A battery inside a kitchen range sells time: years before a home or a building must pay for a bigger wire. On the draft's own numbers that time is worth less than the battery almost everywhere, except in kitchens no load controller can reach. On an AI campus the same equation runs the other way, and time is the dearest thing there is.
Copper's battery-equipped induction range, the Classic Charlie, lists at $5,999, against the $2,000 the draft behind this note assumes for a comparable conventional range, and the electricity its battery stores is worth little: in one California field study the best time-of-use arbitrage came to $71.08 a year. What the battery sells is time. Running from an ordinary 120-volt outlet, it lets a household or landlord put off the electrical work a conventional range would need, until the battery wears out or another appliance forces the work anyway.
On the draft's own inputs (a $5,000 upgrade, a ten-year battery, money at 8% a year) the time falls short of its price. Even with nothing else ever forcing the upgrade, the postponement is worth $2,684.03 against a hurdle of $4,083.29, a loss of $1,399.26; from the household's side, the battery path is a loan at 16.4% a year. It pays only where the avoided work costs more than $7,606.63. That rules out the typical house, where a metered load calculation or code-recognized load management is cheaper, and leaves the older apartment kitchen with no 240-volt circuit, where no controller can help.
The same equation, with its terms reversed, explains why batteries and on-site power matter to AI data centers. There the indivisible upgrade is a substation years away, and what decides the case is the months a bridge pulls forward, not the bill it postpones: in an illustrative 100-megawatt case, a $150 million bridge repays a chip owner in under three months and a landlord in a little over eighteen. Where that points in listed markets is set down as hypotheses, each with its falsifier.
The battery began with a quarrel over frogs. In 1791 Luigi Galvani, an anatomist at Bologna, published his account of dissected frogs' legs twitching at the touch of his scalpel, which he took for a vital force he called animal electricity. Alessandro Volta repeated the experiments, read them differently, and in 1800 wrote to the president of the Royal Society in London to describe an apparatus that gave shocks like a weakly charged battery of Leyden jars, except that its charge, after each discharge, restored itself. Gaston Planté's lead-acid cell, invented in 1859 and introduced in 1860, was the first practical rechargeable battery, and it still starts petrol cars. Carl Gassner patented a dry cell in 1887. Lithium-ion took three more steps: Stanley Whittingham's lithium battery in the 1970s, John Goodenough's cobalt-oxide cathode in 1980, which doubled its voltage, and Akira Yoshino's carbon anode in 1985, which made it practical. Sony began selling the cells in 1991, and the three chemists shared the 2019 Nobel Prize in Chemistry. Goodenough later replaced the cobalt with iron phosphate, the chemistry in Copper's range.
The curve is the useful part. Battery cost has followed Wright's law (Wright, 1936): each doubling of cumulative production cuts unit cost by a roughly fixed share. Lithium-ion cells fell about 97% in real price per kilowatt-hour from 1991, 13% a year to 2016 and 20% per doubling of cumulative market size (Ziegler and Trancik, 2021). Against the calendar that looks exponential, because production grew exponentially, which is why Wright's law and the exponential forecast almost equally well across 62 technologies (Nagy, Farmer, Bui and Trancik, 2013). Adoption traces an S-curve (the Library's Bass diffusion); energy density improves more slowly than price. In December 2025 BloombergNEF put the average pack at $108 a kilowatt-hour, and iron-phosphate packs at $81.
Here the history pays off. At $108 a kilowatt-hour, the cells in Copper's 5-kilowatt-hour battery are worth about $540, 13.5% of the $3,999 premium. The other $3,459 is everything that is not a cell, and it falls only with the appliance's own volume. For the ten-year bridge to break even on a $5,000 upgrade, the premium must fall 35%, to $2,599.74: 2.65 doublings of the stove's cumulative output at a 15% learning rate, 1.93 at 20%, both assumptions. The cell curve has done most of its work. The stove's own curve has hardly started.
The kitchens at Emerson Arms, a 32-unit affordable housing complex in Martinez, California, have no 240-volt circuit and no dedicated circuit for appliances. When its gas ranges came out, in an electrification project that also brought ductless mini-splits, a central heat-pump water heater and electric dryers, the replacements were induction ranges plugged into the kitchens' shared 120-volt, 15-amp circuits, each with a 5-kilowatt-hour battery in its base. From November 2024 to May 2025 the 23 households who agreed to be measured cooked through 6,175 cooking events, and only one drew a battery below 1 percent (CalMTA, report MR26-009).
The ranges worked. The puzzle is why anyone buys one, since the stored electricity is worth little: a CalNEXT study of 36 battery stoves on California's north coast put the best arbitrage under 2024 time-of-use rates at $54.60 to $71.08 a year, and $71.08 a year for ten years is worth $476.95 at 8%.
The answer, and the question this note asks, come from Casey Crownhart's column in MIT Technology Review on October 1, "How smaller, distributed batteries could help the grid". She followed startups putting small batteries into e-bike swapping cabinets, food carts, air conditioners and stoves, and noted that a battery stove could help homeowners avoid expensive electrical upgrades. Copper's chief executive, Sam Calisch, told a Climate Week event the battery is not only for resilience: "we’re selling it for cost savings."
Russ's draft turned that sentence into a hypothesis: appliance batteries sell the avoidance of indivisible electrical upgrades, not stored kilowatt-hours sold cheaply, so a model of the shadow price on connection capacity should find the buyers just across an upgrade threshold. The Vista Research Desk challenged it over several rounds, with corrections in both directions. What survived is narrower and more useful: avoidance has a lifetime, and the cheapest way to stay under the limit may not be a battery.
Start with the physics, which is not in dispute. A 120-volt, 15-amp household circuit is held to 1.44 kilowatts, the limit the CalMTA study takes from UL 858, the household-range standard; the tested range could draw 10 kilowatts from its battery at peak. The battery is a buffer between a small, steady inflow and a large, intermittent outflow, like a water tower. Because the outlet keeps charging during cooking, the study's busiest day, 14.5 kilowatt-hours on one range, 2.9 times its battery, measures throughput, not cycles, as Russ reminded the Desk.
The economics begin where the physics ends. A building's electrical capacity is a constraint, and every binding constraint has a shadow price, the value of one more unit of it (Kantorovich, 1960; the Library's constrained optimization). For most homes the shadow price on their connection is zero: they have headroom. For a home just over a threshold it is large, and lumpy. One more amp is not for sale, only the next panel, the next service, the next transformer. So a battery's value is not its kilowatt-hours times a price but the difference between two fully worked strategies, one that crosses the threshold and pays, and one that stays below it.
And the upgrade is not avoided, only postponed. That makes the battery a real option of the kind McDonald and Siegel (1986) analyzed: the household holds the right to make an irreversible investment, and pays to keep the right alive (real options). The option ends at the earlier of the battery's end of life, H years out, and the day another load, an electric car, a heat pump, a renovation, forces the upgrade anyway. If that load arrives at a steady yearly hazard h, the postponement is worth
where U is the cost of the work the range would otherwise trigger, ρ = ln 1.08 is the continuous form of the 8% discount rate, h is the yearly chance that another load forces the upgrade, and H is the battery's life. It is the precise version of the story: money paid later is cheaper, but only until the later date arrives. Add D, the value of the delay a slow upgrade would have caused, subtract K, the battery's premium and running costs, and the battery wins when
where NPVA is the value of the best compliant alternative. The second condition is the one most stories leave out. Hold on to the three terms, B, D and K; on an AI campus they trade places.
The formula is the same everywhere; its inputs divide the market by the constraint that binds.
| The constraint that binds | The cheapest compliant alternative | What the battery does |
|---|---|---|
| No 240-volt circuit reaches the kitchen, common in older apartments | A new range circuit in each unit, often with the panel and building capacity to feed it | Avoids the circuit for its whole life; load management cannot create one |
| The home's panel or service, as calculated | A load calculation from a year of meter readings; load management at $599 to $7,550 installed | Avoids it too, but so do they, for less |
| The building's service, as in Martinez | One shared upgrade, split across the units | Only with a battery in every unit, and only until other loads force the upgrade |
| A campus's grid connection, years away in a queue | Waiting, or a connection that accepts curtailment in a few hours a year | Rides through those hours; round-the-clock bridging needs on-site generation |
Costs from CalNEXT's load-balancing pilot (ET24SWE0063), which also reviewed the code; Martinez from CalMTA.
This reminds me of the eleventh chapter of the Tao Te Ching, which observes that clay is shaped into vessels, yet "it is on their empty hollowness, that their use depends" (Legge's translation). Electrical capacity is valued the same way: what a household pays for is headroom, the room left under the limit, and a battery in a range makes room without making the vessel bigger. Like all room, it is worth most to those who have none.
The draft prices one illustrative customer on scenario inputs, not averages: the $5,999 range against a comparable $2,000 one, a $3,999 premium; $100 a year of net running costs for ten years; six months of avoided cash delay at $100 a month; money at 8% a year; and a $5,000 upgrade. A 30% federal battery credit cut the premium until it ended for ranges installed after 2025. The hurdle is the premium plus running costs, less the delay:
where $671.01 is ten years of running costs at 8% and $586.72 the present value of the delay. Two tempting answers are withdrawn. The Desk's first survival model let the avoided upgrade run forever against ten years of costs; Russ caught it, and its figures are gone. His own headline, a permanently avoided upgrade worth $916.71, made the same error from the other side, as the Desk's second review found: it survives only if the battery side is renewed in year 10 for less than $1,979.11 with nothing charged after, and a battery that never wore out, its running costs charged as long as the avoidance lasts, would leave $337.72. Permanent avoidance is a series of renewals nobody has priced.
What remains is a bridge. The corrected arithmetic is Russ's own:
| Yearly chance another load forces the upgrade | 0 | 5% | 10% |
|---|---|---|---|
| Chance it happens within ten years | 0% | 39.35% | 63.21% |
| Years the battery buys, on average | 10.00 | 7.87 | 6.32 |
| Value of waiting, B, in dollars | 2,684.03 | 2,179.39 | 1,803.98 |
| Less the hurdle of $4,083.29 | −1,399.26 | −1,903.90 | −2,279.31 |
| Upgrade cost that breaks even, in dollars | 7,606.63 | 9,367.97 | 11,317.43 |
| The battery path as a loan, interest a year | 16.37% | 24.56% | 36.19% |
At zero hazard the upgrade still comes in year 10, when the battery's job ends: ten years of postponement, not permanence. The loan row is the discount rate at which the battery path and upgrading now cost the same.
Even the best column loses: paying $5,000 ten years late is worth $2,684.03 at 8%, short of the hurdle by $1,399.26, and the value only falls as the hazard rises. The question is how much the upgrade costs. The last row is the plainest way to read the trade. Next to upgrading now, the battery path hands the household about $1,600 in its first six months ($1,001 saved on the purchase, $600 of avoided delay) and takes back $6,000 over ten years: a loan at 16.4% a year when nothing else is coming, dearer when something might. For a household that can borrow for less, the battery is the expensive way to wait; for one that cannot borrow, the draft's 8% is a choice, not a fact, and at 20% the zero-hazard case turns positive.
This reminds me of Joseph in Egypt, who stored the grain of seven good years as "a reserve for the land during the seven years of famine" (Genesis 41:36, BSB). The granary worked because the famine had an end. A battery earns its cost the same way, when the shortage it bridges is temporary: a building waiting for the utility, a campus waiting for its substation. Bought to avoid an upgrade forever, it is a granary for a famine without end, and the grain runs out in year 10.
The Desk's second review named three assumptions that flatter the battery or hide the crux. Escalation: a nominal 8% against a flat $5,000 assumes the work gets cheaper in real terms each year it waits; at 3% a year of cost growth, an illustration rather than an estimate, the zero-hazard break-even rises to $10,816.47 and the loan rate to 19.4%. Scope: a later upgrade forced by a car or heat pump is often larger than the range's own, and the battery postpones only the range's share, so U should be that smaller share. Timing: a constant hazard spreads the risk evenly, but the loads that force upgrades arrive on dates, a car purchase, a failed furnace, a lease that turns over. Hold the chance of a trigger within ten years at 39.35% and move it: if all of it falls in years 1 and 2, the break-even upgrade is $11,598.79; in years 2 to 5, $10,170.95; spread evenly, $9,367.97; in years 7 to 10, $8,119.32. The finding is the shape of the hazard, not its average, which is why the evidence calls for Cox's (1972) model, h(t | X) = h0(t) × eXβ, in which a home's facts X (a car on order, an aging furnace) scale a baseline hazard h0(t) free to bunch around dates (survival analysis).
The ten-year life is a judgment too (a five-year warranty; 10 to 15 years by Copper's estimate; more than 20 by CalNEXT's reading of the cell maker's tests, without observed degradation), but none of it rescues a $5,000 case: with nothing else coming, the battery would need to last 31.3 years.
Every figure so far compares the battery with upgrading now, but the draft rightly values each path on its own cash flows: a battery's value is its outcome against the path that would actually have been chosen, and only one is ever observed (Rubin, 1974; potential outcomes). The obvious rival is code-recognized load management. Since its 2023 edition the National Electrical Code has let a listed energy management system's setpoint stand as a service's load (section 220.70), and it lets an existing home's load come from a year of meter readings (220.87), though not every jurisdiction accepts that yet. CalNEXT priced load management installed at $599 to $649 for a plug-in circuit splitter and $5,250 to $7,550 for a smart panel. On the battery's ten-year horizon and zero hazard, with a $500 range circuit, the battery beats a control only if it costs more than $4,401.60: it is $3,752.60 to $3,802.60 worse than the splitter (which shares a circuit, so it can mean no dryer while dinner cooks) and $848.40 to $3,148.40 better than a smart panel that only bridges the same ten years.
Where, then, does the upgrade cost clear the bar? The best public evidence, Lawrence Berkeley National Laboratory's August 2026 report on a survey of 140 contractors, utility staff and building officials, puts the single-family medians at $3,500 for a panel replacement and $2,800 and $3,500 for the customer and utility sides of a service increase, all below the $7,606.63 the battery needs with nothing else coming. Some multifamily projects run to medians of $11,500 to $18,750, but a battery must avoid a whole-building project in every unit at once: thirty-two units at $7,606.63 each come to $243,412, far above the building-project costs the survey reports. The battery pays where the work it avoids is per unit and expensive, a 240-volt circuit run into each occupied apartment: the Martinez kitchen, which no public survey prices.
A saving to the customer is not a profit to the supplier, as the draft insists. Each side's outside option sets the split (Nash, 1950; the Library's outside options). A buyer whose alternative is a $599 splitter will not pay a $3,999 premium for the same headroom; one whose alternative is opening thirty-two kitchen walls will pay a great deal. The largest such buyer has shown its hand. On November 13, 2025, the New York Power Authority, NYSERDA and the New York City Housing Authority announced a contract with Copper to develop a stove for older buildings' standard 120-volt, 20-amp outlets: 100 pilot units and, if they succeed, a purchase of up to 10,000, backed by the housing authority's $30.8 million commitment, about $3,080 a stove, whatever it covers, roughly half Copper's retail price. An institution that runs a competition keeps much of the saving. On September 22, 2026, the city said the pilot units were still to be produced. Elsewhere the saving goes to whoever controls the cheapest path: the household, the electrician who installs a controller instead of a panel, or the utility that skips transformer work.
The same equation governs a far larger machine. An AI data center campus faces the kitchen's problem at ten thousand times a stove's peak power: its indivisible upgrade is a substation or a transmission connection, and the queue for it can run to years. At Grant County PUD in central Washington, the Western Electricity Coordinating Council reported in 2024, a new data center faced a wait of seven to ten years, with 2,000 megawatts of load in the queue, half of it data centers. What changes is the size of the terms. In the kitchen the delay term D was $586.72 against a $3,999 premium, about 0.15 of it, and the battery mostly bought a postponed bill, which lost. On a campus the delay term is nearly the whole story.
Take a 100-megawatt campus, with inputs that are illustrations, not estimates. A bridge of on-site generation costs $1.5 million a megawatt, $150 million, and lets the campus open in month 6 instead of month 24, when the grid connection arrives. A landlord earning $150 a kilowatt-month at a 60% margin gains $9 million a month; at 8% a year the bridge needs 18.4 months of that to repay, a shade more than the 18 it pulls forward, and 10.8 months if 40% of it stays on as backup power. An owner of the chips, whose idle accelerators cost, say, $600,000 a megawatt each month, repays the same bridge in 2.6 months. Five lessons carry over from the kitchen, each changed on the way.
If the campus arithmetic holds, a month of energized capacity can be worth more than the bridge that buys it, and the rent goes to whoever controls the binding constraint. Five hypotheses follow, each with what would falsify it in the companies' own disclosures; none is a recommendation.
In the kitchen story, Enphase Energy sells constraint relief as software: Power Control lets installers avoid main-panel and transformer upgrades by curtailing battery and solar output. That is a capability, not an earnings line.
Likely, we can get some of it from primary research, which is what Vista is all about. People who planned services and large-load connections, quoted kitchen circuits and inspected installations can say now what the record will take years to show.
We would askAs a matter of general past practice, when a home postponed a service upgrade with load management or a low-power appliance, how often did a later load force the upgrade within five years, and which loads did it?
The answer that would change the view"Hardly ever. Once a home found headroom, it kept it."
We would askPer apartment, what did it typically cost to bring a 240-volt range circuit into an occupied unit, including the panel and building work it triggered?
The answer that would change the view"Under $3,000 a unit, building work included."
We would askHow did your office treat battery appliances without a UL 858 certification, and did it accept energy management systems or a year of meter data in place of a service upgrade?
The answer that would change the view"We rarely accepted controllers or metered loads; the upgrade was the default."
We would askIn general past practice, how much sooner could a large load connect if it accepted curtailment for a few hours a year, and what limited it at the substation?
The answer that would change the view"No sooner. The substation was the limit either way."
Every conversation would draw on past, general experience and public examples only, with no current employees of the companies, authorities or utilities named here or their suppliers, nothing confidential, and every Advisor screened for conflicts first. A question no plausible answer could act on is not worth asking (the value of information, Howard, 1966). For a client engagement, Vista would combine this note with those interviews: two or three Advisors for each question that matters, structured conversations, and one written brief that says where they agree, where they split, and what would change the answer.
We will score this note on July 1, 2027, and publish the result on the scorecard whether it flatters us or not. The tests are fixed today, so they cannot drift, and each can fail. For the kitchen, each would side with the broad version of the bet:
For the campus:
In the kitchen group, four or five passes and battery appliances are leaving their niche faster than this note expects; two or fewer and the niche reading holds; three is too early to say. In the campus group, two passes and the scarcity the model describes still binds; none, and it is easing faster than the note assumes.
In Martinez the ranges kept their promise: one cooking event in 6,175 drew a battery below 1 percent. What they bought the building was time. Where time is what a building lacks, the stove is worth its price; where it lacks only room, there are cheaper ways to make it. On a campus the same arithmetic runs the other way, and time is the dearest thing there is. The decision, as always, belongs to the reader.
The bet. That a battery inside an appliance sells its owner out of an electrical upgrade, widely enough and at a high enough premium to build a business: the stove paid for by the wiring it avoids.
The payoff. In the customer's economics, on the draft's inputs (a $5,000 upgrade, a ten-year battery, 8% a year), the battery path loses $1,399.26 if nothing else ever forces the upgrade and $2,279.31 at a 10% yearly chance that something does: a loan at 16.4% to 36.2% a year. It pays where the avoided work costs more than $7,606.63, or $11,317.43 at a 10% yearly hazard. On an illustrative 100-megawatt campus the same equation pays a $150 million bridge back in 2.6 months for a chip owner and 18.4 for a landlord.
Our read. The physics is proven and the arbitrage is small. In homes the economics can work only in the older apartment kitchen with no 240-volt circuit; elsewhere a metered load calculation or code-recognized load management is cheaper, and the saving goes to whoever holds that outside option. On campuses the months decide, not the bill: a bridge pays where queues are long and a month is dear, as it is to a chip owner far more than to a landlord, and the rent goes to whoever controls the binding constraint.
What settles it, and when. By June 30, 2027: for the kitchen, New York's pilot stoves in kitchens, UL battery rules for ranges, a Charlie at $4,999, a Power Control figure from Enphase and CalMTA's data; for the campus, GE Vernova's gas backlog at its 125-gigawatt target and Fluence's data center business past $850 million. Before then, we would talk with a former utility service planner about how often postponed upgrades come back, a former electrical contractor about what a range circuit costs in an occupied apartment, and a former large-load interconnection planner about how much sooner a flexible campus can connect.
The draft's model, rebuilt in code and reproduced to the cent before anything was added. The hurdle is the price premium plus ten years of running costs, less the avoided delay: C = (P − P0) + o × A10 − D, with A10 = (1 − 1.08−10) / 0.08 = 6.710081 and D the six month-end delay costs discounted at the effective monthly rate, $586.717045. Another load forces the upgrade at a random time T with hazard h, the battery ends at H = 10, and the upgrade is paid at the earlier of the two; the benefit is U × (1 − E[e−ρτ]), which reduces to the formula in the note, with ρ = ln 1.08. Running costs continue to year 10 even after an early upgrade, because the range stays in service; a common appliance replacement at year 10 cancels between the paths. The loan rate is the annual rate at which the battery path's cash flows against upgrading now (+$1,001 at the start, +$100 at each of the first six month-ends, −$100 at each year-end for ten years, and −$5,000 at the upgrade, in expectation) are worth zero. The timing figures use a piecewise-constant hazard set so that the chance of a trigger within ten years is 39.35% in every case, integrated numerically; the constant case reproduces the closed form to the cent. The escalation case grows the upgrade cost at 3% a year, which at zero hazard multiplies the deferral by 1 − (1.03 / 1.08)10 = 0.37751. The battery-life case charges running costs for as long as the battery lasts. The control path pays a $500 range circuit and the control now and postpones the remaining $4,500 on the battery's own terms; it is credited the same avoided delay. The figures the draft withdrew, from an infinite-horizon benefit paired with ten years of costs, were reproduced and are not used. The battery-curve figures price 5 kilowatt-hours at BloombergNEF's December 2025 average pack price of $108; the premium at which the bridge breaks even is B(0) less running costs plus the delay, 2,684.03 − 671.01 + 586.72 = 2,599.74 dollars; the doublings are ln(2,599.74 / 3,999) / ln(1 − learning rate), at learning rates of 15% and 20% that are assumptions, not estimates. The campus figures are illustrations: 100 megawatts, $1.5 million a megawatt, a bridge from month 6 and the grid from month 24, each month's value received at month-end and discounted at 8% a year; the landlord's month is $150 a kilowatt-month at a 60% margin, the chip owner's $0.6 million a megawatt-month; keeping 40% as backup means charging only 60% of the bridge's cost to the months it buys. The months-moved case compares a fifteen-year stream that starts in month 6 with the same stream starting in month 24. The battery for curtailment events is sized at half the campus's load for two hours and priced at BloombergNEF's 2025 stationary pack price, before installation. The National Electrical Code's own text sits behind a sign-in we did not use; the provisions cited are as stated in CalNEXT's code review and in Lawrence Berkeley National Laboratory's summary for the Department of Energy. Company figures are as the companies reported them, and the arithmetic on them is recomputed in the model. Every figure in this note was recomputed in code from these inputs.
| Input | Value | Source |
|---|---|---|
| Battery range list price | $5,999 (Classic); $6,799 (new) | copperhome.com, read October 3, 2026 |
| Comparable conventional range | $2,000 | The draft's assumption |
| Net running costs | $100 a year, ten years | The draft's assumption |
| Avoided delay | $100 a month, six months | The draft's assumption |
| Upgrade cost U | $5,000 | The draft's illustration; survey medians from LBNL |
| Battery life H | 10 years | The draft, on Copper's 10 to 15 year expectation |
| Discount rate | 8% a year, effective | The draft's assumption |
| Hazards | 0, 5%, 10% a year | Sensitivities, not estimates |
| Load management, installed | $599 to $7,550 | CalNEXT ET24SWE0063, Tables 21 to 23 |
| Range circuit | $500 | CalNEXT ET23SWE0064, as modeled there |
| Battery pack price | $108 a kWh (average); $81 (LFP); $70 (stationary) | BloombergNEF, December 9, 2025 |
| Appliance learning rate | 15% and 20% per doubling | Assumptions |
| Campus | 100 MW; bridge $1.5 million a MW; months 6 and 24 | Illustration |
| Value of a month | $9 million (landlord); $60 million (chip owner) | Illustration |
Two smaller corrections to the record. In the CalMTA report, the figure shows 32.1% of oven-only events within a 120-volt circuit's capacity where the text says 33.6%, and the 94.9% of events a 4 kilowatt-hour battery would cover sits beside a sum of shares, 5.3%, that implies 94.7%; weighted by event counts, the share above 4 kilowatt-hours is 0.80%. Neither changes the note.
This note is research, not investment advice. It states what the economics assume under labeled assumptions; the hypotheses about listed companies are statements to be tested, not recommendations, and nothing here is a recommendation to buy, sell or hold any security or to buy any product. The decision belongs to the reader. As of October 3, 2026, Russ Rosenzweig, Vista's founder, owns no shares of American Electric Power, Bloom Energy, Caterpillar, CoreWeave, Digital Realty, Dominion Energy, Eaton, Enphase Energy, Fluence Energy, GE Vernova, Murata Manufacturing, Oracle or Sony Group. Copper provided the ranges at the CalMTA test site and shared the cost of the installation pilot; the report states that Copper did not fund or sponsor the study. The note began as Russ Rosenzweig's research question and draft, which the Vista Research Desk challenged over several rounds, with corrections in both directions, and which he then asked us to extend to data centers; the models and every figure were rebuilt in code and checked against the primary sources.