Civilization Still Boils Water

Here is the machine that runs your life. Something burns. The heat boils water. The steam pushes a turbine. The turbine spins a magnet inside a coil of copper. Electricity comes out the other side.

That is a coal plant. It is also a gas plant, a nuclear plant, a biomass plant, and most concentrated solar. The fuel changes. The magic trick does not. We found it in the 19th century and we never replaced it.

The consequence is a hard ceiling on how much energy civilization can make, and almost nobody talks about it, because the ceiling is not a policy or a permit. It is thermodynamics.

The number nobody checks

Global electricity demand hit a record 29,471 TWh in 2023, according to Ember's Global Electricity Review. Renewables crossed 30% of that mix for the first time, which was rightly celebrated.

Look at where the rest comes from.

Coal, gas, nuclear and bioenergy together are roughly 69% of world electricity. Every one of them is a heat engine. Most of them make that heat do work through steam.

And electricity is the small half of the story. Heat is the largest end use of energy on the planet. The IEA puts it at about half of global final energy consumption, against 30% for transport and 20% for electricity. Half of that heat goes into industry, most of the rest into buildings. Fossil fuels supply the bulk of it and produce over 40% of global CO2 emissions in the process.

Total primary energy production sits near 172 PWh a year. That is 19.6 TW of continuous power. The overwhelming majority of it passes through a boiler.

The ceiling is latent heat

No heat engine converts all of its heat into work. Carnot set the theoretical limit in the 1820s: efficiency depends only on the temperature difference between your hot source and your cold sink. Real plants do worse, because Carnot assumes zero entropy generation and reality does not offer that.

Real plants run the Rankine cycle. Water is pumped, boiled, superheated, expanded through a turbine, then condensed back to liquid so the pump can start again. The condenser is where the money burns.

Water needs an enormous amount of energy to change phase into steam. That energy - the latent heat of vaporization - has to be dumped again during condensation, at close to constant temperature, into a cooling tower or a river. It does no work on the way out. It is a structural tax on every steam plant ever built, paid every second of operation.

The cycle is squeezed from the other end too. Superheat is capped by what turbine metallurgy survives. Expansion is capped by moisture: let steam expand too far and water droplets form and erode the blades, so engineers add reheat stages and accept diminishing returns.

Add it up and the global average efficiency of thermal power generation is about 34%. Best available coal reaches 46%. The best combined cycle gas plants reach 61%, and they only get there by bolting a steam cycle onto a gas turbine to catch the exhaust.

Two thirds of the heat we make, we throw away. That is the ceiling.

Chemistry ran out of room

Before touching the machine, look at what we feed it. Energy density has two halves that matter: energy per kilogram, and energy per litre. Chemical fuels lose on one axis or the other, because chemistry only rearranges electrons.

Hydrogen is the best chemical fuel by mass, three times gasoline. By volume it is a disaster. Compressed to 700 bar it holds 5.3 MJ per litre. Chilled to minus 253 degrees C it holds 10. Gasoline holds 34 at room temperature in a thin steel can. Every hydrogen system pays that gap back in compressors, cryogenics, pipelines and losses, forever.

Now look one row up. Uranium in a light water reactor delivers around 3.5 million MJ/kg. In a fast breeder, burning the fertile U-238 that today's reactors leave behind, natural uranium reaches 86 million MJ/kg. That is roughly two million times gasoline.

This is not an argument that nuclear is nice. It is an argument that nuclear is a different physical category. Nothing else moves terawatts without moving mountains.

Replacing the boiler

The first fix is to make heat hotter and safer, which is the whole point of the Generation IV reactor designs.

A light water reactor holds water at 75 to 150 atmospheres so it stays liquid at 300 degrees C. That pressure is the source of most of what can go badly wrong, and the 300 degrees is the source of the mediocre efficiency.

A molten salt reactor dissolves the fuel into a liquid fluoride salt and runs at roughly atmospheric pressure, at around 700 degrees C. There is no pressurized water to flash to steam, so there is no explosion to contain. The fission products that poison solid fuel, xenon-135 above all, bubble out of a liquid continuously instead of forcing shutdowns. If the plant overheats or loses power, an actively cooled plug of frozen salt melts and gravity drains the fuel into tanks where the chain reaction cannot continue. No operator, no pump, no decision required.

I have written in detail about how China's TMSR-LF1 actually works and about the materials science that still limits it. Short version: the physics was proven at Oak Ridge in 1965, and the remaining work is metallurgy, salt chemistry and regulation.

Paired with thorium, the fuel cycle gets better again. Thorium-232 absorbs a neutron, becomes protactinium-233, decays into fissile uranium-233. U-233 gives back enough neutrons per fission that a thermal-spectrum reactor can breed its own fuel, with designs targeting breeding ratios slightly above 1.0 and low starting fissile inventories. It also produces far fewer long-lived transuranics than the uranium-plutonium cycle, which shrinks the waste problem that dominates every public argument about nuclear power.

This is the reactor line Vantar Energy is building for Europe: containerized molten salt units sized for industrial heat and grid baseload, using proven TMSR technology adapted to EURATOM rules rather than invented from scratch.

Replacing the turbine

Getting to 700 degrees C only pays off if the power block can use it. Steam cannot, not efficiently. So the replacement working fluid is carbon dioxide.

CO2 turns supercritical at 31.1 degrees C and 7.37 MPa, conditions that are almost mild. Above that point it has the density of a liquid and the flow behaviour of a gas, and it never changes phase again inside the cycle. Three things follow.

There is no latent heat to throw away. The fluid is cooled down without condensing, so the structural tax of the Rankine cycle disappears.

Compression gets cheap. Just above the critical point, CO2 is highly compressible, so the compressor consumes far less of the turbine's output than it would with an ideal gas. That recovered work is most of the efficiency gain.

The machine shrinks. Because supercritical CO2 is dense, a turbine of a given power output is roughly one tenth the size of its steam equivalent. A power block that fits in a room instead of a hall changes what can be factory built and shipped.

The catch is that supercritical CO2 cycles run low pressure ratios, so the turbine exhaust stays hot and the cycle needs heavy recuperation to work at all. The best configuration is the recompression closed Brayton cycle, which splits the flow before the cooler and routes part of it through a second compressor to fix the heat capacity mismatch inside the recuperator. At turbine inlet temperatures of 550 to 750 degrees C - exactly the range a molten salt reactor delivers - published cycle analyses put net thermal efficiency above 50%.

The cost sits in the heat exchangers, not the turbine. In one recompression cycle assessment the printed circuit heat exchangers accounted for over half of the power block capital cost. Techno-economic studies still project a 6 to 8% reduction in levelized cost of electricity against steam. And whatever heat the CO2 cooler rejects between 80 and 350 degrees C can be caught again by an organic Rankine bottoming cycle, using a working fluid that boils where water will not.

Skipping the machine

The deeper move is to delete moving parts. Turbines have bearings, fatigue limits, spin-up times and maintenance crews. Solid-state conversion has none of that.

Thermophotovoltaic cells are photovoltaics tuned to infrared. Point one at something glowing hot and it makes electricity directly. The historic problem was that a hot object radiates a broad spectrum, and every photon below the cell's bandgap passes through and turns into waste heat.

The fix was spectral recycling. Modern tandem cells with highly reflective back mirrors send 96 to 99% of sub-bandgap photons back to the emitter, where the energy stays as heat instead of leaking away. Lab devices now exceed 40% conversion at emitter temperatures near 2,000 degrees C, and stay above 30% near 1,000 degrees C. That is already competitive with a steam turbine, from a device with no moving parts at all.

The obvious application is thermal storage. Heat a large graphite block with surplus wind and solar, keep it insulated, and convert it back on demand with TPV panels that respond instantly. Cheap materials, no lithium, no spin-up, dispatchable output.

On the solar side the same logic applies to the input. Single junction silicon is approaching its Shockley-Queisser limit near 29%. Perovskite-silicon tandems stack a wide-bandgap top cell over silicon to catch a broader slice of the spectrum, and certified lab cells have now passed 33%. Stability was the historic objection and it is receding, with tandem devices demonstrating thousands of hours under standardized ISOS ageing tests.

The end state of this direction is aneutronic fusion. Deuterium-tritium fusion, the ITER path, releases around 80% of its energy as uncharged 14 MeV neutrons. Neutrons cannot be steered by magnetic fields, so their energy has to be caught in a lithium blanket as heat, and then boiled into steam. The most advanced machine humanity has ever attempted ends in a 19th-century power block. Proton-boron-11 and deuterium-helium-3 reactions release almost all of their energy as charged particles instead, which can do work directly against a decelerating magnetic field and induce current with no thermal cycle in between. Theoretical direct conversion efficiencies run from 70 to 90%. It is decades out. It is also the only path that leaves the Carnot ceiling behind entirely.

The bottleneck is not physics

Every technology above exists in a lab or a demonstration plant. What does not exist, in the West, is fuel.

Today's reactors run on low enriched uranium at 3 to 5% U-235. Most advanced designs - small modular reactors, fast reactors, many molten salt concepts - need high assay low enriched uranium, enriched above 5% and below 20%. HALEU buys smaller cores, longer cycles and higher burnup. It has no commercial Western supply chain.

The physics is not the problem. Around 85% of the separative work needed for HALEU is already done by the time you have made standard reactor fuel. The problem is that only Russia's Tenex and Chinese state entities produce and ship it at scale, that the United States banned Russian uranium imports in 2024, and that everyone else is stuck in a standoff. Reactor developers cannot finance construction without guaranteed fuel. Enrichers cannot justify billions in cascades without signed long-term reactor contracts.

Above 10% enrichment the material moves into IAEA Category II security, which means new certified transport casks designed against criticality and heat, and heavier physical security at every facility that touches it. Fuel fabrication has its own constraint that no subsidy fixes quickly: nuclear-qualified machinists, welders and inspectors are scarce, and they sit directly on the critical path.

The US Department of Energy has committed $2.7 billion over ten years to domestic enrichment. Europe has Urenco and Orano and a far less decisive policy. This is the part of the transition that gets decided by industrial policy rather than by engineers, and it is currently the binding constraint on everything else.

Land is the other ceiling

There is one more physical limit worth naming, because it decides what "scale" can mean.

Sunlight arrives at the surface averaging about 170 watts per square metre. After panel efficiency, spacing to avoid self-shading, and the fact that nights exist, utility-scale solar nets under 20 watts per square metre of site. Wind is more diffuse still, because turbines must sit five to seven rotor diameters apart or they eat each other's wind. The Alta Wind Energy Center averages around 4 watts per square metre. Walney offshore is near 6.

Thermal and nuclear plants are measured in thousands of watts per square metre. Counted across the entire chain, including mining, enrichment and waste storage, nuclear uses roughly 27 times less land than coal and 34 times less than solar PV per unit of electricity.

If global demand doubles or triples this century - and with electrification, AI data centres, desalination and carbon removal, it will - then the sources you scale determine how much of the planet becomes equipment. A civilization that wants both abundant energy and an intact biosphere has to generate at high power density. There is no version of the arithmetic where that is optional.

What this means

Four shifts, in the order they gate each other.

Move from chemical fuel to nuclear fuel, because 141 MJ/kg cannot carry a growing civilization and 86 million MJ/kg can.

Move from pressurized water to high-temperature molten salt, because 700 degrees C at one atmosphere is both safer and thermodynamically worth more than 300 degrees C at 150 atmospheres.

Move from steam to supercritical CO2, because the latent heat of water is a tax we have been paying since Watt.

Then move from turbines to solid state, first for thermal storage, eventually for direct conversion from fusion.

None of this is blocked by a scientific unknown. It is blocked by fuel supply, by fabrication capacity, by a licensing system built for one reactor design, and by a European habit of waiting for someone else to build it first.

We still boil water. We have known better for sixty years. Vantar Energy exists because the machine that replaces the boiler should be built here, and because every year we do not is a year of paying 19th-century prices for 21st-century demand.