Power Density: The Spatial Constraints Behind Every Decarbonisation Bet
Vaclav Smil argues that land will determine how fast civilisation can decarbonise. A decade later, his framework explains why hyperscalers are buying nuclear plants.
This is Part 2 of a five-part Libido Sciendi series on Energy, Technology, and Investment. Part 1 examined Energy and Civilization and the 50-year rule governing transitions. This instalment covers Power Density and the spatial constraints on deployment. Part 3 covers Ludovic Subran’s calculation of the 84% investment gap. Part 4 traces Daniel Yergin’s geopolitical maps. Part 5 asks whether AI can accelerate the transition it complicates.
TL;DR
Power Density: A Key to Understanding Energy Sources and Uses by Vaclav Smil
Smil’s companion volume to Energy and Civilization makes three arguments that should reshape how anyone models the energy transition or underwrites infrastructure investment:
1. Power density does not improve as costs fall. Fossil fuels deliver 1,000 to 10,000 watts per square metre. Solar panels manage 5 to 10. Wind farms, accounting for turbine spacing, rarely exceed 1. These gaps span three to four orders of magnitude and do not compress as costs fall. The gap in W/m² is structural, not incidental.
2. For dense economies, the spatial arithmetic leaves little slack. Replacing American fossil fuel consumption with renewables would claim 25% to 50% of national territory, versus under 0.5% today. Japan would need 60% more land than it possesses. The UK would require essentially its entire surface. The constraint is not political will or capital. It is geometry.
3. Nuclear is the only low-carbon source operating at fossil density. At 2,000 to 6,000 W/m², it is the sole proven technology capable of powering cities, data centres, and heavy industry without restructuring the landscape. This ratio is why hyperscalers are restarting cold reactors.
| Energy Source | Power Density (W/m²) | Context |
|---|---|---|
| Middle East oil field | 10,000–40,000 | Peak production years |
| Natural gas field | 2,000–50,000 | Varies by geology |
| Large coal mine | 1,000–15,000 | Surface mining, thick seams |
| Nuclear plant (full site) | 2,000–6,000 | Including exclusion zones |
| Solar thermal (rooftop) | 40–100 | Best renewable performer |
| Solar PV (ground array) | 3–11 | Actual generation, not peak |
| Hydropower | 0.5–20 | Alpine stations at high end |
| Wind farm (with spacing) | 0.5–1 | After capacity factor |
| Sugar cane ethanol | 0.4 | Brazilian production |
| Corn ethanol | 0.25 | US average yields |
| Rapeseed biodiesel | 0.12–0.18 | European average |
A cliff separates fossil and nuclear sources from renewables. Solar occupies a middle ground. Biofuels sit at the bottom, constrained by photosynthesis itself.
Why This Number Matters
Smil opens his book with 1548. A King’s Commission in Sussex examined damage from iron mills to English forests. Charcoal for smelting was stripping the Weald bare. By 1700, producing 12,000 tonnes of bar iron annually required harvesting 100,000 hectares of coppiced forest. England had hit a ceiling.
Coke shattered it. A single coal mine delivered fuel at 800 W/m², four orders of magnitude above charcoal’s 0.07. The industrial revolution followed: not because coal contained more energy per kilogram (charcoal and coke are nearly identical at roughly 30 MJ/kg), but because coal concentrated extractable energy in space.
This distinction between energy density (joules per kilogram) and power density (watts per square metre) structures the book. Crude oil has the highest energy density among fossil fuels at 42 megajoules per kilogram. But if oil existed only in tiny pockets scattered across vast areas, its power density would be too low to exploit. Middle Eastern fields supply the world because they concentrate high-energy-density fuel at unmatched spatial densities.
The practical consequence: every civilisation’s energy system is ultimately a spatial problem. How much usable power can you extract from a given piece of territory? Smil spent a career watching energy analysts ignore this question. The rest of his book explains why they should not.
The Renewable Hierarchy
Smil works through every technology with the patience of an engineer and the scepticism of a historian. The hierarchy that emerges is separated by orders of magnitude rather than incremental differences.
Solar thermal performs best among renewables. These are the simplest solar systems: dark panels that heat water directly, no electricity involved. Rooftop water heaters exceed 700 W/m² during peak hours in sunny climates, averaging 40 to 100 annually. Smil considers them chronically underappreciated in mainstream energy planning.
Solar photovoltaics sit one tier below. Published figures routinely mislead for two reasons. First, manufacturers quote peak capacity under ideal laboratory conditions, but real-world capacity factors run 10% to 25% depending on latitude and weather; peak solar irradiance hits 1,000 W/m² at noon; panels convert a fraction (150 to 200 W/m²), and annualised output falls to single digits. Second, panels cannot tile a site edge-to-edge: access roads, inverter stations, and maintenance clearances dilute coverage. Ground arrays in high-sun regions like Spain or California deliver 7 to 11 W/m² of total project area. Rooftop installations avoid land competition but not spatial dilution; dense German rooftops, despite weaker sunlight, achieve similar figures per square metre of building footprint because panels pack more tightly without ground-level infrastructure.
Wind drops further. A single turbine blade sweeps impressive power from the air: a 3-megawatt Vestas V90 captures 470 W/m² of rotor disc. But turbines create wake turbulence, and downwind machines would spin in degraded air. To avoid this, wind farms space turbines far apart, roughly five rotor diameters side-to-side and ten diameters front-to-back. For a turbine with 90-metre blades, that means nearly a kilometre between rows. The result: large American wind farms operate at just 2 to 4 W/m² of installed capacity across the whole site. After capacity factors of 25% to 33%, actual generation falls to 0.5 to 1.5 W/m².
The wind industry argues that only turbine pads and access roads should count toward land use, since farmers can graze cattle between towers. Smil is unmoved. Spacing is the operative constraint: power density measures the energy flux derivable from a given area, regardless of how close or far apart the converting facilities may be. The field is spoken for. You cannot simply add more turbines.
Hydropower spans the widest range of any source. Ghana’s Akosombo Dam floods 8,500 km² to create Lake Volta, generating 0.06 W/m². Switzerland’s Grande Dixence, with a small reservoir and 2,000-metre vertical drop, reaches 500 W/m². Most large projects fall between 1 and 20 W/m². Hydro claims over half of all land occupied by global energy infrastructure while supplying under 3% of primary energy.
Biofuels anchor the bottom. US corn ethanol delivers 0.25 W/m². Brazilian sugar cane reaches 0.4. European rapeseed biodiesel manages 0.12 to 0.18. Cellulosic ethanol, perpetually five years from commercialisation, might reach 0.4 under optimistic assumptions. These numbers cannot improve dramatically: photosynthesis converts under 1% of incident solar radiation into harvestable biomass. The thermodynamic ceiling is fixed by biology, not engineering.
The Nuclear Exception
Nuclear receives surprisingly brief treatment given its importance to Smil’s argument. Reactor cores achieve 10⁶ W/m², but downstream energy flows, through steam and turbines, match fossil plants, so station layouts are similar. Complete facilities operate at 2,000 to 6,000 W/m², including structures, cooling, and exclusion zones.
The full fuel cycle, covering mining, milling, enrichment, and waste handling, brings densities to 200 to 1,000 W/m², overlapping coal’s 100 to 1,000+ when mining and transport are included. High-productivity underground uranium mines exceed 10,000 W/m²; in-situ leaching falls to 400 to 600.
Nuclear is the only proven low-carbon technology operating at fossil fuel densities. Everything else requires one to four orders of magnitude more land. This is not a regulatory or political judgement. It is a consequence of physics.
EROI: The Surplus That Funds Civilisation
Buried in the final chapter, Smil introduces Energy Return on Investment (EROI): energy delivered divided by energy consumed to build and operate the system. An EROI of 10 means ten units out for one unit in.
Why does EROI matter? Below a threshold of roughly 7, too much of the energy produced must be reinvested in energy production itself. The net surplus available to power everything else — food, housing, healthcare, education — shrinks below what complex societies require. Pre-industrial societies operated at EROI of 5 to 10 and remained trapped in subsistence. Fossil fuels broke the ceiling: early Pennsylvania crude was extracted at EROI ratios above 100.
Research by Weissbach and colleagues (2013) calculated EROI across technologies on a consistent basis. Their key contribution was distinguishing between unbuffered systems (electricity fed straight to the grid, intermittency someone else’s problem) and buffered systems (paired with storage to deliver power on demand). The distinction matters because solar and wind produce electricity when conditions allow, not when consumers need it. Making them comparable to dispatchable sources requires adding batteries or other storage, which themselves consume energy to manufacture.
Nuclear, coal, and gas need no buffering: they ramp on demand. Wind unbuffered exceeds the EROI threshold comfortably; add storage, and it falls below. Solar PV, even without storage, sits at the margin with an unbuffered EROI of 3.9 in German conditions, and 1.6 buffered. This does not mean solar and wind cannot contribute substantially. It means their contribution requires careful integration with dispatchable sources, whether hydro, nuclear, or gas, or with storage systems that do not consume the gains they generate.
| Technology | Unbuffered EROI | Buffered EROI |
|---|---|---|
| Nuclear (PWR) | 75 | – |
| Coal plant | 30 | – |
| Gas turbine (CCGT) | 28 | – |
| Wind (Europe) | 16 | 3.9 |
| Solar PV (Germany) | 3.9 | 1.6 |
Source: Weissbach et al. (2013). Energy intensities, EROIs, and energy payback times. Energy, 52, pp.210–221.
Consumption Densities
Understanding supply requires understanding demand. How much power does a society draw from each square metre of its territory?
The global average is 0.125 W/m² of ice-free land. But averages obscure the range. A densely populated, highly industrialised country like Germany or the Netherlands consumes around 3 W/m². France, with more land and a nuclear-dominated electricity mix, sits closer to 1.5 W/m². Desert nations with sparse populations approach zero.
Cities concentrate demand sharply. The global urban average runs roughly 20 W/m². Central London or Paris exceeds 100 W/m². The densest blocks of Manhattan or Hong Kong approach 1,000 W/m² — matching peak solar irradiance at noon, yet solar panels deliver only single-digit annual averages.
Here lies a paradox that urban planners tend to celebrate without following to its conclusion. Dense cities are efficient per capita: shared walls, public transport, shorter distances. A Manhattanite consumes less energy per person than a Phoenix suburbanite. For a given population, concentrating people in cities reduces total energy demand, a genuine win for decarbonisation. But it does not ease the spatial mismatch. Stack enough efficient people vertically, and power demand per unit of land soars beyond what any local renewable source can match. Cities must import power from distant generation, regardless of how efficiently their residents behave.
The current system handles this mismatch through concentration. Fossil fuels are extracted at 1,000 to 10,000 W/m², then distributed via pipelines and transmission lines to dispersed consumers. The entire infrastructure claims under 0.5% of American territory.
A renewable system reverses the flow: gathering diffuse energy from vast areas, concentrating it for delivery. The question is whether there is enough land.
Smil calculates full renewable replacement for US 2012 consumption across two scenarios. In a biofuel-heavy scenario, where solar and wind replace electricity while biofuels replace liquid and gaseous fuels, the EU would require approximately 200 million hectares, roughly 50% of its total land area. Picture the entirety of France, Germany, Spain, Italy, and Poland covered with energy infrastructure. In an electrification-heavy scenario, where half of fuel uses electrify and the remainder comes from biofuels, the land claim drops to around 100 million hectares, approximately 25% of EU territory, still equivalent to France and Germany combined, and before conversion losses are accounted for. The current system claims under 1% of EU land.
The challenge is not that renewables cannot scale. They can. The challenge is that scaling them means restructuring the landscape at a pace and extent without historical precedent.
Countries That Cannot
Large nations might accommodate these land claims. Smaller ones cannot.
The UK needs roughly 240,000 square kilometres at an average generation density of 1 W/m². UK total area: 242,500 square kilometres. David MacKay, in Sustainable Energy Without the Hot Air, reached the same conclusion independently: in a renewable-powered world, the land area required to maintain British energy consumption would have to approximate the area of Britain itself. Germany needs around 350,000 square kilometres even with every rooftop covered. Germany’s total area: 357,000 square kilometres. Japan needs roughly 600,000 square kilometres across its four main islands, which together cover 378,000.
Taiwan, Singapore, South Korea, and the Netherlands face equivalent arithmetic. International trade offers partial relief, as two-thirds of crude oil already crosses borders. But trading biomass at 0.4 W/m² from Brazil is a different proposition from trading crude at 10,000+ W/m² from Saudi Arabia. The logistics differ by orders of magnitude, and those orders of magnitude translate directly into shipping capacity, port infrastructure, and cost.
Political ambition does not dissolve these constraints. They require either technological breakthroughs in generation density, or an honest reckoning with the fact that energy trade patterns will restructure substantially over the coming decades.
100% Renewable by 2030
Smil reserves particular attention for a 2011 paper by Jacobson and Delucchi claiming 100% renewables by 2030. Their plan: 3.8 million 5-MW wind turbines, 49,000 central solar plants, 1.7 billion rooftop installations, and assorted geothermal, wave, and tidal capacity. Barriers, they argued, were “primarily social and political, not technological or economic.”
Smil’s response is a list of multipliers. Global wind installed in 2013: 330 GW. Plan requirement: 13 TW. Factor: 40x. Global PV: 100 GW. Requirement: 17.1 TW. Factor: 170x. Solar PV plants of 300 MW or more in existence: zero. Required: 40,000. Wave devices deployed globally: six. Required: 720,000.
Scaling anything by one to five orders of magnitude in under two decades has no precedent. The coal-to-oil transition took half a century. Electrification took longer. Asserting comparable change at unprecedented speed, with barriers merely political, is aspiration dressed as engineering. This matters for investors: mandates anchored to 2030 timelines carry real financial risk when the underlying physics have not moved.
What Transition Actually Requires
Smil is not arguing against renewables. He is arguing against fantasy timelines. His framework points toward what would need to be simultaneously true for a genuine transition.
- Efficiency gains narrow the gap between supply and consumption densities. A building needing 15 W/m² instead of 30 is twice as tractable to power from diffuse sources, and efficiency improvements compound across decades in ways that cost projections for new generation often understate.
- Storage at genuine scale, from household batteries to utility facilities, buffers intermittency. Pumped hydro already stores gigawatts seasonally; grid batteries remain at megawatts. The gap between those two figures is large and will not close quickly.
- Power-to-fuels pathways convert surplus electricity to storable hydrocarbons for aviation, shipping, and high-temperature industry. Hydrogen and synthetic fuels show promise, but neither is commercially mature at the required scale.
- Transmission expansion connects distant generation to consumption centres: Arizona solar serving Boston, North Sea wind reaching Bavaria, undersea cables linking continents. These projects take decades to permit and build.
- High-density low-carbon sources, meaning nuclear and large hydro, must continue operating throughout the transition. Eliminating them while eliminating fossil fuels multiplies renewable land requirements with no compensating benefit.
And the timelines are multi-decade: not 2030, probably not 2050 in any comprehensive sense. Smil’s work on transitions, detailed in Part 1 of this series, suggests 50 to 75 years for fundamental shifts in primary energy.
The Decentralisation Illusion
Distributed generation appeals to those who distrust centralised infrastructure. Rooftop solar, community wind, microgrids. Tokyo is Smil’s worked example.
Tokyo’s 23 central wards consume roughly 100 W/m². Available rooftop area: approximately 64 square kilometres. With panels covering every accessible surface at 12% efficiency, output equals roughly 10% of metropolitan demand, and under 1% in the dense core. The Global Energy Assessment concluded that local renewables “can only supply urban energy in niche markets, but can provide less than 1% of a megacity’s energy needs.” When consumption exceeds 100 W/m² and supply reaches perhaps 10, decentralisation cannot close the gap. Megacities need concentrated generation via high-voltage transmission, regardless of the primary source.
This does not mean distributed generation is without value. For households, small businesses, and communities in lower-density settings, it contributes meaningfully. The illusion is assuming that what works at the periphery scales to the core.
Landscape Transformation
Modern energy infrastructure is nearly invisible. Compact extraction sites, underground pipelines, power plants on city outskirts. The entire system claims a trivial share of land.
A renewable system spreads across the landscape. Wind farms on ridgelines, solar arrays covering deserts and agricultural land, biofuel plantations competing with food production, transmission lines crisscrossing continents. Energy becomes the dominant feature of the built environment, competing with agriculture, housing, and wilderness simultaneously.
“Energy studies have accomplished a remarkable feat by largely ignoring space as a key organising determinant of modern systems supplying fuels and electricity,” Smil writes. “New energy arrangements are both inevitable and desirable, but without any doubt, if they are to be based on large-scale conversions of renewable energy sources, then the societies dominated by megacities and concentrated industrial production will require a profound spatial restructuring of the existing energy system.”
The early moves are already visible in every infrastructure planning document that mentions small modular reactors (SMRs), high-voltage direct current (HVDC) lines, and offshore wind corridors in the same breath.
Ten Years On
Published in 2015, Power Density predates the solar cost collapse, the offshore wind scale-up, and the nuclear retreat. The intervening decade is worth examining.
Solar PV costs collapsed roughly 90%, faster than Smil predicted. Global capacity grew from 180 GW to over 1,500 GW. But power density improved only marginally. Module efficiency rose from 15–17% toward 20–23%, pushing project densities from 5–8 toward 8–12 W/m². Same order of magnitude. The economic story changed substantially; the spatial story did not.
Wind turbines grew larger, with offshore units exceeding 15 MW, improving capacity factors to 40–50% offshore versus 25–35% onshore. But spacing scales with rotor diameter. Larger turbines require more distance between them. The fundamental density constraint persists.
Battery storage scaled from under 1 GWh to over 100 GWh globally, with costs falling roughly 80%. This addresses timing, not land. Storage buffers intermittency; it does not increase W/m².
Nuclear moved in the opposite direction across the West. France’s Flamanville 3 EPR, originally budgeted at €3.3 billion for completion in 2012, eventually connected to the grid in December 2024 at a cost exceeding €13 billion. Germany completed its nuclear phase-out in 2023. Global generation flatlined while demand grew. The high-density low-carbon option contracted where it was most needed.
Smil’s ratios hold. Solar and wind are dramatically cheaper than in 2015. They are not more spatially efficient. The geometric constraints are independent of cost curves.
The AI Power Problem
One development Smil did not anticipate: electricity demand from artificial intelligence.
Data centres consumed roughly 1% of global electricity in 2015. By 2024: 2 to 3%, rising fast. AI training compute grows approximately 10x annually. A frontier model training run consumes energy matching thousands of households across a year. The IEA projects global data centre electricity consumption reaching 945 TWh by 2030, more than doubling the 2024 figure.
Data centres operate at 200 to 500 W/m² of building footprint. AI-optimised facilities push toward 1,000 W/m² as GPU density increases. Hyperscale campuses concentrate loads rivalling small cities. Microsoft, Google, Amazon, and Meta each plan gigawatt-scale facilities.
These loads sit outside what distributed renewables can serve. A 500-MW data centre campus cannot rely on rooftop solar or nearby wind. A data centre needs firm power, 24 hours a day, at densities that only fossil fuels, nuclear, or large hydro can provide locally. Alternatives exist: massive renewable overbuild with storage, which multiplies land requirements, or long-distance transmission, which adds losses and cost. Neither is straightforward at the required pace.
The capital markets have registered the tension. Microsoft signed to restart Three Mile Island’s undamaged reactor. Amazon purchased a nuclear-powered campus in Pennsylvania. Google announced SMR procurement agreements with Kairos Power. After decades of retreat across the West, nuclear returns, pulled by the physics of power density.
Smil’s framework explains why. When consumption density rises from 20 W/m² (urban average) toward 500+ W/m² (AI data centre), the mismatch with renewable supply widens. A solar farm at 10 W/m² needs 50 times the land to match a data centre’s footprint. At some threshold, geometry forces a reckoning.
Intelligence may be substrate-independent. Power is not.
Orthogonal Insight: The Concentration Reversal
Viewed across long timescales, the history of energy is a story of increasing spatial concentration. Hunter-gatherer societies drew energy from territories many times the size of their settlements. Agriculture condensed it into cultivated fields. Coal mines extracted it from seams a few hundred metres wide. Saudi fields concentrate energy at densities the rest of the world has never replicated. The modern transmission grid is infrastructure built around a single premise: energy is dense at source and diffuse at destination.
The energy transition inverts this logic. Instead of extracting concentrated energy from small areas and moving it outward, a renewable system must gather diffuse energy from vast areas and move it inward. This reversal has a direct infrastructure consequence: the relevant metric is no longer cost per kilowatt-hour, but square kilometres per gigawatt installed, and kilometres of high-voltage line per terawatt delivered. Levelised cost of energy calculations were not designed to capture either of these quantities.
The implication for capital allocation is underappreciated. Infrastructure investors who model renewable projects primarily through LCOE and capacity factor are working with a framework built for a concentration-first world. The transition they are financing runs in the opposite direction.
Conceptual Toolbox
| Concept | Definition | Excerpt from Smil | Implication |
|---|---|---|---|
| Power density | Rate of energy production or consumption per unit area, measured in W/m² | ”Cities consumed 20–30 W/m² while forests produced only 0.6 W/m²” | Land requirements for data centres, solar farms, and hydrogen production must be evaluated in W/m² terms, not just cost |
| Energy density | Energy content per unit mass (MJ/kg), distinct from power density | ”Crude oil has the highest energy density among fossil fuels at 42 MJ/kg” | A fuel can be energy-dense but spatially dilute; both dimensions constrain deployment |
| EROI | Energy delivered divided by energy consumed to build and operate the system | Below EROI of ~7, “too much economic output must be reinvested in energy production” | Buffered solar and wind approach the minimum viable EROI threshold; integration with dispatchable sources is structurally necessary |
| Buffering penalty | EROI reduction when intermittent sources are paired with storage to become dispatchable | Wind falls from 16 (unbuffered) to 3.9 (buffered); solar from 3.9 to 1.6 | Storage is not free; it consumes part of the energy gain it is meant to preserve |
| Consumption density | Power demand per unit area of built environment (W/m²) | Global urban average ~20 W/m²; dense city centres exceed 100 W/m² | Megacities structurally require imported concentrated power regardless of local renewable potential |
| The concentration reversal | Fossil systems extract concentrated energy from small areas and distribute outward; renewable systems gather diffuse energy from vast areas and concentrate inward | ”Modern civilisation has evolved as a direct expression of the high power densities of fossil fuel extraction” | Transition requires restructuring landscape use at unprecedented scale; LCOE alone does not capture this |
| Photosynthetic ceiling | Biofuel productivity capped by photosynthesis converting under 1% of incident radiation into harvestable biomass | Corn ethanol 0.25 W/m²; sugar cane 0.4 W/m²; “the thermodynamic ceiling is low and fixed by biology” | No engineering improvement can make biofuels spatially competitive with fossil or nuclear sources |
Closing Note
The power density framework does not argue that the energy transition cannot happen. It specifies what it would physically require: more land, more transmission, more storage, and more nuclear than current plans tend to acknowledge. Politics slows the transition. Physics constrains it.
Near the end of Power Density, Smil observes: “Modern civilisation has evolved as a direct expression of the high power densities of fossil fuel extraction… New energy arrangements are both inevitable and desirable, but… will require a profound spatial restructuring of the existing energy system.”
That restructuring is now underway. The question is not whether it will happen, but at what pace, through which technologies, and at whose expense when the timelines prove longer than projected.
Energy sets the boundaries. What we build within them remains our own affair.
Next in the series: Part 3 examines Ludovic Subran and Marco Zimmer’s Investing in a Changing Climate, calculating exactly how large the gap is between current climate investment and what a genuine transition requires, and what it implies for capital allocation.
This is the second article in a five-part Libido Sciendi Deep Digest series on the essential books for understanding energy, technology, and investment:
1. Vaclav Smil, Energy and Civilization: A History - How energy transitions shaped human societies across 10,000 years
2. Vaclav Smil, Power Density (this article) - Why land, not cost, constrains the energy transition
3. Ludovic Subran and Marco Zimmer, Investing in a Changing Climate - The 84% gap between current climate investment and what transition actually requires
4. Daniel Yergin, The New Map - Energy geopolitics and the clash of nations
5. The Synthesis - Jevons Meets Jensen: can AI realistically be powered at projected scale?
