The kWh Token

The Cost of Ascent

Energy, Cognition, and the Economics of the Substrate War

Tinkering with Time, Tech, and Culture #50

In 1984 I started a small company called NRG.

The name was simple. NRG, energy. Our tagline was even simpler:

The capacity for doing work.

At the time it was just a definition from physics that I liked. Energy is the ability to make things happen, to move matter, to push electrons, to transform the world. It felt fundamental.

I didn't realize then how much that definition would follow me.

Four decades later, after building networking stacks, wireless infrastructure, identity-based remote access systems, and now watching the rise of AI compute, that same definition keeps resurfacing in a new context.

Because if you strip away the abstractions of modern finance, digital economies, and speculative markets, civilization still runs on the same primitive:

Energy.

Which is why I've had a simple intuition for years:

The most honest monetary token is the kilowatt-hour.

Not dollars. Not gold. Not Bitcoin. Energy.

The Universal Token of Potential Work

A kilowatt-hour is not just power. It is the universal token of potential work.

The potential to mine. The potential to farm. The potential to transport. The potential to manufacture. The potential to compute. And now, increasingly, the potential to think.

Energy is the one resource that underlies every transformation civilization performs. Whether we are smelting steel, growing crops, transmitting signals across the planet, or running large-scale AI inference, the underlying act is the same: energy being converted into work.

For most of human history, the forms of that work were physical. Energy powered muscles, engines, and turbines. It pushed ships across oceans and trains across continents. Energy became motion.

But something new is happening now.

For the first time in history, we have built machines that convert energy directly into cognition-like behavior. Electricity flows into GPUs. Matrix multiplications occur at massive scale. Out comes inference. Language. Prediction. Planning. Decision-making.

In other words, the pipeline of civilization has expanded. Energy no longer produces only motion. Now it produces cognition.

The Calorie Ladder

This is not the first time civilization has bumped into an energy ceiling and climbed past it.

Two centuries ago, Thomas Malthus looked at the relationship between population and food and saw a wall. Calories grew arithmetically. Mouths grew geometrically. The math was unforgiving. Civilization, in his view, was permanently bounded by the rate at which sunlight could be converted into grain and grain into muscle.

He was right about the constraint and wrong about the ceiling.

What Malthus missed was that civilization keeps escaping its energy walls by climbing to a new substrate. Each escape looks the same in structure: an old energy regime hits a hard limit, a new one becomes available, and the ceiling moves up.

Muscle gave way to steam, steam to electricity, electricity to computation. Each transition rebuilt the Malthusian wall at a higher altitude and moved civilization onto a substrate the previous regime could not have powered. Now compute is giving way to cognition. The same electricity that ran the spreadsheets is running the inference engines. The substrate transition is in progress as you read this.

Each rung of this ladder consumed dramatically more energy than the one before. A human body runs on roughly 100 watts. A steam locomotive ran on megawatts. A modern datacenter runs on hundreds of megawatts and the largest training clusters are pushing into gigawatt territory. Every climb up the ladder is paid for in energy, and the cost is not metaphorical. It is measured in joules, dissipated as heat, drawn from grids that have to be physically built.

Which means the constraint Malthus identified never actually went away. It got renegotiated, repeatedly, at higher and higher levels of abstraction. Each new substrate raised the ceiling but also raised the floor. The civilization that runs on cognition cannot run on muscle, cannot fall back to steam, cannot survive a collapse of the grid that powers its compute.

We are now at the fourth wall. AI's energy hunger is not a novel problem. It is the recurring problem in its newest costume.

Energy → Cognition → Agency → Value

To understand why this matters economically, it helps to look at the deeper stack of how civilization creates value.

At the bottom is energy. Energy powers machines, infrastructure, and networks. Without energy, nothing moves and nothing computes.

From energy comes computation. Computation turns raw signals into structured information, routing packets, optimizing systems, simulating environments.

From computation emerges cognition. Cognition is the ability to model the world, to recognize patterns, make predictions, and generate plans.

And once cognition exists, something even more powerful appears: agency.

Agency is the ability to act in the world. An entity with agency can observe, reason, and take action that changes outcomes.

For most of human history, humans were the primary carriers of agency. Machines amplified our labor but did not generate decisions. AI changes that. Now electricity can power systems that write code, analyze markets, discover drugs, design components, and coordinate complex logistics.

Electricity is no longer just turning motors. It is producing decisions.

Which means the economic pipeline increasingly looks like this:

Energy → Computation → Cognition → Agency → Value.

That shift has enormous implications for how we think about economics, infrastructure, and power.

The Return of the Energy Constraint

For roughly seventy years, civilization got to pretend energy was a solved problem.

The early twentieth century delivered an extraordinary infrastructural transformation. Electricity spread across the planet. Automobiles reshaped transportation. Aircraft collapsed global distances. Nuclear energy promised near-limitless power. Early computers began the digital revolution.

Then progress narrowed. We kept shrinking transistors and writing better software. The information world accelerated. But the deeper physical infrastructure of civilization, energy especially, slowed dramatically. Nuclear stalled. Fusion remained perpetually twenty years away. Energy abundance plateaued.

Through that long stall the economy could keep expanding because computation became more efficient and global supply chains became more optimized. The information layer carried the weight while the physical layer coasted.

That arrangement is over.

AI systems are extraordinarily energy-hungry. Training large models requires massive GPU clusters operating for weeks or months. Running global-scale inference systems requires datacenters consuming gigawatts of power. The technology industry is rediscovering something earlier industrial eras understood very well: infrastructure matters.

Compute is constrained by electricity. Datacenters are constrained by power grids. AI capability is constrained by the ability to generate and deliver energy.

This is why we are seeing a quiet but dramatic shift in industrial priorities. Technology companies are investing in nuclear power. Solar farms and battery systems are expanding rapidly. Datacenters are being colocated directly next to power plants.

The underlying reality is simple. Energy is returning as the fundamental constraint of civilization, and the abstraction layers that floated free of it for seventy years are about to reacquaint themselves with their substrate.

The Substrate War

This is where the idea of the Substrate War enters the picture.

For much of the twentieth century, geopolitical power revolved around control of physical resources and industrial capacity. Coal. Oil. Steel. Factories.

In the twenty-first century, power is increasingly determined by control of the infrastructure that produces cognition. Semiconductor manufacturing. Datacenters. Global network infrastructure. And the electricity that powers all of it.

AI models themselves are only the visible layer. Underneath them lies a deeper physical stack:

Energy → Compute → Inference → Cognition.

Control the bottom layers of that stack and you control the production of intelligence. That is the real battlefield of the coming decades.

But here is where most analysis of the Substrate War stops, and where it needs to go further.

The Two Substrates

The Substrate War has two fronts, not one.

The physical substrate is what we have been discussing: energy, compute, network infrastructure, the literal hardware that converts electricity into inference. Whoever owns the physical substrate owns the means of producing thought.

But there is a second substrate underneath everything AI does, and it is at least as contested as the first.

The ideological substrate is the layer of meaning, attribution, and canonical reference that determines what the physical substrate produces thoughts about. The training data. The citation graph. The provenance chain. The set of facts and frames a model treats as real. The narrative ground truth that gets baked into weights.

That is why provenance, attribution, and machine-readable authorship are not peripheral concerns. They are the other half of the war. AnchorID is one piece of that infrastructure, built specifically to give humans a way to assert authorship in a form that survives ingestion into training pipelines.

You can own all the GPUs in the world and still lose if the data those GPUs train on has been shaped by someone else.

History rhymes here. The twentieth century offered a clear case of physical control without ideological alignment turning out to be hollow.

The Cold War was nominally a contest between two industrial superpowers, but the actual battlefield was alignment. The Soviet Union had nuclear weapons, oil, factories, and a continental landmass. What it didn't have, by the end, was a story its own population still believed in. The physical substrate held; the ideological substrate collapsed; the whole structure came down.

The pattern is consistent. Physical substrate without ideological alignment is brittle. Ideological substrate with even modest physical backing is durable. The interesting question for the AI era is what happens when both fronts are contested simultaneously, at machine speed, by systems trained on data whose provenance nobody fully controls.

This is what I've been calling the Substrate War in the broader sense. Not just the fight over GPUs and grid capacity, but the fight over what the GPUs end up thinking about. Both fronts use energy. Both fronts produce cognition. But they operate on different layers of the stack and require different defenses.

The kWh token addresses the physical front. It makes the cost of computation legible, prevents the markup-on-permission gap that current AI economics depends on, and ties the money supply to actual generative capacity. It is a tool for the physical substrate war.

The ideological front needs its own tools, and that work is the running thread through most of what I've published this year. AnchorID is one piece. Other people are building other pieces. The work is not done and the front is not stable.

What matters for this essay is the recognition that both fronts exist and that they reinforce each other. A civilization that owns its physical substrate but loses control of its ideological substrate ends up running its own GPUs to produce someone else's worldview. A civilization that secures its ideological substrate but cannot generate its own electricity ends up renting cognition from whoever owns the grid. Both fronts have to be held.

The kWh token is one piece of one front. It is a foundational piece, because energy underlies everything else. But it is not the whole war.

The Cost of Ascent

I've written elsewhere about what I call the Atlas of Cognition, a vertical map tracing the descent of matter into form and the ascent of form into mind. The full ladder runs from raw physics at the bottom through thermodynamics, physical computation, and statistical cognition, then upward through neural computation, conscious integration, language, machine cognition, and finally into what I call the reflective continuum, the universe studying itself through every system capable of modeling its environment.

That map matters here because it reveals something the economics conversation usually misses.

Every layer of cognition is funded by the layer below it. And every layer below it is funded by energy.

When a GPU trains a neural network, the temperature monitor climbs as the loss curve drops. The chip gets hotter as the model gets smarter. You can watch it happen in the thermal readout: energy becoming inference. I saw the same thing decades ago building TCP/IP stacks for AVR microcontrollers, voltage regulators warming up as packets moved through state machines. Electrons flowing through silicon, becoming logic, becoming protocol, becoming meaningful communication.

Thinking has a temperature. Understanding has a power budget. Every act of cognition, whether in silicon or neurons, follows the same arc: energy constrained into pattern, pattern refined into prediction, prediction becoming something we recognize as understanding.

This is what makes the kWh the honest token.

Not because it measures electricity. Because it measures the cost of ascent.

Every step up that ladder, from raw physics to computation to inference to meaning, burns energy. The kilowatt-hour is the unit that makes that cost legible. It is the price of climbing from matter toward mind.

Fiat currencies and speculative tokens obscure this. They float above the physical substrate, disconnected from the thermodynamic cost of the work they represent. They function only as long as the energy substrate beneath them continues to hold. When it doesn't, when the grid fails or the supply chain breaks or the datacenters go dark, the abstraction collapses back to what was always underneath it. Energy.

The kWh doesn't borrow its certainty from anywhere. It is the certainty. One kilowatt-hour delivered is one kilowatt-hour of potential work, potential computation, potential inference, potential cognition. You cannot fake it. You cannot print more of it. You can only generate it or you cannot.

That is what makes it the bottom of the trust stack. You can fabricate narratives. You can manipulate search rankings. You can forge attribution chains. But you cannot forge a kilowatt-hour of delivered work. Energy is the layer where trust becomes physics.

The Lords of Zero

There is an economic dimension to this that goes beyond philosophy.

I've written about what I call the Lords of Zero, the entities that sit at the point where AI drives costs toward zero but access stays gated. Inference costs collapse. Training costs collapse. Design, simulation, modeling, all collapsing. But API pricing stays sticky. Model access stays permissioned. Compute stays controlled. Someone is capturing the delta between abundance and permission, and that delta is where the new power concentrates.

The mechanism is simple and old. When the marginal cost of producing something falls toward zero, the price doesn't follow unless competition forces it. If access is gated, the price stays high, the cost stays low, and the gap becomes pure extraction. This is what every platform business has done, but AI does it at a new scale because the underlying cost compression is unprecedented.

The kWh token is the antidote to that delta.

You cannot create a permission gap on a kilowatt-hour the way you can on inference. A kilowatt-hour delivered is a kilowatt-hour delivered. You cannot mark it up by controlling the gate between what it costs and what you charge for it, because the unit itself is the work, not the access to the work.

The Lords of Zero exploit the abstraction gap between the thermodynamic cost of cognition and the price of access to that cognition. An energy-denominated economy collapses that gap. It makes the cost of thinking legible and the markup visible. It turns the invisible tax of cognitive feudalism into something you can read on a meter.

This is why the kWh token is not just a philosophical preference. It is a structural defense against the most predictable failure mode of the AI economy. A failure mode that is already here, already operating, already extracting.

The Lords of Zero are not confined to the physical front. The same entities that gate access to inference also shape what inference gets produced about, by controlling which models reach scale and what those models were trained on. They are dual-substrate operators, which is part of why their position is so durable.

Energy as Sybil Defense

There is one more dimension worth naming, because it becomes critical as autonomous agents proliferate.

When millions of micro-agents can generate pitches, proposals, and transactions at machine speed, signals drown in noise and trust collapses under abundance. The cost of action falls to zero, which means the cost of fake action also falls to zero. Spam stops being a delivery problem and becomes an ontological problem. How do you tell a real intent from a generated one when both are produced by the same process at the same cost?

The kWh is the answer the physics already provides.

Every transaction denominated in energy carries an implicit proof of work, not the Bitcoin kind, but the thermodynamic kind. You cannot mint a kilowatt-hour without doing real physical work: burning fuel, capturing photons, splitting atoms. That makes it a natural defense for an economy increasingly populated by agents.

A thousand spoofed bots can flood an API with worthless requests, but a thousand kilowatt-hours represent a thousand units of real physical work that someone, somewhere, had to generate. Energy is the one token that cannot be counterfeited by computation alone, because computation itself is what it pays for.

This is the friction the system needs. Not artificial friction imposed by gatekeepers, but natural friction baked into the laws of thermodynamics. The kind no one can repeal.

The Honest Inflation

There is an obvious objection to the kWh token, and it is worth addressing directly: isn't it inflationary?

Yes. It is. And that is the strongest argument in its favor.

As civilization builds more energy infrastructure, more solar farms, more nuclear plants, more fusion reactors if we get there, the supply of kilowatt-hours increases. The token inflates. Each individual kWh becomes nominally cheaper relative to the total supply.

But here is the difference between honest inflation and the kind we are used to.

Fiat currencies inflate when central banks expand the money supply. That expansion is disconnected from productive capacity. Printing more dollars does not mean more work can be done. It means each dollar buys less. The inflation is a tax on stored labor, imposed by institutions, with no necessary relationship to the underlying capacity of civilization.

A kWh token inflates only when more energy is actually produced. The money supply grows precisely when, and only when, the civilization's ability to do work grows. Inflation becomes a direct signal that the economy's productive capacity has expanded. Not a signal that someone with access to the printing press decided to dilute the pool.

It is the difference between watering down the soup and building a bigger kitchen.

This distinction matters enormously when you think about what happens if we break through the next energy ceiling.

If fusion arrives, or orbital solar, or any of the energy breakthroughs that have been perpetually twenty years away, the kWh token inflates massively. The unit price of energy drops toward zero. And that sounds like a problem until you realize what it actually means: unlimited energy is unlimited potential work. Unlimited computation. Unlimited inference. Unlimited agency. The token devalues in nominal terms but the economy it denominates explodes in real capacity. Everyone gets richer in capability even as the unit price drops.

That is the Kardashev window opening, denominated in honest units.

Compare that to what happens under fiat when energy becomes abundant. The energy producers sell cheaper. The financiers and platform owners capture the delta between collapsing costs and sticky prices, the Lords of Zero dynamic again. The price signal gets laundered through layers of abstraction until nobody can tell whether civilization got more capable or just more leveraged. The abundance is real but the accounting obscures who benefits.

Now consider the incentive structures.

A deflationary token like Bitcoin rewards hoarding. The rational strategy is to hold, not to spend, because your tokens will be worth more tomorrow than today. That is a recipe for economic paralysis dressed up as sound money.

A kWh token rewards building generators.

If your money supply grows by producing energy, the incentive structure of the entire economy tilts toward building more energy infrastructure. Solar farms become literal money printers, but the kind that increase civilization's actual capacity rather than just its nominal wealth. Fusion reactors become the most valuable machines ever built, not because they produce scarce tokens, but because they produce abundant ones backed by real work potential.

Gold rewards digging holes. Fiat rewards proximity to the printer. Bitcoin rewards burning energy for proof-of-work puzzles that produce nothing useful. A kWh token rewards producing energy that can do actual work: computation, manufacturing, transportation, cognition.

That is a fundamentally different incentive structure than anything civilization has operated under before. It aligns the monetary system with the physical trajectory of a species trying to climb the Kardashev scale. The more energy you produce, the more money exists, and the more work civilization can do. Inflation and expansion become the same thing.

The honest inflation of a kWh token is not a flaw in the design. It is the design.

The Local Cognition Factory

Over the years I've become something of a collector of used solar panels.

Not the newest, highest-efficiency models. Often they are panels removed from older grid-tied systems, equipment that someone else considered outdated or surplus. They end up stacked around my property, quietly producing electricity whenever the sun rises.

It started as a hobby. It has become something closer to a working theory.

Two of my homes run on NEM2 solar systems I built before the PG&E policy window closed. My primary residence and the lake house both have rooftop arrays that feed back to the grid under the old net metering rules, which means the kilowatt-hours I generate during the day are credited at retail rates against the kilowatt-hours I draw at night. The first system nearly paid for itself in the first year. The second one is following the same curve. Behind both is a bank of Enphase inverters doing the actual work of converting DC to AC, syncing to grid frequency.

There is also a small off-grid cabin system: a 14 panels, a EG4 split phase inverter/charge controller, and a modest 16kWh battery. No grid connection. No utility account. Just sunlight in, electrons stored, work done.

What I have built, without quite meaning to, is a small-scale demonstration of the entire essay.

Each solar panel is a machine that mints kilowatt-hours from sunlight. Each kilowatt-hour represents potential work. The panels feed the inverter. The inverter feeds the house. The house feeds the workstations and the network gear and the local model boxes I run for inference and experimentation.

The full pipeline is right there, running in my yard:

Sunlight → Photovoltaic conversion → Inverter → Compute → Inference.

When my local language model generates a token, that token was paid for in photons that hit my roof some number of hours earlier. The energy moved through silicon twice: first in the panel, where photons knocked electrons into a current, then in the GPU, where that current drove matrix multiplications that produced language. Two layers of silicon, separated by an inverter and a few hundred feet of copper, doing the same fundamental thing the universe has been doing since the first stars formed: turning raw energy into structured pattern.

I am not arguing this is the future of civilization. Most people will not stack used solar panels in their yards. Most people will not run their own inference. But the architecture is the point. The full vertical stack from photons to cognition can fit on a residential lot. The substrate war is not abstract. It is something you can build, in pieces, with parts.

That changes what the kWh token means. It is not just a unit of account in some hypothetical future economy. It is the actual measure of what comes out of my roof and goes into my GPUs. The thermodynamic ledger is already running. The only question is whether the economy above it will eventually be denominated in the same units, or whether it will keep pretending the substrate doesn't matter.

Closing the Loop

Forty years ago I started a company called NRG.

The tagline was simple:

Energy: the capacity for doing work.

That definition still holds. But now we can see the full arc of its implications, not just the four-word pipeline, but the entire vertical stack from physics to mind.

Energy powers computation. Computation enables cognition. Cognition generates agency. Agency creates value.

And at every step of that ascent, there is a cost. A thermodynamic cost, measurable in joules, visible in heat dissipated, legible in the thermal signature of a GPU learning to predict the next token.

My about page has always carried a version of this pattern:

Energy → Flow → Computation → Meaning.

I wrote that before AI made the pattern obvious to everyone else. Before datacenters started competing for grid capacity. Before the Substrate War made energy the contested resource of the century.

But the pattern was always there. In NRG's tagline. In the voltage regulators warming up as packets moved through my embedded stacks. In the solar panels stacked around my property, quietly minting potential work from sunlight. In the local inference running on photons that hit my roof a few hours earlier.

The kilowatt-hour is not waiting for permission to become the unit of account. The substrate is already running. Every token a model generates, every inference served, every decision an agent takes is being paid for in kilowatt-hours whether the economy above it acknowledges the bill or not.

What's left to decide is whether the accounting catches up to the physics, or whether civilization keeps pricing its cognition in abstractions that float free of the substrate paying for them. The first path is honest. The second is borrowed time.

Or to put it the way I first understood it, watching a GPU get hotter as a model got smarter:

Energy becoming inference. Matter learning what it means to think.