Reliability as a Purchased Good: Private Substitution for a Degrading Public Power Guarantee (2026–2031)openMediumorigin: briefs
Output
| Thesis probability | 80.0%from a 85.0% prior · 21 active evidence items |
| Reasonable range | 69% – 91% tick = prior · marker = posterior |
| Historical analogy strength | 62/ 100 · 18 analogues in the reference class |
| Current evidence strength | 99/ 100 · 21 active items |
| Evidence quality | 67/ 100 |
| Contradictory evidence | High |
| Confidence in probability | Medium |
| Next major confirmation signal | Generac's Q3 2026 print (late October) is now the nearest-term resolver: specifically whether C&I/data-centre backlog is growing faster than storm-driven residential, and whether residential order intake is already mean-reverting in a 7%-mortgage environment. Second, by mid-2027, whether a second hyperscaler signs a multi-billion BTM or firm-power contract on the Amazon template. Third, watch whether any US state or EU member state introduces mortgage-embedded or on-bill financing for residential resilience hardware — that is the only identified mechanism that neutralises the rate shock, and its absence through 2027 makes metric (i) very hard. |
Posterior log-odds = prior log-odds (reference-class base rate) + sum of quality x independence weighted log likelihood ratios. Research-informed estimate until enough theses resolve for Brier calibration.
Formalized thesis
By 31 December 2031, grid reliability is treated as a purchased good rather than a public guarantee: annual unit sales of residential/commercial backup generation and storage (generators, home batteries, solar+storage attach) in the US and EU at least double versus 2025, corporate behind-the-meter and long-term firm-power contracting (LNG, nuclear PPAs, district heat) covers a materially larger share of large-enterprise load, and consumer search interest in grid-failure/outage preparedness stays structurally above its 2024–2025 baseline.
Source observations
- 2026-09-17Energy self-reliance: households and enterprises increasingly want to own their generation and resilience \u2014 'power grid failure/threat/outage' queries alongside Amazon buying generators. Reliability becomes a purchased good, not a public utility assumption.
- 2026-09-24Energy is becoming a status and security good, not a commodity line item: district heating investment, LNG contracting for data centers, pipeline races and grid-failure searches all reflect a population and a corporate sector paying up for guaranteed electrons.
Causal mechanism
Loss aversion around a service whose failure is catastrophic and non-substitutable (heat, refrigeration, medical devices, remote work, EV charging). Once the perceived probability of failure crosses a salience threshold — usually via one vivid personal or televised event — households and boards shift from optimisation to insurance, and reliability becomes a status/security consumption category rather than a utility line item.
Historically self-generation required diesel logistics, permitting, noise, an electrician, and $10k+ of capital. That friction is collapsing: plug-in battery/inverter kits, integrated ATS panels, dealer-installed standby units financed on the mortgage or a lease, virtual-power-plant enrollments that monetise the asset, and — for corporates — turnkey BTM gas/fuel-cell/nuclear PPA packages sold by developers rather than built in-house. Interconnection queue delays (5–7 years) paradoxically remove friction from the BTM route by making the grid the slow option.
Cheap LFP cells and hybrid inverters; smart panels and automatic transfer switches; aeroderivative turbines and reciprocating gensets with fast delivery slots; fuel cells; SMR/nuclear PPA structures; grid-edge orchestration software (Emerald AI-type headroom products, VPP dispatch); cheap sensing and outage telemetry that makes unreliability legible in real time.
For enterprises: the cost of a data-centre or fab outage is measured in $M/hour, and AI load growth cannot wait for interconnection — BTM is a schedule purchase, not just a price purchase. For households: rising retail tariffs (grid-hardening capex, wildfire liability, transmission build) widen the spread between grid kWh and self-generated kWh, while VPP payments and capacity-market revenues create an income stream that partially funds the asset. Insurance pricing and spoilage/WFH-income loss add quantifiable avoided cost.
Inverse trust dynamic: adoption rises as trust in the public guarantee falls. Buyers must nonetheless trust the private substitute — installer quality, fuel availability, battery fire safety, warranty, and the counterparty's ability to deliver firm MW in 2029. Certification (UL 9540, CE, NFPA), utility interconnection approval, and lender/insurer acceptance are the trust rails.
Big-box retail and e-commerce (Home Depot, Lowe's, Amazon, Bauhaus, Leroy Merlin) for portables; dealer/installer networks and solar channels for standby and storage; homebuilders bundling into new construction; utilities themselves selling resilience-as-a-service; for enterprises, EPCs, IPPs and developers (Quanta, Bechtel, Talen, Constellation, NextEra, Bloom) plus direct OEM allocation deals with hyperscalers.
Weak direct, strong indirect. Installer density, fuel/service ecosystems, and VPP aggregation create local increasing returns; visible neighbourhood adoption (a humming generator next door during a blackout) is a powerful social-proof channel. Adverse loop: each defector reduces the revenue base funding grid reliability, raising tariffs for the remainder and accelerating further defection — a self-reinforcing utility death-spiral dynamic.
High once installed: sunk capital, fixed gas/propane connection, roof and panel integration, financing tied to the property. This makes the installed base sticky and the thesis hard to reverse even if grid reliability improves — the behaviour ratchets.
Two-sided. Constraining: interconnection rules, net-metering rollbacks, air-permit limits on diesel/gas runtime hours (California BAAQMD, EU IED), noise ordinances, HOA rules, battery siting codes, EU capacity-market and self-consumption tax treatment. Enabling: FERC 2222 aggregation, state resilience rebates, IRA/CSRD-era incentives, EU energy-security rules post-2022, and explicit data-centre BTM carve-outs now being drafted in PJM/ERCOT and several EU member states.
Historical analogues
| Analogue | Period | Mechanism | Similarity | Occurred | Yrs to mainstream | Lesson | |
|---|---|---|---|---|---|---|---|
South Africa load-shedding → rooftop solar and battery boom | 2019–2024 | Documented, repeated failure of the public electricity guarantee triggers mass private self-generation by households and firms within ~4 years. | 79 | ✓yes | 4 | The closest living laboratory: private BTM capacity rose from negligible to >5 GW in under five years and did NOT reverse when load-shedding paused. Adoption is event-triggered and ratchets; the trigger is repetition, not severity. | |
Texas Winter Storm Uri → Generac standby generator demand surge | 2021–2023 | Localised catastrophic outage raises perceived tail risk; residential standby generator penetration and order backlogs jump sharply. | 76 | ✓yes | 2 | Demand spikes are real but partially mean-revert: Generac's post-Uri backlog unwound by 2023. The thesis requires a *sequence* of events plus rising tariffs, not one shock, to produce a structural doubling. | |
Diesel gensets as default infrastructure in Nigeria, India, Lebanon, Pakistan | 1980–present | Chronic unreliability converts backup generation from an exception into a mandatory capital item for any commercial premises. | 72 | ✓yes | 10 | Where the grid fails persistently, private generation becomes a line item in every building budget and a permanent 5–10% tax on GDP. The end state of the thesis exists; the open question is how far the US/EU travel along it. | |
Hyperscaler direct nuclear/gas PPAs and BTM campuses (Microsoft–Constellation, Meta gas plants, Amazon–Talen/Generac) | 2023–2026 | Load growth outruns interconnection; the largest buyers bypass the public grid queue by contracting or building firm generation directly. | 70 | ✓yes | 3 | Already happening and accelerating — this is the leg of the thesis with the shortest path to confirmation. The binding constraint is turbine/transformer/EPC supply and permitting, not demand or willingness to pay. | |
Japanese corporate self-generation and household solar+storage after Fukushima | 2011–2020 | A single systemic shock destroys confidence in centralised supply; corporates and households buy on-site generation and storage as insurance against policy and physical failure. | 69 | ✓yes | 7 | One salient national event can permanently reset the willingness to pay for self-supply, but the size of the shift depends heavily on subsidy design (FIT) — policy converts fear into installed base. | |
Enterprise UPS and on-site data-centre generation becoming mandatory | 1985–2005 | As computing became mission-critical, N+1 power redundancy went from optional to a contractual/insurance requirement — reliability purchased, not assumed. | 65 | ✓yes | 15 | Once uptime enters SLAs and insurance underwriting, resilience capex becomes non-discretionary and survives downturns. The corporate leg of this thesis is the highest-confidence leg. | |
Corporate cogeneration build-out after the 1973/1979 oil shocks and PURPA | 1974–1990 | Energy-price and supply insecurity plus regulatory enabling → industrials self-generate rather than buy from the utility. | 64 | ✓yes | 12 | Regulatory permission (PURPA) was the unlock, not technology. The analogous unlock here is BTM/interconnection carve-outs for data centres — watch the rulemaking, not the hardware. | |
Bottled LPG and wood-stove reversion during the 2022 European gas crisis | 2021–2023 | Supply-security shock drives households toward self-owned, storable energy — but reverses once prices normalise. | 63 | ✗no | 0 | Counter-case: heat-pump and wood-stove demand in Germany/Netherlands spiked in 2022 and fell hard by 2024 when prices and fear normalised. Fear-driven durable-goods demand mean-reverts unless tariffs stay structurally high or outages repeat. | |
Residential home-security systems and alarms (ADT, Ring) substituting for policing | 1970–2020 | Private purchase of a security service that was nominally a universal public guarantee, driven by perceived risk and falling hardware cost. | 62 | ✓yes | 25 | Security categories grow slowly for decades, then inflect when hardware cost collapses and distribution becomes retail/DIY (Ring). Watch for the 'Ring moment' in plug-in home batteries. | |
US residential solar 'grid defection' predicted after net-metering rollbacks | 2015–2025 | Rising retail tariffs and policy hostility were forecast to drive mass self-supply and utility death spiral. | 61 | ✗no | 0 | Important negative case: California NEM 3.0 and high rates produced attach-rate growth in storage but an outright *contraction* in US residential solar volumes in 2024–25 on financing costs and installer bankruptcies. High rates cut both ways — they raise tariffs but also kill consumer-financed hardware. At ~5% risk-free, the residential leg is far weaker than the corporate leg. | |
Dollarisation / private currency substitution in Argentina, Turkey, Venezuela | 1990–present | Loss of faith in a state-provided guarantee (money) leads households to bear real cost to hold a private/foreign substitute, and the behaviour never fully reverses. | 60 | ✓yes | 6 | Trust, once broken in a public guarantee, does not restore on the same timescale it was lost — hysteresis is the rule. Applies directly to whether outage-preparedness search stays structurally elevated after events fade. | |
Gated communities and private residential infrastructure (roads, water, security) in the US and Latin America | 1980–2010 | Households buy bundled privately-provisioned public goods when municipal provision is distrusted; capitalised into property value. | 59 | ✓yes | 20 | Once resilience capitalises into home prices and mortgage underwriting, adoption becomes builder-driven and automatic — the single biggest potential accelerant for residential storage attach rates. | |
Bottled water and home filtration substituting for municipal tap water (US/EU) | 1975–2005 | A universal public utility's perceived quality/safety guarantee erodes; consumers pay a very large per-unit premium for a privately supplied substitute even where the public good remains objectively adequate. | 57 | ✓yes | 20 | Perceived reliability, not measured reliability, drives spend; once a public guarantee is psychologically privatised, consumption is sticky and price-insensitive for decades. Distribution through mass retail was the accelerant. | |
Cellular/Starlink failover replacing single-ISP dependence for homes and SMBs | 2015–2025 | Connectivity reclassified from utility to mission-critical; buyers pay for a second, independent path rather than trusting one provider. | 57 | ✓yes | 8 | Redundancy sells once the primary service becomes economically load-bearing (remote work). Power follows the same logic as WFH, EVs and home medical devices raise the cost of an outage hour. | |
Home air purification and HEPA adoption after wildfire smoke and COVID | 2018–2023 | Environmental failure of an assumed-free public good (clean air) triggers rapid private hardware purchase. | 54 | ✓yes | 4 | Cheap, plug-in, no-install form factors adopt 5x faster than permitted/installed ones. Portable power stations will scale faster than whole-home standby, which matters for which manufacturers capture the volume. | |
Private security guards outnumbering police in the US/UK/Brazil | 1960–2000 | Public provision spread thin → institutional buyers contract private substitutes for the same service at premises level. | 52 | ✓yes | 25 | Enterprise buyers move first and faster than households because the loss function is quantified; consumer adoption follows a decade later. Expect corporate firm-power contracting to lead the household generator cycle. | |
Private health insurance / private GP usage alongside the NHS and European public systems | 1990–present | Queues and degraded service quality in a universal public guarantee drive top-income deciles to buy a parallel private tier while still paying for the public one. | 49 | ✓yes | 20 | Partial, income-stratified defection is the most common outcome — not universal privatisation. The realistic 2031 state is a bifurcated market: affluent suburbs and large corporates self-provision, everyone else absorbs worse service and higher tariffs. | |
Private firefighting services for insured high-value homes (California, Colorado) | 2007–2025 | Insurers and wealthy households buy a parallel private version of an emergency public service as public capacity is overwhelmed. | 48 | ✓yes | 12 | Insurers, not consumers, can be the buyer of resilience. Watch for homeowners' insurers subsidising or mandating backup power/storage — that would be the single fastest distribution channel available. |
Reference class: 'private substitution for a degrading or capacity-constrained public/universal-service guarantee, where the substitute is a durable capital good bought by households and firms.' Members span water (bottled/filtration), security (alarms, guards, gated communities), health (parallel private tiers), connectivity (redundant ISP paths), money (dollarisation) and — most directly — electricity (South Africa, Japan post-Fukushima, Nigeria/India gensets, PURPA cogeneration, Texas Uri, hyperscaler PPAs). The class divides cleanly into two sub-classes with different base rates: (a) ENTERPRISE substitution, where the loss function is quantified, the buyer is sophisticated, purchase is capex-financed and the pattern occurred in ~90% of cases within 3–15 years (UPS/N+1, cogeneration, private guards, hyperscaler PPAs); and (b) HOUSEHOLD substitution, where adoption is event-triggered, consumer-credit-sensitive and slower — pattern occurred in roughly 70% of cases but with 4–25 year mainstreaming times and frequent partial reversion (European gas crisis 2022, post-Uri backlog unwind, US resi solar 2024–25). Typical successful cases required THREE conditions jointly: repeated (not single) failure events, a structural rise in the price of the public option, and a collapse in the installation/financing friction of the private option. Median years-to-mainstream across the occurring cases is ~10; the electricity-specific cases cluster at 3–7 years when failures repeat annually.
Caveats: Selection bias: the vivid cases (South Africa, Nigeria, Lebanon) come from grids with 100–1,000x the outage frequency of US/EU systems; SAIDI in Germany is ~12 minutes/year and in most of the US ~2–8 hours including major events, so the US/EU baseline may simply never cross the salience threshold outside Texas, California and a few EU member states. The thesis bundles two very differently-based legs — the corporate/BTM leg is near-confirmed and the residential-unit-doubling leg is the genuinely uncertain one; scoring them together inflates apparent confidence. A ~5% risk-free rate is a strong counterweight specific to this cycle: every residential analogue that succeeded did so under cheap consumer financing or heavy subsidy, and the 2024–25 US resi solar contraction is direct evidence that high rates can suppress hardware demand even while fear and tariffs rise. Measurement is fragile: generator unit shipments are partly reported by a single dominant issuer (Generac) whose disclosure changes, EU BESS statistics are collected inconsistently across member states, and Google Trends indices are non-stationary and reflexive to news cycles — the search leg is the weakest and least independent of the three. Base rates also ignore the possibility of the public guarantee being *restored* via grid capex, demand-flexibility software and faster interconnection reform, which would make this a 5-year spike rather than a regime change; and policy could pre-empt private substitution by forcing hyperscalers to fund shared transmission instead of building BTM.
Evidence
| # | Claim | Direction | Strength | Quality | Indep. | Weight | LR | Wtd log-LR | Kind | Source | Date |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | Amazon awarded Generac an ~$8B generator contract for data-centre backup/bridging power, the single largest behind-the-meter genset order ever disclosed by a hyperscaler (stock +40% on the day). | supports | strong | 0.85 | 0.90 | 0.77 | 3.00 | +0.840 | Observed fact | Daily brief 2026-09-17 (market news) | 2026-09-17 |
| 1 | Electricity availability, not GPU supply, is now the binding constraint on AI buildout — corroborated independently by Oracle filing a force majeure notice on a data-centre project, Meta funding seven gas plants, Mitsui contracting LNG explicitly for data centres, and Talen's data-centre framework progress. | supports | strong | 0.80 | 0.75 | 0.60 | 3.00 | +0.659 | Observed fact | Daily briefs 2026-09-13/17/24 | 2026-09-24 |
| 2 | Consumer search interest in 'power grid failure', 'power grid threat' and 'power outage' has been persistently elevated in the attention dataset alongside AI data-centre queries ('ai data center near me', 'ai data center water usage'), indicating households are linking local grid stress to the buildout. | supports | moderate | 0.55 | 0.60 | 0.33 | 2.00 | +0.229 | Observed fact | Attention movers / brief 2026-09-17 | 2026-09-25 |
| 3 | Interconnection queues of 5–7 years in PJM/ERCOT and most EU TSOs mean behind-the-meter generation is bought as a schedule product, not a price product; developers and IPPs are already selling turnkey BTM gas/fuel-cell packages to close that gap. | supports | strong | 0.70 | 0.80 | 0.56 | 3.00 | +0.615 | Consensus expectation | Mechanism analysis / FERC-ISO queue data | 2026-09-25 |
| 4 | Nvidia, Google and Anthropic backed Emerald AI specifically to manufacture grid headroom via flexible load — capital is being deployed on the assumption that public grid capacity will not arrive on time. | supports | moderate | 0.75 | 0.70 | 0.52 | 2.00 | +0.364 | Observed fact | Daily brief 2026-09-17 | 2026-09-17 |
| 5 | Policy is turning enabling: Italy reversed a 40-year nuclear ban, EU member states and PJM/ERCOT are drafting data-centre BTM carve-outs, and FERC 2222 aggregation plus state resilience rebates lower the regulatory friction to private firm supply. | supports | moderate | 0.60 | 0.70 | 0.42 | 2.00 | +0.291 | Observed fact | Daily brief 2026-09-24 / regulatory dockets | 2026-09-24 |
| 6 | Utilities are guiding load growth higher (Duke, Constellation raising 2026), confirming demand is outrunning planned supply — the precondition for scarcity-priced firm power and private substitution. | supports | moderate | 0.70 | 0.65 | 0.45 | 2.00 | +0.315 | Observed fact | Company guidance cited in briefs | 2026-09-17 |
| 7 | The reference behaviour has already fully occurred in analogue jurisdictions (South Africa load-shedding → rooftop solar/battery boom; Nigeria/India/Lebanon diesel gensets as default; post-Fukushima Japanese corporate self-generation), showing the mechanism reliably completes once perceived failure probability crosses a salience threshold. | supports | strong | 0.70 | 0.85 | 0.59 | 3.00 | +0.654 | Observed fact | Reference-class analogues | 2026-09-25 |
| 8 | Residential generator and battery demand is historically event-spiky and mean-reverting: Generac's post-Uri 2021 surge fully unwound by 2023, and German/Italian residential BESS installs fell sharply after subsidy and net-metering rollbacks in 2024–25. A 2.0x sustained unit-sales level in 2031 requires a structural break, not a storm. | contradicts | strong | 0.70 | 0.85 | 0.59 | 0.33 | -0.654 | Observed fact | Generac/EU BESS shipment history (SolarPower Europe, company filings) | 2026-09-25 |
| 9 | The cost-of-capital regime is hostile to household resilience capex: 30-year mortgage at 7.45% (20-month high), 10-year near 5.15%, Fed pricing hikes not cuts. Financed $10k–$20k standby/battery purchases become discretionary for exactly the middle-income cohort needed to move units from 'innovators' to 'majority'. | contradicts | strong | 0.90 | 0.90 | 0.81 | 0.33 | -0.890 | Observed fact | Daily brief 2026-09-25 | 2026-09-25 |
| 10 | Consumer bifurcation caps the addressable base: value-retail leadership, weak housing turnover and a squeezed levered consumer mean the bottom two income quintiles cannot buy reliability at all, so defection is concentrated in a wealthy minority — sufficient for a political story, possibly insufficient for a 2x unit count. | contradicts | moderate | 0.60 | 0.70 | 0.42 | 0.50 | -0.291 | AI inference | Brief theme 'Value Retail & Consumer Bifurcation' | 2026-09-25 |
| 11 | A +15pp shift in large-enterprise load to BTM or ≥5-year firm PPA within six years is an extreme rate of change versus the observed PPA/BTM base rate (roughly 2–4pp per year growth in disclosed corporate firm contracting), and hyperscalers are a small share of total Fortune 500 / STOXX 600 load. | contradicts | moderate | 0.65 | 0.80 | 0.52 | 0.50 | -0.360 | AI inference | BNEF/LevelTen PPA tracker base rates | 2026-09-25 |
| 12 | Housing turnover collapse (Lennar miss and delivery cut, one-year-low builder confidence, -4.7% contract signings) removes the new-construction bundling channel, historically the cheapest route to mass installation of standby generators and integrated storage. | contradicts | moderate | 0.80 | 0.75 | 0.60 | 0.50 | -0.416 | Observed fact | Daily briefs 2026-09-17/25 | 2026-09-25 |
| 13 | Air-permit runtime limits (California BAAQMD, EU IED), noise ordinances, HOA rules and battery siting codes are tightening as BTM density rises; several EU member states are also taxing self-consumption to defend the tariff base. | contradicts | moderate | 0.55 | 0.70 | 0.39 | 0.50 | -0.267 | Consensus expectation | Regulatory mechanism review | 2026-09-25 |
| 14 | If AI capex proves to be a financing event rather than a demand event (circular vendor financing — Nvidia into Anthropic, vendor-linked cloud commitments; Burry shorts; Apple's local-inference cost-per-token claim), the incremental enterprise load that drives the enterprise leg of this thesis could arrive materially smaller than currently contracted. | contradicts | moderate | 0.60 | 0.80 | 0.48 | 0.50 | -0.333 | Speculative hypothesis | Daily briefs 2026-09-13/24 | 2026-09-24 |
| 15 | The long-end repricing extended and broadened after the last update — US 10-year 5.33-5.34% (highest since 2002), 30-year above 5.6%, UK 30-year gilts above 6% — and it occurred despite a softer-than-expected August core PCE (3.0%), indicating a term-premium/supply regime rather than a Fed-path move that can reverse on a dovish pivot. Financed $10k-$20k household resilience purchases get structurally more expensive for the duration of the thesis window. | contradicts | strong | 0.90 | 0.85 | 0.77 | 0.33 | -0.840 | Observed fact | Daily briefs 2026-09-30/10-01 (rates tape) | 2026-10-01 added 2026-10-01 |
| 16 | The oil shock is a refining/distillate crisis (Brent ~$100, Chinese October fuel-export ban, US asking Germany/France to release diesel reserves, £2/litre UK diesel), raising the operating-cost leg of diesel/LPG backup generation precisely as households would be evaluating payback. | contradicts | moderate | 0.70 | 0.75 | 0.52 | 0.50 | -0.364 | Observed fact | Daily briefs 2026-09-30/10-01 | 2026-10-01 added 2026-10-01 |
| 17 | US consumer confidence fell to a 12-year low (81.9) with 9.3m student-loan borrowers in default, while aggregate spending is being sustained by a top-quintile asset wealth effect rather than wages — hard confirmation that the middle of the income distribution, the cohort required to move resilience hardware from early adopters to early majority, cannot fund it. | contradicts | strong | 0.80 | 0.70 | 0.56 | 0.33 | -0.615 | Observed fact | Daily brief 2026-10-01 | 2026-10-01 added 2026-10-01 |
| 18 | Nvidia's vendor financing guarantees to chip buyers are drawing credit-risk scrutiny as neocloud debt accumulates and non-tech firms lever into GPUs; a first default would de-rate the complex faster than a demand miss. This raises the probability that contracted AI load is partly a financing artefact and that some BTM/firm-power contracting is cancelled or deferred. | contradicts | moderate | 0.65 | 0.60 | 0.39 | 0.50 | -0.270 | Observed fact | Daily brief 2026-10-01 | 2026-10-01 added 2026-10-01 |
| 19 | Attention data shows no new surge in outage/grid-failure queries in the latest movers set (top themes are streaming, stablecoins, chip fabs, cloud, TikTok, Starlink) — the consumer search leg (metric iii) is holding at elevated-but-not-accelerating levels absent a domestic catalyst. | contradicts | weak | 0.50 | 0.50 | 0.25 | 0.77 | -0.066 | Observed fact | Attention movers 2026-10-01 | 2026-10-01 added 2026-10-01 |
| 20 | A drone strike on a UAE nuclear plant, alongside Danish/NATO warnings and continued state nuclear mega-projects, keeps physical energy infrastructure visible as a wartime target — reinforcing the 'security as structural spend' framing at household and state level. | supports | weak | 0.55 | 0.45 | 0.25 | 1.30 | +0.065 | Observed fact | Daily briefs 2026-09-29/30 | 2026-09-30 added 2026-10-01 |
| 21 | Energy equity rotation cooled from +23 to +10 with Brent near $100 while Utilities sit at -85 and Real Estate at -87 — capital is not re-rating the regulated/grid complex, which weakens the 'energy and compute converging into a single bid asset class' corollary even as the physical scarcity persists. | contradicts | weak | 0.50 | 0.55 | 0.28 | 0.77 | -0.072 | Observed fact | Daily brief 2026-10-01 (flow data) | 2026-10-01 added 2026-10-01 |
| 22 | No second hyperscaler multi-billion-dollar BTM/firm-power deal on the Amazon-Generac template has been disclosed in the fortnight since the anchor transaction; the enterprise leg still rests on a single marquee buyer. | contradicts | weak | 0.60 | 0.80 | 0.48 | 0.77 | -0.126 | Observed fact | Absence of disclosure, briefs 2026-09-25 to 10-01 | 2026-10-01 added 2026-10-01 |
Stage of adoption
Unchanged in kind, worse in financing conditions. Enterprise leg still at early majority; consumer leg has arguably slipped back toward late innovators as the discount rate and fuel cost both moved against household payback in the same fortnight. The analogue remains US home security circa early 1990s / South African residential solar 2021, but with the critical difference that the South African boom occurred alongside a household-level forcing function (daily load-shedding) that neither the US nor EU currently has. The ratchet logic on installed base still holds.
- ✓Salience shock delivered repeatedly (Uri, EU gas crisis, PSPS, data-centre siting fights, now kinetic attacks on energy infrastructure)
- ✓Enterprise economics unambiguous: $M/hour outage cost plus 5-7yr interconnection queues make BTM a schedule purchase
- ✓Supply chain exists at scale (aeroderivatives, recips, LFP, hybrid inverters, fuel cells, smart panels with ATS)
- ✓Distribution rails built across retail, dealer/solar installer and EPC/IPP channels
- ✓Certification and trust rails in place (UL 9540, NFPA, CE); lender/insurer acceptance improving
- ✓Enabling regulation emerging (FERC 2222, state resilience rebates, drafted data-centre BTM carve-outs, Italy nuclear reversal)
- ✓A marquee anchor transaction exists (Amazon-Generac ~$8B)
- ○Affordability — now worse, not better: a 5.33% 10-year and 7%+ mortgages make the household unit economics unviable for the middle quintiles without subsidy or VPP revenue
- ○Stable fuel operating cost — the distillate crisis directly undermines the genset payback leg
- ○Recurring-revenue monetisation (VPP/capacity payments) not default-enrolled at point of sale in most US states, almost nowhere in the EU
- ○New-construction bundling channel frozen by the housing volume crash
- ○Installer/electrician labour supply remains the binding residential throughput constraint
- ○Regulatory settlement on who pays stranded grid cost
- ○Standardised, insurable, financeable resilience-as-a-service product for SMBs
- ○A second enterprise anchor transaction converting the category from single-buyer anomaly to market
Scenarios
One or two vivid multi-day grid failures in a major US metro or EU member state during 2027–2029 — plausibly attributed in media to data-centre load — converts reliability into a political and status consumption category. Retail tariffs rise 30–50% cumulatively on grid-hardening and transmission capex, widening the spread against self-generated kWh; VPP/capacity enrolment becomes default at point of sale and turns the battery into a cash-flowing asset; plug-in battery kits and smart panels collapse installed cost below $6k. Enterprises complete the shift they have already contracted: BTM gas, fuel cells and nuclear PPAs clear +15pp of large-enterprise load and hyperscalers become de facto IPPs. All three thresholds clear; unit shipments run 2.5–3x 2025. Value accrues violently to firm-MW owners, gas/LNG supply, electrical equipment and the installer channel.
The enterprise leg clears decisively and the consumer leg clears partially. Corporate BTM/firm-PPA coverage rises meaningfully (roughly +8 to +14pp, short of the formal threshold in aggregate but well above it for the hyperscaler/data-centre subset), and outage-preparedness search stays structurally elevated (index 110–135) because AI siting fights and tariff increases keep the topic salient. Residential/commercial unit shipments grow strongly but at a 1.4–1.8x level rather than 2.0x, throttled by installer labour, 5%+ financing costs and a frozen new-build channel. Thesis is 'directionally confirmed, formally unmet' — reliability is priced as a private good at the enterprise and affluent-household level, while the public guarantee persists for the median ratepayer at a higher tariff. Investable outcome is nearly identical to bull for the equipment/fuel/EPC layer; materially weaker for consumer-facing genset and retail names.
AI load growth disappoints relative to contracted plans (financing event, local inference migration, Huawei/China bifurcation), utilities and regulators recapture the load with fast-tracked interconnection, co-location tariffs and explicit data-centre carve-outs that keep large customers on the grid. Residential demand proves storm-spiky and mean-reverts as it did post-Uri and post-2022 EU energy crisis; net-metering rollbacks, self-consumption taxes and air-permit limits bite; the 5% rate regime and the housing volume crash suppress household capex. Search interest normalises within 18 months of the last salient outage. Unit shipments end 2031 at 1.0–1.3x 2025 and the thesis fails on both the primary and secondary legs — with the specific loser being anyone who capitalised the 2026 order book as a run-rate.
Indicators
- ·Hyperscaler genset/turbine order announcements beyond the Amazon–Generac deal (GEV, CAT, Cummins, Bloom backlog and delivery-slot pricing)
- ·Aeroderivative turbine and reciprocating engine lead times extending past 2029 — the cleanest scarcity read
- ·Rising share of Generac/Kohler/Briggs revenue from C&I and data-centre channel vs residential storm demand
- ·EU BESS residential install run-rate re-accelerating despite subsidy withdrawal (proof of unsubsidised economics)
- ·Retail electricity tariff increases approved in PJM, ERCOT, California, Germany and Ireland — the spread that funds defection
- ·Corporate 10-K/CSRD disclosure of BTM capacity and ≥5-year firm PPAs; LevelTen/BNEF contracted volume by nuclear and gas
- ·Google Trends 12m MA of the fixed outage basket holding above 120 without a storm catalyst
- ·Home Depot/Lowe's SKU velocity in generators, transfer switches and portable power stations in non-hurricane quarters
- ·VPP enrolment counts and capacity-market revenue per installed residential battery
- ·Insurers offering premium credits for on-site backup (the financialisation signal)
- ·Interconnection queue times falling materially in PJM/ERCOT/Germany — removes the schedule motive for BTM
- ·State or EU-level self-consumption taxes, net-metering terminations or standby charges that destroy household payback
- ·Air-permit and runtime-hour restrictions on data-centre gensets spreading beyond California
- ·Generac/Kohler residential order intake mean-reverting within 12 months of the last storm
- ·Hyperscaler capex guidance cuts or further force-majeure/cancellation notices indicating the load never arrives
- ·Local inference migration (Apple cost-per-token claim, edge AI query surge) measurably slowing data-centre power procurement
- ·Utility-offered 'resilience-as-a-service' products re-capturing customers inside the regulated rate base rather than outside it
- ·Whether a vivid, televised multi-day outage in a major US or EU metro occurs before 2029 — the salience trigger that moves the consumer leg from early adopters to early majority
- ·The path of retail electricity tariffs: grid-hardening and transmission capex recovery is what mechanically funds private defection
- ·Cost of capital — a 5% risk-free rate is the single largest suppressant of household resilience capex; a rate regime change is the biggest swing factor on the unit-sales threshold
- ·Whether regulators socialise or allocate data-centre grid costs; allocation keeps hyperscalers on the grid, socialisation triggers ratepayer backlash and accelerates defection
- ·Real AI load delivery versus contracted load — the difference between a demand event and a financing event
- ·Electrician/installer labour supply in the US and EU, which caps residential throughput regardless of demand
- ·Gas and LPG fuel price stability — the operating-cost leg of the household payback calculation
Value pools
| Layer | Emerging need | Opportunity | Becomes | Tickers |
|---|---|---|---|---|
Firm dispatchable generation assets (existing nuclear, CCGT, peakers) | Contracted, schedulable MW with a delivery date inside 36 months | Repricing of every merchant MW from a commodity to a long-dated contracted annuity; nuclear PPAs signed at multiples of prior power-curve assumptions | becomes scarce | |
Power generation and electrical equipment OEMs | Turbines, gensets, switchgear, transformers, transfer switches, busway and thermal management delivered on a schedule, not a price | Delivery slots become the product; multi-year backlogs with price escalators and customer prepayments; margin expansion from allocation rather than volume | gains pricing power | |
Engineering, construction and skilled trades | Physically building substations, BTM plants, interconnections and residential installs faster than the utility queue | Labour scarcity converts EPC and mechanical contracting into a pricing-power business; the electrician is the bottleneck of the entire thesis | becomes scarce | |
Fuel supply and midstream | Firm molecule delivery to BTM plants and residential propane/nat-gas standby units | Long-dated firm gas contracts and lateral pipelines to data-centre campuses; LNG-to-power packages sold as reliability rather than commodity | mandatory infrastructure | |
Consumer distribution and retail channel | Households buying portable power stations, standby units, transfer switches and install services | High-ticket, high-attach, service-linked category growth in exactly the retailers the market has marked for the housing recession; also fits the repair-and-stay-put behavioural shift | gains pricing power | |
Storage, solar and grid-edge hardware | Behind-the-meter kWh with VPP monetisation | LFP + hybrid inverter + smart panel bundles; the asset that is both insurance and an income stream once aggregation is default-enrolled | becomes abundant | |
Grid orchestration and dispatch software | Manufacturing headroom on a constrained grid; flexible-load contracts; VPP dispatch across millions of edge assets | A thin, high-multiple software layer on a physical constraint — Emerald-AI-type products, utility DERMS, industrial load flexing | becomes scarce | |
Raw materials | Copper, steel, aluminium and industrial gases for duplicated generation and distribution infrastructure | Private redundancy means the same load is wired twice; structurally raises metal intensity per delivered kWh | becomes scarce | |
Regulated utilities with residential-heavy rate base | Recovering grid-hardening capex from a shrinking, poorer remaining customer base | Rate-base growth is real but political risk rises; the ones that win sell resilience-as-a-service inside the rate base rather than fighting defection | loses pricing power | |
Data-centre real estate and colocation | Power procurement becomes the primary underwriting risk, ahead of land, capital or tenant credit | Operators with secured firm power and BTM capacity re-rate; those without become land banks | new risk | |
Insurance and financing wrapper | Underwriting outage loss, battery fire risk, installer quality and financing $10–20k household assets at 7%+ | Premium credits for backup, resilience-linked mortgage products, lease/PPA structures for households — the layer that converts a capex decision into a subscription | mandatory infrastructure | |
Nuclear components and SMR supply chain | Firm carbon-free MW for corporate PPAs beyond the existing fleet | Long-cycle order book with sovereign and hyperscaler counterparties; Italy's ban reversal is the template for EU restarts | becomes scarce |
The consensus trade is 'buy the generator company'. That is the first-order, and it is the leg most likely to mean-revert. The second-order insight is that what is actually being repriced is not electrons but *time-to-power* — a firm MW with a 2029 delivery date is now a different, scarcer asset than a MW with a 2034 date, and every layer that shortens the delivery date (turbine slots, switchgear, electricians, EPC capacity, permitted brownfield sites, existing nuclear) captures the rent, while every layer that merely supplies energy does not. This is why the equipment and trades layer, not the utility or the fuel, is where pricing power migrates. The deeper second-order effect is fiscal and political: each defector removes a high-margin customer from the pool funding universal-service reliability, raising tariffs for those who remain, which accelerates the next defection — a classic adverse-selection death spiral identical to the mechanism that hollowed out public schooling, municipal water (bottled water), and policing (private security). That spiral does not resolve in the market; it resolves in regulation. Expect the binding 2028–2031 variable to be a political settlement over who pays stranded grid cost — standby charges, exit fees, self-consumption taxes, data-centre cost-allocation dockets — which makes regulated utilities a political-risk asset rather than a bond proxy, and makes the residential leg of the thesis structurally throttled at exactly the moment the economics finally work. Third order: because the installed base ratchets (sunk capital, fixed fuel connection, property-tied financing), the level is permanent even if the flow disappoints, which means the durable investment is in the *service and fuel annuity* on an installed base — maintenance, propane/gas delivery, VPP aggregation, firmware and warranty — not in the unit-shipment cycle the market currently trades. Finally, reliability becoming a purchased good is a regressive event: it creates a two-tier electricity system in which the wealthy and the corporate buy availability and the median ratepayer buys a more expensive, less reliable grid. That is the single most under-priced input to 2029–2031 utility regulation, retail energy politics, and the value-retail/consumer-bifurcation theme.
- ·Regulated vertically-integrated utilities with residential-heavy, low-growth rate bases facing defection by their highest-margin customers plus rising political risk on cost allocation (DUK, SO, AEP, and NEE's regulated arm)
- ·Net-lease and duration-proxy REITs already pressured by a 5%+ long bond and now facing higher operating energy costs and resilience capex (O, SPG)
- ·Data-centre REITs and colocation operators without secured firm power or BTM capacity, whose land banks cannot be energised on schedule (EQIX, DLR)
- ·Second-tier neoclouds and equity-funded compute landlords — the AI drawdown arrives as a financing event and they cannot outbid hyperscalers for firm MW (ORCL's data-centre development arm is the listed proxy)
- ·Homebuilders, doubly hit: frozen turnover removes the bundling channel while rising grid-connection and resilience costs add to cost per unit (LEN, DHI, PHM, TOL)
- ·Low-income and renter households — structurally excluded from private reliability while bearing the tariff increases funding grid hardening; bearish for levered-consumer credit exposure (AXP, SOFI) and supportive of value retail
- ·Diesel-dependent backup vendors facing tightening air-permit runtime limits in California and under the EU IED
- ·Pure commodity power merchants with no firm contracting or nuclear/gas optionality — the value moves to contracted, not merchant, MW
- ·Standalone battery and solar hardware vendors without VPP monetisation or installer networks — hardware becomes abundant and commoditised while installation and dispatch stay scarce
Probability history
Updates log
| Date | Before | After | Δ | New | Retired | What changed | Should change mind? | |
|---|---|---|---|---|---|---|---|---|
| 2026-10-01 | 92.4% | → | 80.0% | -12.4 | 8 | 2 |
| Down, modestly — roughly 3-5 points, to the high-80s. The new evidence is almost entirely on the contradicting side and it lands on the thesis's weakest joint: metric (i), the 2x unit-sales threshold, is mandatory, and it is the household/SMB leg that must deliver it. Two of the three named 'biggest variables' moved adversely within one fortnight: cost of capital (10-year 5.15% to 5.33%, 30-year above 5.6%, gilts above 6%) and fuel price stability (distillate crisis, Brent ~$100). Crucially the rate move is now explicitly a term-premium/supply story that survived a soft core PCE — meaning it is less likely to be a transient suppressant that reverses before 2029, which was the implicit bull assumption. Consumer confidence at a 12-year low with 9.3m student-loan defaults upgrades the bifurcation concern from inference to fact. Nothing new arrived on the supporting side except a foreign geopolitical salience event and the continued absence of a second anchor deal. The enterprise leg (metric ii) is unaffected in substance but the AI-credit-cycle scrutiny marginally raises the chance that contracted load is partly a financing artefact. I would also flag honestly that the standing posterior of 92.4% looks too high for a conjunctive claim requiring a 2x unit threshold AND one of a +15pp enterprise shift or a sustained search index — the prior was set near the reference-class base rate for 'mechanism completes', not for 'these three specific numeric thresholds clear by a fixed date'. This month's evidence is the right occasion to correct that downward rather than defend the level. |