Ugrás a tartalomhoz
All analysis
The Altar of Efficiency · Part 3

The Gigawatt War and the Cantillon Effect

The human brain runs on 20 watts — AI server farms are wired to millions. Who pays?

The Danube Lens·3 June 2026

For years we lived in a comfortable, sterile and reassuring illusion: that the digital revolution was unfolding somewhere above us, in the weightless, pristine and invisible "Cloud". Society believed software, gigabytes and code had no physical footprint.

When autonomous AI agents tore through the white-collar workforce (as we showed in Part 1), and then $10,000 physics-literate humanoid robots entered factories and lights-out facilities (Part 2), the world narrowed its gaze to almost a single question: how many people would lose their jobs to the machines?

Yet behind the scenes, beyond the black walls and industrial fencing of restricted-access data centres, a darker, rawer and more destructive war had already begun. A fight for the hardest, most unforgiving currency of physical reality. It is no longer the dollar, no longer digitally minted Bitcoin, and no longer the gold of central banks.

The true, ultimate currency of the 21st century is the Gigawatt.

The 20-Watt Mind vs. the Megawatt Monster

To grasp the infrastructural and economic strains bearing down on us, we must strip human performance and artificial intelligence back to pure physics and thermodynamics.

Take an average middle manager sitting at a desk for eight hours. They pore over a complex logistics spreadsheet, conduct video calls, drink coffee and make critical business decisions. The human brain doing this unfathomably complex intellectual work draws roughly 20 watts. That is the draw of a dim fridge bulb, or a modern LED lamp. Biology is the most efficient, most brilliant mechanism in the universe.

Set that against what happens when a corporation sacks that worker and replaces them with Anthropic's Claude or Google's Gemini 3.1 Pro autonomous agents. The split-second, brilliant response comes neither from thin air nor from some ethereal "Cloud". It is produced by tens of thousands of Nvidia GPUs whirring inside football-pitch-sized server farms, cooled by tens of thousands of litres of water and ranks of industrial fans. Their energy demand is measured not in watts but in megawatts.

Average energy consumption of a single complex online operation (Wh)
Traditional Google search
0.3 Wh
Generative AI (LLM) prompt
2.9–5.0+ Wh
Source: IEA (International Energy Agency) and industry estimates, 2024–2026
The human brain vs. an AI data centre — live energy consumption
🧠
The human brain
20 watts
during your time on this page
0.0 J
A neural network of 86 billion neurons — a flickering green light
🏭
One LLM data centre
200 megawatts
during your time on this page
0.00 kWh
Infrastructure serving a single GPT/Claude query
Ratio — the AI data centre draws 10,000,000× more energy
Human
AI
The green sliver (~2 px) is the human brain's draw — to scale against the full red bar.
UK household annual consumption: ~3,500 kWh
200 MW data centre — per hour: 200,000 kWh
Households equivalent to one AI data centre: ~57×
Sources: IEA Energy Efficiency Report 2024; Goldman Sachs AI Power Demand study. Data-centre consumption is a modelled average (±50% depending on query type).

The tech-PR machine has spent years running colossal campaigns to reassure the world. Silicon Valley lobbyists have repeated the same mantra at every climate conference: "AI is so efficient that it saves energy globally! If algorithms replace coders and managers, we will no longer need to heat office buildings, and commuting emissions will vanish."

A perfect piece of greenwashing. And, by the iron laws of economics, a colossal, deadly lie.

THE JEVONS PARADOX — WHY GREATER EFFICIENCY MEANS MORE POWER CONSUMPTION

To understand why the grid is crumbling beneath AI, we have to go back to the 19th century. The English economist William Stanley Jevons noticed something curious: when more efficient steam engines were invented — consuming far less coal — people assumed England's coal use would fall.

Instead, aggregate demand for coal exploded. Because the steam engine had become cheaper and more efficient, it was suddenly deployed everywhere.

In 2026, the same Jevons paradox is playing out in artificial intelligence. AI may produce a single spreadsheet with a smaller carbon footprint than a commuter in an office. But because coding and analysis have become virtually "free" and instantaneous, corporations are running ten thousand operations, video generations and synthetic simulations in the cloud every minute. That efficiency has bred such hyper-consumption that it has sent the planet's aggregate power demand through the roof.

Silicon Valley Buys Nuclear Reactors

By 2024, the tech giants — Microsoft, Amazon, Meta and Google — had awoken in near-panic to a brutal physical ceiling. They understood that it mattered little how clever the world's algorithms became, or how many trillions they spent on Nvidia's latest rack servers, if there was nothing to plug them into the wall. Unless they secured their own dedicated, immediately available power source, their astronomically expensive AI models would simply shut down.

Solar and wind are too intermittent for a data centre that must orchestrate billion-dollar processes with 99.999% uptime. Coal is politically untenable. One path remained: nuclear.

  • Microsoft and Three Mile Island: the tech giant signed a 20-year exclusive power-purchase agreement to restart the infamous Three Mile Island nuclear plant in Pennsylvania (under the name Crane Clean Energy Center). The 835 MW of carbon-free electricity produced there will feed exclusively into Microsoft's data centres.
  • Amazon and Talen Energy: Amazon Web Services (AWS) struck a 1.9 GW nuclear deal with Talen Energy, securing data-centre capacity directly adjacent to the Susquehanna nuclear plant and stable nuclear baseload through a PPA (power-purchase agreement).

These companies are no longer merely software developers. They are the new energy cartels of the 21st century. Whoever owns the compute capacity and the gigawatts owns the world. But how did they afford all this? Here the system's darkest economic secret is laid bare.

THE CANTILLON EFFECT AND THE GREATEST THEFT IN HISTORY

Seen through the lens of Austrian economics, the Gigawatt War is not merely a technological race but a textbook case of the Cantillon effect (the way new money enriches whoever receives it first).

Richard Cantillon, an 18th-century economist, laid it out: when governments create new money out of thin air, it does not spread evenly. It reaches those "closest to the printing press" — the banks, the venture capitalists and the Big Tech oligarchs — first. They can buy real, physical assets (nuclear plants, data centres) with this new money before inflation drives market prices up.

By the time the trillions in global monetary expansion and tech subsidies trickle down into the real economy, the money has lost its purchasing power. The average person — at the very end of the chain — is left holding nothing but a steeper electricity bill. The tech elite has used state-created money to buy up physical reality; the tab is picked up by an impoverished society through currency debasement.

Big Tech's energy buying spree — nuclear and renewable deals (2023–2025)
Combined capacity: 7+ GW
Deals: 7
Buyers: Microsoft · Amazon · Google
🌎North America4 deals
Microsoft2023
Three Mile Island
Pennsylvania, USA
835 MW Nuclear
Active (2024 restart)
Amazon2024
Susquehanna / Talen Energy
Pennsylvania, USA
1.9 GW Nuclear
Data-centre campus + PPA
Google2023
Vogtle nuclear plant
Georgia, USA
500 MW Nuclear
PPA (through 2030)
Amazon2024
Pacific NW hydro
US Pacific Northwest
~2 GW💧 Hydro
In progress
🖥️
AI DATA
CENTRE
🌍Europe3 deals
Microsoft2024
Rolls-Royce SMR
United Kingdom
470 MW/unit Small modular reactor
In development
Google2024
EDF nuclear
France
~1 GW Nuclear
PPA signed
Amazon2025
Vattenfall
Sweden
~800 MW🌬 Wind
In negotiation
MicrosoftAmazonGoogleNuclear   ⚛ SMR   💧 Hydro   🌬 Wind
Sources: Reuters, Bloomberg, ESG Today, S&P Global Market Intelligence. Capacity figures reflect announced or contracted volumes.

The "Toxic Combo": In the Vice of Chaos

The independent macro analyst Lyn Alden's theory of the "Toxic Combo" is no longer an abstract financial warning. It is the hard reality of everyday life. The formula is a two-way trap that slowly but surely crushes the 99%.

While Microsoft and Amazon buy themselves stable, cheap and clean nuclear power for their private data centres, the domestic grid, the average person and small businesses are left to fend for themselves on the open market. As geopolitical chaos and the AI-driven drain on grid capacity intensify, energy prices keep climbing for the ordinary consumer. Stagflation (economic stagnation coupled with high inflation) sets in.

The middle class — the former white-collar manager and the blue-collar warehouse worker alike — finds itself caught in the same relentless squeeze:

From Below: AI (the deflationary squeeze)
  • The machine strips away the worker's wage-bargaining power
  • Job numbers fall and the real value of remaining wages collapses
  • The worker is made redundant by an algorithm
From Above: stagflation (the inflationary squeeze)
  • Power-hungry AI and geopolitical turmoil push prices higher
  • The rising cost of housing, food and electricity
  • Savings are eroded by the Cantillon effect

The real victim is the average person who simultaneously loses their job to algorithms and cannot pay the electricity bill — because the very same Microsoft server farm has siphoned power from the grid that powers the agent responsible for replacing them.

Without power, the smart robot is just a dead lump of metal in the corner. But the world's richest corporations and governments have made sure the server parks never go dark. And when the masses lose both their financial stability and their economic utility relative to the machines, history's most dangerous, darkest psychological experiment begins.

The Toxic Combo's two-way trap — two forces, one victim
↓ PRESSURE FROM ABOVE — Stagflation & energy inflation
Energy prices: +80%
Food inflation: +35%
Bond yields: ↑ debt burden
Inside the squeeze
👤 Office worker
👤 Factory hand
👤 Service staff
👤 Bookkeeper
👤 Administrator
👤 Call-centre agent
"Losing the job and the purchasing power in the same breath."
↑ PRESSURE FROM BELOW — AI wage compression
Mass layoffs: ↓ wage level
AI as labour: $0.002/task
RPE pressure: Automation-first
⚠️Toxic Combo: Stagflation-driven price rises and AI-driven wage compression hit the middle class at the same time — a double bind of real-wage collapse and employment insecurity.

UP NEXT: in Part 4 of this series (The Aimless Society and the UBI Mirage), we explain why Universal Basic Income (UBI) is a stillborn, destructive mirage seen through the lens of Austrian economics. We turn the spotlight on shocking statistics from the medical journal The Lancet on the link between unemployment and mortality, and ask the ultimate question: what becomes of humanity when "work" ceases to exist forever?

Share
Comment filter
1
14710
Pick a level to filter comments
0 comments

No comments yet.