FERC imposes a pro-forma large-load security agreement, every utility’s pipeline discount factor gets re-derived at once.
⚑ Two Jurisdictions Show the Full Cost of Sequencing the Coordination Layer Wrong
Topic: DERMS Procurement Sequencing / Integration Debt / Program Design
Relevance: Hawaii’s PUC opened Docket 2026-0084 to design VPP grid services across an already installed base of >1 GW of rooftop solar and ~410 MW of attached batteries that the Commission says are largely unoptimized — retrofitting orchestration after saturation. CalChoice is doing the inverse, building “the DERMS framework first,” finalizing contracts and technical roles before enrollment, then staging rollout starting with the existing battery base. Same technology, opposite order, and the cost difference is the entire argument for funding the platform early. The MarketScale capital-planning item names the mechanism: DER volume is outpacing existing integration toolsets, and integration debt compounds.
Action Signal: Implement — Specify the DERMS and coordination layer as a funded line inside the load-growth capital plan rather than a discretionary item competing against supply-side builds. Use Hawaii as the exhibit: a jurisdiction that reached high DER penetration first is now paying to retrofit coordination onto assets that are already installed and already paid for.
⚑ The Rate Case Is Now the Flexibility Case
Topic: Large-Load Rate Design / Customer Economics / DERMS Business Case
Relevance: SEPA counts 77 large-load tariffs across 60 utilities (51 approved), spanning 36 states, with 29 approved in 2025 alone; the shared structural features are minimum-take provisions, firm capacity commitments, exit fees, and long terms. TVA’s Capacity Commitment Charge on new loads above 5 MW is the concrete instance, effective October 1. Once a large customer faces a firm capacity charge, curtailable operation becomes a rate-management tool the customer has reason to pursue — which moves the orchestration platform that certifies flexibility into the customer’s cost case rather than only the utility’s.
Action Signal: Engage — Pair every large-load tariff filing with a flexibility product the same customer can buy to reduce its committed capacity, and file them together. A capacity charge without an adjacent flexibility offering is a pure cost increase that invites litigation; a capacity charge with one is a price signal with a response available. SEPA’s database is also the comparables set for defending your own terms.
⚑ ISO-NE Order 2222 Goes Live in Nine Weeks and the Blocker Is on the Distribution Side
Topic: Wholesale Market Access / DER Data Architecture / Revenue Stack
Relevance: FERC’s May 29 approval restored the NRC/CNRC qualification process, and DER aggregations can participate in ISO-NE energy and ancillary-services markets from November 1, 2026, with capacity following February 1, 2027 for the 2028/29 capacity year. But distribution utilities are still standing up DER-eligibility review processes and third-party (non-host-utility) meter-reader accommodation — the exact telemetry and settlement functions a DERMS must supply. ISO-NE moves ahead of MISO’s 2027–2029 phased schedule.
Action Signal: Engage — New England utilities should confirm this quarter that DER-eligibility review and third-party meter-reader accommodation are operational, not planned. An aggregation that cannot be metered by a third party cannot clear, and November 1 is not a target date that moves. Elsewhere, note the failure mode: ISO-NE’s DER qualification pathway was accidentally deleted during unrelated Order 2023 compliance work and had to be restored by a separate FERC filing. Audit your own compliance filings for the same class of collateral damage.
⚑ ERCOT’s Load Forecast Is Now a Casualty of the Interconnection Pause
Topic: Load Forecasting Integrity / Resource Adequacy / Planning Risk
Relevance: ERCOT committed at the August 20 PUCT open meeting to auditing ~300 data-center proposals ≥75 MW in Batch Zero, targeting a December 10 filing, against a 474 GW queue that is ~90% data centers — more than five times record peak. Officials warned the pause could push back ERCOT’s long-term load forecast and reliability assessment, and that the Batch Zero study will slip past its April 9, 2027 deadline. Seventeen large loads totaling ~6.6 GW sit at the final energization gate.
Action Signal: Watch — Track the December 10 filing as a hard checkpoint. The verified-load number ERCOT publishes will be the first regulator-audited discount applied to a large-load queue anywhere in the country, which makes it a citable benchmark far beyond Texas — potentially more durable than Ascend’s 55.4% success rate because it is an official verification rather than a modeled assumption. Plan for the possibility that ERCOT’s next long-term forecast carries an explicit uncertainty band rather than a point estimate.
⚑ Surplus Interconnection Is an Underused Route to Dispatchable Capacity
Topic: Interconnection Reform / Storage Siting / Capacity Shortfall
Relevance: PJM’s first Surplus Interconnection Service reform produced 8 applications and 2 approvals, against MISO’s ~14.8 GW and SPP’s ~14.3 GW of surplus requests — mostly batteries, with 44 SIS projects online in MISO and 22 in SPP. PJM is reopening the rules because it missed reserve margins in two consecutive auctions, with the 2028/29 shortfall at ~6.8 GW. The proposed fix would let hybrid resources clear capacity as one unit while bidding energy and ancillary services as separate resources, unlocking CIR access for co-located storage. UC Berkeley estimates PJM’s existing sites could host ~150 GW; many PJM solar projects use only 40–60% of nameplate CIRs. Indiana and Virginia passed 2026 study laws.
Action Signal: Engage — Inventory your own fleet’s unused capacity interconnection rights before the next capacity procurement, and participate in PJM’s rule redesign if you hold underutilized interconnection points. A battery sited behind an existing interconnection with 40–60% headroom is faster and cheaper than greenfield, and it belongs in the same avoided-cost comparison as a new peaker rather than in a separate innovation track.
⚑ Program Duration Is a Storage Procurement Instrument
Topic: DR Program Design / Storage Enrollment / Contract Terms
Relevance: New York’s DLM overhaul spans CSRP, DLRP, DLC, Term-DLM and Auto-DLM, and two structural parameters do most of the work: a 50 kW minimum keeps aggregated commercial load eligible, and the state’s minimum three-year program term is what makes battery storage economic to enroll — converting incentive-driven signups into durable dispatchable capacity. DLRP dispatches on network contingency, so it is sized against deferred T&D and contingency capacity rather than a flat energy price.
Action Signal: Implement — Review the term length of every DR program in your portfolio. A one-year program cannot underwrite a customer capital asset, which means short-term programs structurally exclude storage no matter how attractive the incentive. Extending term is often cheaper than raising the payment, and it changes which technologies can participate at all. Pair it with contingency-based dispatch triggers so the avoided cost is a locational deferred asset rather than a system average.
⚑ Two Design Forks Are Open Right Now in Three Jurisdictions — Same Question
Topic: VPP Compensation Design / Advocacy Targeting / M&V
Relevance: New Jersey’s BPU straw proposal (Docket QO26030099, 2027 launch) will direct utility rebates to customers with grid-connected batteries, thermostats, EV chargers, and heat pumps. Hawaii’s Docket 2026-0084 is designing grid-services compensation across an installed 1 GW / 410 MW base. California’s GridLab blueprint recommends compensation on verified grid performance rather than enrollment. All three are deciding the identical question — flat enrollment payment versus performance-paid dispatch — and only New Jersey still has a blank page.
Action Signal: Engage — File the same argument in all three: compensation tied to verified dispatch performance, with revenue-grade measurement specified in the program rules rather than deferred to implementation. This is the cheapest advocacy available right now because the argument is identical across venues and the record in one is citable in the others. A program that pays for enrollment produces a headcount; a program that pays for performance produces capacity a resource plan can count.
📌 Sources
August 24, 2026 Entry
– Utility Dive (Robert Walton) — ERCOT, Texas PUC Data Center Audit (August 21, 2026)
– Utility Dive (Robert Walton) — Duke’s Carolinas Resource Plan: Solar, Gas, Nuclear (August 20, 2026)
– Utility Dive (Brian Martucci) — California DER Wholesale Market Participation Could Grow 2 GW From CAISO Change (August 5, 2026)
– Utility Dive (Ethan Howland) — PJM Surplus Interconnection Service (August 21, 2026)
August 25, 2026 Entry
– Utility Dive (Ethan Howland) — Microsoft, PowerHouse Hillwood Data Center Service Agreements at FERC (August 24, 2026)
– Chattanooga Times Free Press — TVA to Charge Data Centers More for Power Under New Rate Class (August 20, 2026)
– California Community Choice Association (CalCCA) — CalChoice Sets Sights on Virtual Power Plant With New DERMS Program (March 18, 2026)
– New Jersey Board of Public Utilities — Virtual Power Plant Straw Proposal, Docket QO26030099 (July 27, 2026)
August 26, 2026 Entry
– Utility Dive (Brandon Owens) — Data Centers Can Vanish From the Grid in Seconds; Reliability Rules Are Coming (August 2026)
– Puget Sound Energy — 2026 Demand Response Request for Proposals (2026)
– Utility Dive — GETs, Demand Response Can Ease Near-Term Data Center Electricity Price Pressure (Columbia commentary, June 23, 2026)
– Hawaii Public Utilities Commission — PUC Opens Docket on Virtual Power Plant Program and Invites Participation, Docket 2026-0084 (April 2026)
August 27, 2026 Entry
– SEPA — U.S. Data Center Gold Rush Drives Surge in New Utility Tariffs (Database of Emerging Large Load Tariffs, updated 2026)
– Data Center Dynamics — Google Totals 1 GW of Demand-Response Capability in US Data Centers (2026)
– Troutman Pepper / Washington Energy Report — FERC Approves ISO-NE Updated Market Rules for Distributed Energy Resources (June 2026)
– T&D World — Why DERMS Is Earning a Place Alongside SCADA (2026)
August 28, 2026 Entry
– Utility Dive — PJM Gets Emergency Approval to Curtail Data Centers, Large Loads During Hot Weather (2026)
– New York Department of Public Service — Commission Improves Customer-Centered Electric Demand Response Programs (2026)
– Insurance Journal — AI’s Volatile Power Demand Is Damaging Its Own Data Centers (August 12, 2026) (previously reported in the August 21 digest; carried here only for the Gartner 40%-by-2027 projection)
– MarketScale — The Grid Investment Surge of 2026 Is Reshaping What Utility Operators Must Evaluate Now (2026)
Verification Sources (targeted search, not from the daily research log)
– California ISO — Large Loads Initiative: Technical Requirements Straw Proposal Posted, Meeting 8/19/26
– mgrid.org — CAISO Answers FERC With Two Flexible Interconnection Services and a 50 MW Large-Load Definition (August 18, 2026)
– NERC — FERC Sets Year-End Deadline for NERC to Finalize Registry Criteria and Standards for Computational Loads
– NERC — Computational Load Entity, Summary of Changes (April 2026 posting)
– American Public Power Association — NVIDIA, Emerald AI, EPRI, PJM and Others to Develop Power-Flexible AI Factory
– Renewable Energy World — EPRI Expands DCFlex Data Center Initiative to Nine Demo Sites Across U.S. and Europe