"The AI revolution is not primarily a software story. It is an infrastructure story."
A sweeping account of the physical systems that power civilisation — from Roman aqueducts and Victorian sewers to submarine cables, GPU farms, and immersion-cooled AI infrastructure.
Every technology revolution in history has rested on a physical foundation. The industrial revolution needed coal, steam, and railways. The internet age needed fibre, data centres, and electricity grids. The AI era needs all of those — plus cooling systems that did not exist five years ago.
Infrastructure traces this story across six parts and thirteen chapters, from the Roman engineers who moved water across mountains to the engineers designing immersion-cooled GPU clusters for the next generation of AI models.
"Every technology revolution rests on a physical foundation. Understanding that foundation is not optional for anyone who wants to understand where the world is going."
6 PARTS · 13 CHAPTERS · 260 PAGES
A complete journey from ancient water infrastructure to AI-era immersion cooling.
How civilisations were built on water, and why that never stopped being true
Roman aqueducts lasted five centuries. Victorian sewers last two hundred years. The infrastructure we build today will outlast every business model built on top of it.
The hidden relationship between water and power that shapes every infrastructure decision
Hydropower generates 16% of global electricity and over 90% in countries like Norway and Iceland. Pumped hydro accounts for over 90% of grid-scale storage capacity worldwide.
The architecture of electricity — a legacy of a nineteenth-century dispute
The 60 Hz/110 V standards that Edison's era bequeathed to North America and the 50 Hz/230 V standards of Europe are now permanent. Infrastructure lock-in is real and long-lasting.
Why solar and wind are economically irreversible, and what that means for data centres
Solar and wind are now the cheapest sources of new electricity generation in most of the world. The energy transition is economically irreversible — the question is only speed.
The engineering marvel that became a geopolitical weapon
Over 1.3 million kilometres of high-pressure pipelines carry oil and gas across the world. Europe restructured its gas supply chain substantially in under two years after the 2022 Ukraine invasion.
Stranded assets, hydrogen, and the finite transition window
The potential stranded asset risk from fossil fuel infrastructure is $1–4 trillion. Natural gas is a transition fuel with a finite transition window — the window is closing.
Containerisation as the template for every infrastructure standard
The twenty-foot equivalent unit (TEU) is one of the most consequential design decisions in economic history. COVID-19 permanently repriced supply chain resilience.
95% of international data travels through submarine cables — not satellites
Google, Meta, Amazon, and Microsoft collectively own or co-own over half of the world's undersea cable capacity. The internet is physical infrastructure with specific, mappable vulnerabilities.
The factories of the information economy — and why they are now critical infrastructure
PUE (Power Usage Effectiveness) is a financial metric, not just a technical one. The difference between a PUE of 1.05 and 1.5 represents hundreds of millions of dollars over a facility's lifetime.
GPU power density is increasing faster than infrastructure can adapt
From 250 watts per GPU five years ago to 700 watts today, heading toward 1,000+ watts. Grid connection has replaced planning permission as the binding constraint in every major data centre market.
Why air cooling has reached its thermodynamic limit for AI workloads
The specific heat capacity of air is fixed at 1.005 kJ/kg·°C. This is not an engineering problem — it is a physics constraint. Air cooling cannot scale to meet AI's thermal demands.
Three mature approaches — choose deliberately
Single-phase immersion, two-phase immersion, and direct liquid cooling each have distinct thermal envelopes, fluid chemistries, and total cost profiles. The fluid selection is a strategic infrastructure decision.
Site decisions made today determine the 2030 landscape
Three-to-five year lead times mean the facilities operating in 2030 are being designed now. The transition requires executive ownership, not engineering delegation.
"The facilities operating in 2030 are being designed today. The decisions being made now — about sites, power, cooling, and suppliers — will determine competitive position for a decade."
Olafur V Sigurvinsson — Infrastructure
Water runs through every chapter of this book. It powers hydroelectric dams, cools nuclear reactors, is consumed by data centres, and — in the form of dielectric fluid — is the medium through which the most advanced AI processors are now cooled. The water thread connects Roman engineers to GPU engineers across two millennia.
Air cooling has been the default for computing infrastructure for seventy years. The book traces why this is ending: the specific heat capacity of air is a physics constant, not an engineering variable. As GPU power density crosses 700 watts per chip, air cooling becomes thermodynamically inadequate.
The 60 Hz standard, the TCP/IP protocol, the twenty-foot container — infrastructure standards, once established, become permanent. The book examines how lock-in works, why it is so difficult to escape, and what it means for decisions being made today about AI infrastructure.
Every infrastructure era has a binding constraint — the resource or capability that limits everything else. In the AI era, that constraint is not compute, capital, or software. It is grid connection and cooling capacity. Understanding this changes how you think about AI strategy.
The $1–4 trillion of potential stranded fossil fuel assets is the largest capital allocation risk in economic history. The book examines how this risk propagates through energy markets, data centre power procurement, and corporate infrastructure strategy.
Total cost of ownership, not capital expenditure, is the correct metric for infrastructure decisions. The difference between a PUE of 1.05 and 1.5 represents hundreds of millions of dollars over a facility's lifetime. Liquid cooling's TCO case is stronger than its PUE case alone.
Infrastructure publishes in Q2 2026. Register your interest to be notified when it becomes available — and receive an exclusive excerpt from Chapter 1.