The third energy argument

“Energy is the only truly universal currency, and nothing (from galactic rotations to ephemeral insect lives) can take place without its transformations.”

In 2025 the International Energy Agency expects capital flowing into clean energy technologies and infrastructure to reach about 2.2 trillion dollars — roughly twice the roughly 1.1 trillion dollars going to oil, gas, and coal.¹ Global energy investment as a whole is set near 3.3 trillion dollars. That is a capital-flow measurement, not a forecast of system substitution. In the same decade, electricity still accounts for only about one-fifth of global energy use.⁵ The scientific task is to keep those objects separate: investment rates, generation growth, electricity’s share of energy, power density, end-use conversion efficiency, and the waste streams each stack leaves behind.

What follows puts Vaclav Smil’s quantitative map of density and historical transition speed in the same frame as manufacturing cost curves for solar, batteries, and electrified demand — without collapsing one into the other.² Waste is in the frame, not afterthought. Dual account is a filter, not a menu: many mixes can respect density and learning. The extra criterion I use is firm low-carbon power that can actually be delivered for dense loads, at a cost and schedule institutions can clear, while the manufacturing stack scales. Under that test my working path is nuclear plus solar (with storage and grids) through the multi-decade window — life extension and existing plants count, not only greenfield units — and, if storage, process heat, and end-of-life loops clear, solar plus batteries as the dominant electricity stack, with residual fuels and feedstocks still a separate molecule problem. Path hypothesis, not inventory.

Two incomplete models

One common model treats the energy system as essentially hydrocarbon extraction plus thermal conversion. Its correct content: modern primary energy is still largely fossil; power density of concentrated thermal systems is high; steel, cement, ammonia, and long-haul freight require high-grade heat and chemical feedstocks that diffuse solar flux does not supply at the same spatial intensity. On that model, annual solar capacity additions can look small next to primary energy totals measured in exajoules.

Another common model treats the energy system as an emissions-abatement problem whose master variable is cumulative carbon. Its correct content: combustion releases carbon dioxide; the cumulative stock in the atmosphere is the climate-relevant integral; delay has physical consequences for concentration pathways.² On that model, any transition metric that is not yet consistent with a mid-century carbon budget can be scored as failure even when module costs and deployment rates are changing quickly.

Both models omit a third, empirically active layer: modular electricity technologies whose costs fall with cumulative manufacturing, and end-uses whose useful work per unit primary energy rises when combustion is replaced by motors and heat pumps. Walter, Butler-Sloss, and Bond call this stack electrotech — factory-built electricity generation, electric end-uses, and storage — and distinguish it from a broader “cleantech” bag that mixes high-learning modules with low-learning process schemes.³ The rest of this note is that distinction, stress-tested against Smil, with waste streams kept on the scorecard.

Give each incomplete model its best number

Before stacking electrotech onto Smil, state each incomplete model at full strength — otherwise the dual account is a slogan.

Best pure-density objection. Field-scale solar and wind deliver energy at power densities that are orders of magnitude below concentrated thermal plants when measured in watts per square metre of footprint. Smil’s tables put many thermal converters in the hundreds of watts per square metre while utility solar fields sit nearer single digits to tens, once roads and spacing count.⁴ Replacing a gigawatt of firm thermal capacity is not “install a gigawatt of nameplate solar.” That said, land is not always the binding constraint — grids, permits, minerals, and politics often are. Density bites hardest for continuous industrial loads and dense cities, not every desert array.

Best pure-learning objection. Module and cell prices have fallen well over 90% across multi-decade series as cumulative production rose (Table I). Clean-side capital already runs near twice fossil supply investment (Fig. 03–04). Electric cars take more than one-fifth of new world car sales and roughly half of China’s (Fig. 06).⁷ Measuring only the fossil stock understates the rewiring. Measuring only module prices understates firm capacity, interconnection, and winter adequacy — pack cost is not system cost.

Both objections are right about their objects. The dual account refuses density erasing learning or learning erasing density. It does not by itself pick nuclear plus solar over every other firm-plus-modular mix; that selection needs cost, build rate, and institutions as well.

Define the measured stack

Electrotech, for measurement purposes: (1) electricity generation that is increasingly modular and factory-built (notably solar panels and wind turbines); (2) end-uses that convert electricity to work or heat without burning fuel on site (motors, heat pumps, electrolytic processes where applicable); (3) batteries and digital control that time-shift and coordinate power. That is a scope boundary, not a brand.

Exclude from the electrotech scorecard, unless they show comparable manufacturing learning: carbon capture as a mainline path, hydrogen from fossil gas with capture as a default fuel, and biomass as bulk primary replacement. They remain live residual options for hard heat, feedstocks, and seasonal firming where electrons alone fail; keeping them off the module curve is bookkeeping, not a claim they cannot win regions.³

Electrification itself is not new. Electric motors and lighting reorganized industry and households for more than a century; semiconductor manufacturing later supplied process precision that solar and battery factories reuse.³ Continuity of industrial technique helps explain speed of module scale-up. It does not cancel power density or process heat requirements.

Smil’s quantitative constraints

Vaclav Smil’s program — Power Density, Energy Transitions, Energy and Civilization, How the World Really Works — is a constraint engine: quantify fluxes, refuse unmeasured claims, and keep industrial materials in view.⁴

Power density (watts per square metre of energy flow) sets spatial requirements. Concentrated thermal conversion and fossil extraction sites operate at densities that field-scale solar and wind do not match without large collection areas and supporting infrastructure. High-density demand (cities, continuous industrial loads) therefore cannot be treated as a drop-in replacement problem without accounting for land, transmission, and storage.⁴ MacKay’s national-scale order-of-magnitude method is the same discipline applied to a single country’s budget: physics first.⁴

Historical transitions are multi-decade and multi-fuel. Coal did not zero wood on a factory timescale; oil did not zero coal. New converters stack onto old ones. Equating a decade of module learning with a decade of primary-energy substitution is a category error.

Four material pillars — steel, cement, plastics, ammonia — dominate modern material throughput and require high-temperature heat and chemical routes that electricity is only beginning to reorganize at scale (direct reduction, electric kilns, electrified steam cracking research, electrolytic hydrogen for ammonia). Process thermodynamics, not narrative, sets the bound.

Electricity remains a minority of global energy. In the Our World in Data energy dataset, electricity’s share of global energy is about 19.3% in 2023, up from 13.6% in 1990.⁵

Electricity share of energy 1990–2023

Electricity share of global energy — 1990 13.6% → 2010 16.4% → 2020 18.9% → 2023 19.3%. Hover points. Still a minority, rising slowly.

Electricity share of energy (%) Hover points for year values

Claim. Electricity is still only ~⅕ of global energy — rising, not dominant. Mechanism. Electrification of end-uses + generation mix change; multi-decade stock turnover, not a one-year flip. Source. Our World in Data electricity_share_energy (World): 13.6% → 16.4% → 18.9% → 19.3%. Primary-energy-equivalent share, not final useful services. n. 5.

Combustion conversion is inefficient for many end-uses. Order-of-magnitude engineering comparisons put on the order of two-thirds of energy input to fossil systems as lost before useful work; battery-electric cars use roughly threefold less energy per kilometre than petrol cars; heat pumps deliver roughly two- to fourfold more heat per unit energy than simple electric resistance or combustion boilers in suitable climates.⁶ Those are conversion efficiencies for defined services. They do not map one-to-one onto blast furnaces or cement kilns.

Inst 02 CONVERT

Schematic: ~⅓ of fossil input becomes useful work vs battery-electric cars (~3× less energy per km) and heat pumps (~2–4×). Toggle services.

System losses Cars Heat

Combustion baseline Electric end-use

Claim. Where services can run on electrons, conversion physics often multiplies useful work per unit energy. Mechanism. Avoid thermal losses in engines/boilers; heat pumps move ambient heat, so more heat out than electricity in. Check. Defined services only — not blast furnaces or cement kilns. Toggle modes · n. 6.

Manufacturing learning versus system substitution

Solar modules and lithium-ion cells are factory goods. Across multi-decade industry series, their prices have fallen on the order of well over 90% — learning consistent with cumulative production, not with fuel extraction curves.⁷ Handmer’s synthesis of the frontier emphasizes solar, batteries, and synthetic fuels as a manufacturing stack, with very high global solar deployment rates by the early 2020s.⁸ Naam’s longer solar cost work and Vernon’s point that demand electrification lags supply additions are the same scientific objects: cost trajectories and stock-turnover rates.⁸

Table I pins the learning object to one named series: Bloomberg’s volume-weighted lithium-ion pack price survey. Silicon modules show a parallel multi-dollar-per-watt to roughly a tenth of a dollar per watt decline. Learning can plateau or reverse under commodity and trade shocks; even when it continues, pack price is not interconnection, firm winter capacity, or multi-day firming.

Capital flows track the manufacturing story at global scale. The charts below hold the Agency’s ~2025 levels: clean infrastructure about 2.2 trillion dollars, oil/gas/coal about 1.1 trillion (~2:1), total energy investment about 3.3 trillion.

Clean vs fossil capital 2025

2025: clean ~2.2 trillion $ vs fossil ~1.1 trillion $ (~2:1). Hover bars. Capital flow is not generation mix.

Clean energy & infrastructure Oil, gas, coal

Claim. Clean-side capital is already ~2× fossil supply investment at global scale. Mechanism. Factory goods (panels, batteries, grids, nuclear) attract multi-trillion annual flows. Source. International Energy Agency, World Energy Investment 2025. Investment is not generation mix and not pack price (Table I). n. 1.

Clean capital vs fossil INVESTMENT

Sankey: total energy investment splits into clean infrastructure (~2.2 trillion $) and oil/gas/coal (~1.1 trillion $). Clean then fans into renewables, grids, nuclear. Hover ribbons.

Clean-side capital Fossil supply capital

Claim. Money already prefers the clean stack roughly two-to-one at global scale. Mechanism. Width = annual capital. Clean internal splits are schematic. Not generation mix. Source. Agency “clean” aggregate includes grids and nuclear — broader than panels alone. n. 1.

Learning and capital can move fast; primary energy and materials substitution usually do not. Both facts are empirical. The error is canceling one with the other. The energy Sankey below draws the cut for 2023: primary energy is a slow stock; electricity is about one-fifth of it; solar is a high-growth flow inside electricity. Absolute growth 2010→2023: primary energy about +22%, electricity generation about +40%, solar electricity about ×52 (from ~32 to ~1,664 terawatt-hours).¹⁷

Where primary energy goes 2023

Sankey: world primary energy (~172,000 TWh) splits into electricity (~30,000 TWh, ~one-fifth) and non-electric uses (~four-fifths). Electricity then splits into fossil power (~18,000 TWh) and low-carbon power (~12,000 TWh). Hover ribbons.

Fossil / non-electric Electricity / low-carbon power

Claim. Most energy still never becomes electricity — that is the stock the transition must move. Mechanism. Width ≈ 2023 terawatt-hours. Left = primary; middle = use class; right = power mix inside electricity. Source. Our World in Data energy dataset (World). Primary-energy-equivalent accounting. 2010→2023: primary +22%, electricity +40%, solar ×52. n. 5, n. 17.

End-use conversion and stock turnover

Where services can run on electricity, conversion physics favors electrotech for the same useful output. Measured markets: world electric car sales share above 20% in 2024; China about 40–50% of new cars (new-energy definition in Chinese stats includes plug-in hybrids).⁹

New car electrification 2024

World electric car sales share >20% (2024); China ~40–50% of new cars. New sales are not the in-use fleet. Hover bars.

Electric share of new sales Remainder of new sales

Claim. New-car electrification is already material globally and leading in China. Mechanism. Module cost + policy + manufacturing scale change the sales mix first. Source. Agency Global Electric Vehicle Outlook 2025; Chinese industry 2024 summaries. Sales share is not fleet share. n. 9.

Fleet and building stocks turn over on multi-decade timescales. Electrifying new sales is not the same as electrifying the in-use stock. Industrial heat and chemical feedstocks remain the binding technical frontier for full electrotech substitution. Ausubel’s dematerialization work measures a related quantity: energy and materials per unit of service can fall as systems develop, which is not identical to absolute primary energy or absolute emissions.¹⁰

Worked pillars: steel and ammonia

Smil’s four material pillars are where density and learning collide hardest. Steel and ammonia make the dual account concrete without leaving thermodynamics.

Steel. Most primary steel still routes through blast furnaces that reduce iron ore with carbon, producing molten iron and carbon dioxide as process chemistry, not merely as “energy waste.” Integrated blast-furnace routes typically sit near ~20 gigajoules per tonne of crude steel in final energy (order of magnitude; plant and scrap mix move the number).¹⁵ Scrap melted in electric arc furnaces already electrifies a large and growing share of secondary steel where scrap stock and power exist; that is electrotech meeting an existing material loop. Primary routes that try to exit coal-based reduction need either (i) hydrogen reduction of iron ore plus electric melting, or (ii) other electrolytic or plasma research paths still far from global tonnage. Pure hydrogen reduction requires on the order of ~50–70 kilograms of hydrogen per tonne of reduced iron — relocating the bill to continuous, cheap, low-carbon electricity and electrolyzer capacity at the scale of millions of tonnes of hydrogen.¹⁵ Module learning on solar panels and battery cells is necessary but not sufficient; the plant still needs firm power, high utilization, and ore logistics.

Ammonia. The Haber–Bosch process synthesizes ammonia from nitrogen and hydrogen under pressure and heat. Stoichiometry alone is about 177 kilograms of hydrogen per tonne of ammonia; real plants run slightly higher with losses. Today’s hydrogen is overwhelmingly from steam reforming of methane. An electrolytic path swaps methane for electricity and water: again, the chemistry is solvable in principle, and pilot and early commercial electrolyzer stacks exist, but at roughly fifty kilowatt-hours of electricity per kilogram of hydrogen the power bill is large and must be cheap and abundant for the product to compete as fertilizer feedstock, not only as a climate demonstration.¹⁵ Ammonia is also a candidate carrier for shipping energy; that use multiplies demand rather than removing it. The dual account says: treat electrolytic ammonia as a learning-rate object for electrolyzers and as a density object for continuous industrial power — not as a slogan that “chemicals will electrify because solar is cheap.”

Cement and plastics have their own heat and feedstock maps (kiln temperature, petrochemical cracking, carbon in the product). The point of steel and ammonia is methodological: if the path hypothesis cannot state what happens to reduction chemistry and hydrogen mass balance, it is not yet an engineering claim.

Storage as a duration ladder, not a single product

“Batteries” in the attractor is a family of services with different physics and cost slopes. Collapsing them into one bar chart is how category errors re-enter.

Day-scale (roughly 1–8 hours). Lithium-ion packs already firm evening ramps where solar is large. Table I’s pack series is most relevant here: cycle life, efficiency, and interconnection still matter, but the manufacturing curve is visible in markets.

Multi-day (roughly 10–100 hours). Weather systems outlast a workday. Covering them with lithium-ion packs alone multiplies energy capacity (megawatt-hours) much faster than power capacity (megawatts). Flow batteries, thermal storage, and demand flexibility compete on multi-day cost per megawatt-hour — not on the car-pack price curve. This is the falsifier zone for “solar + batteries drop nuclear tomorrow.”

Seasonal and strategic (weeks to seasons). Winter–summer imbalance, drought years for hydro, and industrial stockpiles are different objects again. Reservoir hydro, hydrogen and ammonia stores, underground thermal storage, and firm nuclear or residual fuel with declining annual hours all appear here. Learning is slower and more site-specific. In this essay, “solar + batteries” means electricity stacks with electrochemical and thermal storage take a rising share of hours only if multi-day and seasonal services clear cost and reliability tests — not that car battery packs solve winter.¹⁶

Table II is explicitly a schematic ladder for teaching: order-of-magnitude levelized storage cost bands, not a bid stack and not a claim that one consulting year applies everywhere. The dual account is the slope. Short duration is already on a manufacturing learning curve; multi-day and seasonal are not the pack-price series in Table I.

A grid vignette: California’s evening ramp

Abstraction is easier to trust when one system has already paid part of the dual account’s bill. California’s “duck curve” — published by the state’s grid operator — is the cleanest public example of high solar penetration creating a steep net-load ramp as the sun sets: midday solar depresses net load; evening demand rises as solar falls. That ramp was met for years mainly by gas and imports. It is now also a battery problem — and opportunity — at day-scale duration.¹⁸

Battery capacity on that California system grew from a few hundred megawatts in the late 2010s to multi-gigawatt scale by the mid-2020s (on the order of about five gigawatts by 2023 and well into double-digit gigawatts by 2024–25 in public tallies). Operator and market reports show batteries charging through high-solar midday hours and discharging into the late-afternoon and early-evening peak — regularly supplying a large share of energy in the early evening in recent years, and on high-storage days reaching multi-gigawatt discharge during the ramp.¹⁸ That is Table I pack learning and the day-scale row of Table II in the field: pack cost and factory scale made short-duration storage operational, and the grid used it.

What the vignette does not show is equally important. Multi-day winter storms, cold-climate weeks, seasonal hydro drought, and continuous industrial heat are not solved by evening lithium-ion. California has mild winters, gas and imports, and rich institutions; batteries reshaped the duck’s neck, not the annual energy budget or the non-electric four-fifths of primary energy. Day-scale firming is real; multi-day and pillars are still open.

Materials and network constraints

Electrotech relocates bottlenecks: from continuous fuel extraction toward minerals, refining capacity, manufacturing throughput, transmission, and distribution. Copper, aluminum, solar-grade silicon, lithium, nickel, graphite, and rare-earth magnets each have mine–refine–factory chains with multi-year lead times. Battery materials can enter recycling loops once stocks are large — but loops need prior stock, collection, and energy for reprocessing. High-voltage grids and interconnection queues limit how fast new generation becomes delivered energy. A cheap module that cannot interconnect is inventory, not power. None of this is resolved by pack price alone.¹¹

Nuclear’s bottlenecks are different but real: licensing, skilled labor, large-component supply chains, multi-year construction. The dual account does not pretend nuclear is modular like solar panels; it treats nuclear as high-density firm power whose deployment rate is a measured constraint, not a preference.

Waste is part of the equation

Density and learning still miss a third measured object if waste is treated as rhetoric rather than mass and isolation. Every energy path produces residues. The dual account I use treats waste management as a first-class constraint — comparable in seriousness to land, minerals, and firm power — not as a moral trump card against one technology alone.

Fossil fuels generate continuous waste in use: carbon dioxide to the air, combustion particulates and nitrogen oxides where controls are weak, ash and sludge from coal plants, produced water and spills from extraction, and mine or wellfield disturbance. That stream is large, ongoing, and mostly diffuse. It is not “free of waste” because the residue is thin and everywhere rather than stacked in a fence-line facility.

Nuclear power produces spent fuel small in volume relative to energy delivered, but long isolation, secure handling, and durable institutions — an intergenerational and political problem, not only an engineering one. Isolation has often failed as politics even where physics is known. My stand: price licensed storage and a disposal path into nuclear’s cost of firm power; do not treat “nuclear waste exists” as unique disqualification while treating fossil air and ash as background.

Solar, wind, and batteries create end-of-life mass — glass, metals, composites, concentrated battery materials. Most is industrial, not high-activity, but scale and toxics still need take-back before the first retirement wave. Pack learning (Table I) without recycling plants is half a stack. My stand: product stewardship with deployment, not after.

I rank waste by controllability and by residual risk structure — concentrated technical waste is not automatically “better” than diffuse fossil waste if isolation never gets built. Still: stop scoring paths as if only one leaves a residue. Fund isolation and take-back at the same seriousness as interconnection.¹⁹

Path, headroom, and instruments

The near-term job is useful terrestrial power while exiting combustion, under density, materials, waste, and deployable cost. Human primary power is about 2×10¹³ watts against roughly 10¹⁶–10¹⁷ watts of sunlight Earth intercepts — large headroom in pure physics, not a technology chooser and not a reason to relax this century’s carbon arithmetic.¹³

Inst 05 · three lessons 3D

1 · Headroom 2 · Density 3 · Path

Three lessons: (1) headroom — humanity ~10¹³ W inside planetary sunlight; (2) density — same continuous power, nuclear pad vs wide solar field; (3) path — Now → nuclear+solar bridge → solar+batteries, residual molecules as a floor. Needs a graphics-capable browser.

Path timeline ~2025 Now
~2025 Bridge
~2035–45 Solar+batteries
~2055–65+

Solar · batteries · equal-power beams Residual fossil · molecules · nuclear body Drag to orbit · lesson 3: scrub the path

Lesson 1. Headroom is huge in pure watts — not a technology chooser. Lesson 2. Same continuous power on both sites: nuclear sits on a tiny pad (~100s W/m²); the solar field needs a pad on the order of 10–100× larger (~1–10s W/m²). Schematic density, not a land ban. Lesson 3. Scrub Now → bridge → attractor. Residual fossil falls; nuclear peaks mid-bridge then eases; solar and batteries rise; residual molecules (heat/fuels) stay as a floor. Teaching schematic, not a forecast.

Working path: nuclear + solar → solar + batteries

Under dual account plus deployable firm low-carbon power, this is the positioning I hold and will revise under the falsifiers below.

Multi-decade bridge: nuclear and solar together where both can actually be built. Nuclear’s role is dense firm power for industry, cities, and seasons when multi-day storage is still expensive — first by running and extending existing plants, then by new units only where cost and schedule are credible. Recent large Western projects often land near ~5,000–15,000 dollars per kilowatt overnight and roughly ~8–15 years from first concrete to commercial operation; some East Asian series have been faster and cheaper, which is the band “credible” must clear against.²⁰ Where new nuclear is blocked, the bridge still needs some firm low-carbon (hydro, residual gas with falling hours, imports, long-duration storage as it clears). Solar is the modular learning engine (Table I). Grids, flexibility, and batteries glue them — day-scale first, multi-day later. Wind stays where resource is strong. Waste isolation and take-back travel with the build.

Electricity attractor I take seriously: solar + batteries as the dominant electricity stack if multi-day firming, process electrification, and recycling clear — not “all primary energy becomes panels.” Non-electric uses are still about four-fifths of primary energy today (~140,000 terawatt-hours-scale); even deep electrification can leave a large residual molecule stack (fuels, feedstocks, hard heat) for decades, via synthetics, captured carbon, or other chemistry.¹⁷ Manufacturing limit of learning curves, not “drop nuclear in 2026” or “car packs solve winter.”⁸

Density forbids a pure storage-limited jump while industrial loads need continuous power. Learning means the next nuclear unit must compete on cost and hours against solar-plus-storage as curves compound. Geography and institutions decide which half of the bridge ships first. The bridge does not deny the attractor; the attractor does not cancel the bridge or the non-electric remainder of the system.

Now → bridge → attractor ~2025–2065

Path hypothesis Sankey. Start ~2025 (fossil residual still large). Mid path ~2035–2045 nuclear+solar bridge. End ~2055–2065+ solar+batteries if storage and process heat clear. Drag the timeline. Schematic only — not a forecast.

Timeline ~2025 Now
~2025 Bridge
~2035–45 Solar+batteries
~2055–65+

Nuclear / residual firm & fossil Solar / batteries

Claim. Start from today’s residual-fossil-heavy mix; thicken nuclear+solar through the bridge decades; finish on solar+batteries only if multi-day storage and process heat clear. Mechanism. Left = supply; right = day-scale, multi-day, and process-heat service. Years on the bar are approximate teaching marks, not commitments. Check. Schematic path hypothesis — not a generation forecast. Drag from Now → Bridge → Solar+batteries.

What would change this positioning

I would drop the dual account if electrotech cost curves reverse for a sustained period without a temporary commodity shock; if minerals and manufacturing prove structural bottlenecks; if the four pillars stay without high-temperature or chemical substitutes at scale; if waste isolation or pack/panel recovery fails at fleet scale; or if fossil primary demand does not respond where electrotech is cheapest and grids can deliver.

I would revise the path if new nuclear never clears cost and schedule outside a few states (then the bridge is solar plus other firm low-carbon, not nuclear+solar as a universal pair); if multi-day and seasonal storage undercut firm nuclear even in the bridge decades; if solar-plus-storage fails real weather years; or if process heat and residual fuels stay cheaper on non-electric routes, including captured carbon where it wins. Field data on build rates, duration cost, interconnection, tonnes of steel and ammonia, and waste politics decide — not headroom.

Density, learning, waste, and deployable cost stay on the table. The personal content is only the joint fit I use until those tests move: nuclear and solar where both can ship; solar and batteries on electricity if the duration ladder, pillars, and end-of-life management clear; molecules as a residual stack, not an afterthought.

1. International Energy Agency, World Energy Investment 2025, executive summary — iea.org. Clean energy aggregate ~2.2 trillion dollars vs ~1.1 trillion dollars oil/gas/coal; total energy investment ~3.3 trillion dollars (2025). 2024 cycle similarly described clean investment as almost twice fossil. “Clean” includes grids, nuclear, efficiency, electrification — broader than solar modules and batteries alone. ↩
2. Atmospheric CO₂ and energy-system emissions as physical objects: IPCC AR6 WG1 physical science baseline; for carbon accounting context on this site see What climate does not entail. ↩
3. Electrotech definition (vs broader “cleantech”), historical electrification, manufacturing drivers: Walter, D., Butler-Sloss, S., and Bond, K. (2025), “Rewiring the energy debate,” The Electrotech Revolution — electrotech-revolution.com. Used for technical scope (electrotech vs mixed “cleantech” scorecards). ↩
4. Smil, V. (2015), Power Density: A Key to Understanding Energy Sources and Uses (MIT Press); Smil, V. (2017), Energy Transitions: Global and National Perspectives (Praeger / 2nd ed.); Smil, V. (2017), Energy and Civilization: A History (MIT Press); Smil, V. (2022), How the World Really Works (Viking) — four pillars (steel, cement, plastics, ammonia), transition speed, materials. Order-of-magnitude national physics: MacKay, D.J.C. (2009), Sustainable Energy — without the hot air (UIT Cambridge), free at withouthotair.com. Quantitative constraints for this essay: density, transition timescales, industrial pillars. ↩
5. Electricity share of global energy: Our World in Data energy dataset, owid/energy-data, field electricity_share_energy, country = World: 1990 13.6%; 2010 16.4%; 2020 18.9%; 2023 19.3%. Accounting conventions follow Our World in Data / Energy Institute pipeline; interpret as share of energy, not final useful services. ↩
6. Conversion efficiency: RMI and related analyses of fossil-system losses before useful work (order-of-magnitude ~⅔), as cited via Electrotech (n. 3); battery-electric cars and heat-pump factors of ~3× and ~2–4× for defined services in the same engineering literature. Not universal to high-temperature industrial processes. ↩
7. Multi-decade solar and battery cost declines “well over 90%” in standard industry series: RMI Cleantech Revolution / Electrotech citations (n. 3); Table I pins batteries to BloombergNEF’s volume-weighted pack survey (n. 14); crystalline-Si modules tracked separately in IRENA solar cost reports and spot indices. Module or pack price ≠ full system electricity cost, firm capacity cost, or multi-day reliability cost — dual account required. ↩
8. Handmer, C., “Future of Energy Reading List” (2023, updated) — caseyhandmer.wordpress.com — solar + batteries + synthetics as manufacturing hypothesis; deployment-rate claims as of 2023. Naam (solar cost path) and Vernon (demand electrification lag) linked there. Stress-tested against Smil (n. 4). ↩
9. International Energy Agency, Global EV Outlook 2025 — iea.org: global electric car sales >17 million in 2024, sales share >20%; China almost half of car sales electric in 2024. Chinese industry summaries: 2024 “new-energy vehicle” sales ~12.9 million, ~40.9% of total new vehicle sales (pure battery-electric plus plug-in hybrids in that definition). ↩
10. Dematerialization / intensity: Ausubel, J.H., long-run work on dematerialization and environmental intensity (e.g. Rockefeller University programs and essays on “the environment liberates”). Complements supply-side electrotech with demand intensity — not a claim that absolute use automatically falls. ↩
11. Materials and network bottlenecks as engineering objects (minerals, refining, grids): see also RMI work on battery mineral loops cited in Electrotech (n. 3). Stock-flow and recycling timescales are physical, not rhetorical. ↩
12. Path positioning in the text (nuclear + solar as multi-decade bridge; solar + batteries as asymptotic attractor conditional on storage and process electrification) is an engineering hypothesis constrained by Smil (n. 4) and learning data (nn. 7–8), not a present generation mix. Residual firm options (wind, hydro, gas with declining hours) remain in the measured system; they do not replace the stated sequence. ↩
13. Kardashev, N.S. (1964), “Transmission of Information by Extraterrestrial Civilizations,” Soviet Astronomy 8: 217–221 — Type I / II / III as orders of magnitude of controllable power (~planetary, stellar, galactic). Human global primary power ~10¹³ W order of magnitude (from ~600 EJ/yr primary energy scale); Earth’s intercepted solar power ~10¹⁷ W order of magnitude. Used here only as a power-budget scaffold for “headroom,” fused into the path section — not as SETI speculation or a technology chooser. See also standard order-of-magnitude treatments in MacKay (n. 4). ↩
14. Table I — single named series: BloombergNEF annual lithium-ion battery pack price survey (volume-weighted global average). Primary citation for the 2024 low: “Lithium-Ion Battery Pack Prices Hit Record Low of $115/kWh” (2024 pack-price release; about.bnef.com — URL path may move; search “battery pack prices”). Historical sequence in the table (2010–2023) follows the same annual survey: 1,160; 650; 373; 176; 137; 151; 139; then 115 in 2024 (dollars per kilowatt-hour). Methodology, chemistry mix, and real vs survey-year dollars can revise historical cells. Parallel silicon solar-module declines (International Renewable Energy Agency cost reports) are cited in prose only. Pack price is not full system electricity cost and not multi-day firming. ↩
15. Steel and ammonia process anchors: integrated blast-furnace final energy intensity often cited near ~18–25 gigajoules per tonne crude steel depending on plant (order of magnitude ~20 used in text). Hydrogen iron reduction: ~50–70 kg hydrogen per tonne reduced iron in engineering and policy literature. Ammonia: stoichiometric hydrogen requirement about 177 kg per tonne ammonia; real plants slightly higher. Electrolyzer electricity intensity ~50 kilowatt-hours per kilogram hydrogen class (plant-dependent). Smil (n. 4) on pillars; Agency and industrial roadmaps on blast furnace vs scrap electric arc vs hydrogen reduction. Not a plant design manual. ↩
16. Storage duration ladder (day-scale / multi-day / seasonal): standard power-system taxonomy. Table II is labeled schematic on purpose — not a bid stack. Day-scale utility lithium-ion levelized storage costs often land in the low-to-mid hundreds of dollars per megawatt-hour in Lazard storage-cost editions (year, financing, cycles, and duration shift the band). Multi-day and seasonal are not on the pack-price curve (Table I). See also Handmer manufacturing stack (n. 8). ↩
17. Stock vs flow (Fig. 05 Sankey and prose): Our World in Data energy dataset, owid/energy-data, country = World. Extracted fields: primary energy 2010 141,602; 2020 157,994; 2023 172,239 terawatt-hours; electricity generation 21,264; 26,723; 29,665; solar electricity 32; 853; 1,664; electricity share of energy 16.4; 18.9; 19.3%; fossil electricity 14,345; 16,596; 17,972. Primary energy uses Our World in Data’s conversion of the Energy Institute substitution method. Solar is generation, not capacity. Complements Fig. 01 share series (n. 5). ↩
18. California grid operator duck curve and battery firming: operator educational note on net-load ramps (“duck chart”); U.S. Energy Information Administration and operator tracking of battery storage growth (sub-gigawatt in the late 2010s toward multi-gigawatt by the early–mid 2020s, continuing into double-digit gigawatts); California Independent System Operator, 2024 Special Report on Battery Storage (May 2025) on battery energy and capacity contributions in late-afternoon / early-evening hours. Used as a day-scale vignette only — not a claim that multi-day or seasonal firming is solved, nor that California substitutes for global primary-energy stock (Fig. 05). ↩
19. Waste by energy path: fossil combustion residues (carbon dioxide, criteria air pollutants, ash, extraction wastes) are continuous and largely diffuse; nuclear spent fuel is low volume relative to energy delivered but requires long isolation and licensed institutions; solar and wind end-of-life hardware is mostly ordinary industrial mass (glass, metals, composites) needing collection and recycling at fleet scale; lithium-ion packs concentrate metals that enter recycling loops only after stock, collection, and plants exist. Positioning in the text: include waste in the dual account; require isolation and product stewardship rather than treating nuclear waste as unique disqualification or panel/pack waste as free. See also materials loops (n. 11) and Smil materials framing (n. 4). ↩
20. Nuclear cost and schedule band (order of magnitude, recent large projects): overnight capital cost often cited in the multi-thousand to ~10,000+ dollars-per-kilowatt range for recent OECD megaprojects, with first-concrete-to-commercial-operation intervals commonly ~8–15+ years when delayed; some East Asian multi-unit series have reported lower unit costs and shorter builds. Used here as a credibility band for “new nuclear in the bridge,” not as a single project forecast. See IEA / NEA and national project reporting; compare to modular solar factory lead times in months, not years. ↩