The Accidental Infrastructure
How the most critical systems get built when everyone's optimising locally
Welcome back to The Ledger - a weekly briefing of what’s happening inside complex systems around industrial decarbonisation. Let’s get straight into Issue #9.
THE OPENING ENTRY
Nobody decided that 70% of the world’s internet traffic should flow through Virginia.
No commission voted for it. No strategic plan recommended it.
What happened instead: Thousands of individual companies made economically rational decisions about where to site data centres. Cheap power. Fibre proximity. Tax incentives. Each decision made perfect sense in isolation.
The aggregate result: A systemic dependency nobody planned for, nobody voted on, and nobody can now easily undo.
This is the pattern I keep seeing: The most consequential infrastructure decisions aren’t decisions at all. They’re outcomes that emerge from accumulated micro-optimisations where nobody with authority actually chose the result.
This week: Five times critical infrastructure got built through a process where nobody was deciding - and what it means when you realise the system you’re depending on was never actually designed.
Everyone optimises locally. The system drifts globally.
FIELD REPORTS
Virginia: The Accidental Internet
The world’s single biggest hub of global internet traffic.
In the late 1990s, a few data centres opened in northern Virginia. Proximity to government agencies, decent fibre infrastructure, competitive power rates. Sensible choices.
More companies noticed. If your data centre needs low-latency connections to other data centres, and many data centres are already in Virginia, then Virginia is where your data centre should be. Network effects.
More data centres arrived. Which justified more fibre investment. Which attracted more data centres. Which justified more fibre. Standard agglomeration dynamics.
Today: An unreasonably large share of global internet traffic passes through northern Virginia at some point (though not the oft-quoted 70%). AWS’s us-east-1 region - the default for millions of applications - sits there. A significant chunk of the world’s digital infrastructure has a single geographic point of concentration.
This wasn’t planned. It emerged from thousands of individual siting decisions that each made local sense.
The result: A systemic risk that’s now weight-bearing. You can’t easily move us-east-1. Too much depends on its existence exactly where it is.
The lesson: Optimising for individual facility economics can create system-level geography that nobody chose and nobody can now easily change.
China: The Rare Earth Bottleneck
85% of rare earth processing.
In the 1980s and 1990s, rare earth mining existed in multiple countries. Processing and refining existed in several places. Reasonable geographic distribution.
Then China invested heavily in rare earth processing capacity. Lower labour costs, fewer environmental restrictions, economies of scale. Economically rational.
Western firms noticed: Cheaper to ship ore to China for processing than to process domestically. Individual companies made spreadsheet decisions. Lower costs, better margins.
More processing capacity moved to China. Which created more expertise there. Which made it harder to maintain expertise elsewhere. Which made China more cost-competitive. Which moved more processing there.
Today: Roughly 85–90% of rare earths (and nearly all heavy rare earths) processing happens in China. Not because anyone decided this was a good geopolitical arrangement. Because thousands of individual procurement decisions optimised for unit cost.
The result: Every mobile phone, every wind turbine, every electric vehicle motor depends on processing capability that’s geographically concentrated in ways nobody planned.
And now you can’t easily rebuild distributed processing. The expertise has atrophied elsewhere. The supply chains are configured around the current geography. The economics only work at scale.
The signal: Supply chain optimisation at the firm level can create strategic dependencies at the system level that nobody intended.
Container Ships: The Port Problem
The rise of megaports.
In the 1960s, container ships were modest in size. Most ports could handle them. Reasonable infrastructure distribution.
Then shipping lines realised: Bigger ships have better unit economics. Each incremental increase in ship size reduces cost per container. Individual firms optimised for scale.
Ships got bigger. Then bigger again. Then bigger again. Each generation of mega-ships made sense for the shipping line ordering them.
Today: The largest container ships can only be handled by about a remarkably small number of ports globally.
Now only a limited set of deep-water ports with the right cranes, berths and channel depths can handle 24k-TEU-class ships, concentrating traffic into a small number of megahubs.
This didn’t happen because anyone decided “let’s make global trade depend on twelve infrastructure chokepoints.” Every spreadsheet told the same story.
The result: A port that can’t handle the largest ships loses traffic to ports that can. Which reduces its revenue. Which makes it harder to invest in upgrades. Which makes it less competitive. Concentration accelerates.
Many mid-tier ports are effectively bypassed by the most efficient services, ceding volume to the megahubs. Nobody made a bad call - just a narrow one.
This isn’t strategy. It’s gravity.
Taiwan: The Chip Chokepoint
90% of the most advanced chip manufacturing.
Semiconductor manufacturing used to be geographically distributed. The US had fabs, Europe had fabs, Japan had fabs, Korea had fabs. Reasonable resilience.
Then Taiwan invested in advanced process technology. TSMC became extraordinarily good at manufacturing leading-edge chips. Better yields, faster process improvements, lower costs at scale.
Fabless chip designers noticed: Better to design chips and have TSMC manufacture them than to maintain your own fab. Individual companies made rational decisions. Lower capital requirements, faster time to market, better economics.
More design firms went fabless. Which gave TSMC more volume. Which improved their economics. Which made them more competitive. Which encouraged more firms to go fabless.
Today: Roughly 90% of the most advanced chip manufacturing happens in Taiwan. Not because anyone thought “let’s make one island the lynchpin of global technology supply chains.” Because thousands of firms optimised their individual manufacturing strategy.
The result: Every smartphone, every data centre, every advanced weapons system depends on manufacturing capacity in a geography that nobody chose as a strategic plan.
And now you can’t easily rebuild that capability elsewhere. It took 30 years to build the expertise. You can’t recreate it with a five-year government programme, no matter how much capital you throw at it.
The mechanism: Outsourcing manufacturing to the most capable provider can create concentration risk that nobody wanted.
Cloud Regions: The Sovereignty Trap
Three companies.
In the early 2010s, cloud computing was optional. Organisations ran their own data centres, used cloud for overflow, maintained control of their infrastructure.
Then cloud providers got very good at operations. Better reliability, lower costs, faster feature velocity. Individual organisations made rational decisions: Move to cloud.
More workloads moved to cloud. Which gave providers more scale. Which improved their economics. Which made them more competitive. Which moved more workloads to cloud.
Today: Critical government services, healthcare systems, financial infrastructure - all running in data centres operated by three companies, in specific geographic regions those companies chose for their own operational reasons.
This wasn’t designed. Individual agencies made procurement decisions that optimised for cost and capability.
The result: Systemic dependencies on infrastructure that wasn’t designed for the criticality it now bears. With many public and private services concentrated on a few hyperscale regions, regional outages (e.g., AWS us-east-1 in 2021; Azure incidents in 2023–24) have cascaded into widespread disruptions.. By the time anyone notices the dependency, it’s load-bearing.
The lesson: Technology adoption decisions that make sense at the organisational level can create sovereignty risks at the national level that nobody chose.
WHAT THE LEDGER REVEALS
Five different systems. Five different decades. Same underlying dynamic:
Nobody decided to create single points of failure. Nobody chose to create geographic concentration risk. Nobody planned to make critical systems depend on infrastructure that wasn’t designed for that criticality.
What happened: Everyone optimised locally. The system drifted globally.
The mechanism:
Each actor optimises rationally. Lower costs, better performance, faster delivery—whatever the local optimisation target is.
Early movers create agglomeration effects. Being where others are becomes advantageous. Network effects, economies of scale, knowledge spillovers.
The system tips. What started as marginally better becomes structurally better. The gap widens. Concentration accelerates.
Alternative paths atrophy. Expertise leaves. Supply chains reconfigure. Infrastructure adapts to the current state.
By the time anyone notices the dependency, it’s load-bearing. Too much depends on the current configuration to easily change it.
The result: Infrastructure that nobody decided to build.
Virginia wasn’t chosen as the internet’s physical home. It became that through agglomeration.
China wasn’t selected as the rare earth processing centre. It became that through cost optimisation.
The largest ports weren’t chosen as trade chokepoints. They became that through ship size optimisation.
Taiwan wasn’t designated as the chip manufacturing lynchpin. It became that through capability concentration.
Cloud regions weren’t selected as critical infrastructure. They became that through adoption acceleration.
The problem isn’t that individual decisions were wrong. Each one made sense in isolation.
The problem is that individual optimisation creates system outcomes that nobody chose - and nobody can easily undo.
When everyone optimises locally, the system drifts to configurations that serve individual economics but create collective fragility.
And there’s no mechanism to stop it. Because nobody’s deciding. There’s no single point where someone looks at the aggregate pattern and says “wait, should we really be doing this?”
What gets built when nobody’s deciding? Whatever emerges from accumulated micro-decisions that optimise for local efficiency and ignore system-level fragility.
And we’re all depending on it.
THE OPERATOR’S PLAYBOOK
Five questions to ask before making decisions that seem locally optimal. Because what makes sense for you might be creating system-level outcomes nobody wants.
1. What’s the aggregate if everyone does this?
Your decision makes sense in isolation. But if every similar organisation makes it, what system-level pattern emerges?
Cheapest provider? Everyone picks the cheapest provider. Concentration risk nobody chose.
Most capable manufacturer? Everyone picks the same one. Geographic dependency nobody wanted.
Best technology platform? Universal adoption creates systemic dependency.
Ask: If my decision makes sense, doesn’t everyone’s similar decision also make sense? If everyone makes this decision, what emerges? Is that an outcome anyone actually wants?
Don’t just optimise locally. Think about the system your local optimisation is creating.
2. Am I creating concentration risk I’ll regret later?
Today’s optimisation becomes tomorrow’s constraint.
Outsourcing to the most capable provider gives you better economics now. But if everyone outsources to the same provider, you’ve created a chokepoint you’ll need to route around later.
Consolidating suppliers reduces complexity now. But if everyone consolidates to the same suppliers, you’ve created systemic dependencies you’ll need to hedge against later.
Ask: Am I creating tomorrow’s problem to solve today’s optimisation? Will I look back in five years and wish I’d maintained alternatives even though they were more expensive?
Short-term efficiency often trades off against long-term optionality. Make sure you’re making that trade consciously, not accidentally.
3. What am I assuming will always be available?
When you optimise based on current infrastructure, you’re implicitly assuming that infrastructure will remain accessible, capable, and cost-effective indefinitely.
But infrastructure that exists because of agglomeration can disappear if agglomeration reverses. Infrastructure that exists because of policy can disappear if policy changes. Infrastructure that exists because one company invested can disappear if that company’s strategy shifts.
When we evaluate thermal infrastructure projects, this is the question that reveals whether we’re depending on industrial heat patterns that might not exist in five years.
Ask: What infrastructure am I treating as given? What would I do if that infrastructure became unavailable, uneconomical, or strategically compromised?
Don’t optimise as if the current state is permanent. It isn’t.
4. Am I contributing to a pattern I don’t want to see continue?
Sometimes you know the aggregate pattern is problematic, but your individual decision seems too small to matter.
“Sure, everyone sourcing from one geography creates concentration risk. But my individual decision won’t change that pattern. I might as well get the cost savings.”
This is rational at the individual level. It’s a disaster at the system level.
Everyone reasons the same way. Everyone free-rides on the assumption that others will maintain alternatives. Nobody maintains alternatives. The concentration you didn’t want happens anyway.
Ask: If I can see the pattern is problematic, am I contributing to it? What if everyone’s making that same assumption?
Sometimes the right answer is to deliberately maintain a less efficient path, even though it costs more, because you need system-level diversity that individual optimisation destroys.
5. Can I see the dependency forming before it’s load-bearing?
The time to worry about concentration risk is before it’s concentration risk - when it’s just “lots of people making the same locally rational decision.”
Virginia wasn’t a single point of failure in 2000. It became one by 2015. But you could see the pattern forming in 2005 if you were watching.
Rare earth processing wasn’t a bottleneck in 1995. It became one by 2010. But you could see the pattern forming in 2000.
Cloud regions weren’t critical infrastructure in 2015. They became that by 2020. But you could see the pattern forming in 2017.
Ask: What patterns of concentration am I seeing form around infrastructure I depend on? What would it look like in five years if this pattern continues?
Don’t wait until the dependency is load-bearing to worry about it. By then it’s too late. Watch for the pattern formation, not the pattern completion.
THE LEDGER ENTRY
What gets built when nobody’s deciding is whatever emerges from everyone optimising locally.
FIELD NOTES
Nine issues in, the patterns are sharpening.
The pattern this week: Systems that emerge from accumulated local decisions that nobody actually chose at the system level. The geography of digital infrastructure, the concentration of processing capabilities, the dependencies we’ve created without planning for them.
If you’re seeing similar patterns - infrastructure dependencies that formed through drift rather than design, concentration risks that nobody chose but everyone contributed to, system-level fragility that emerged from individual efficiency—I want to understand them.
Where are you seeing local optimisation create system outcomes nobody wants? What dependencies are forming that nobody’s choosing? What concentration risks are building that everyone can see but nobody’s stopping?
Hit reply or reach me at hello@tldgr.com. The most useful observations show what changed and when.
Thanks for reading, and please share The Ledger if you have found it useful or insightful in any way.
Here’s to what’s possible
Dom
The Ledger is a weekly field report on where power actually sits in complex systems.
Subscribe at tldgr.com · Read the archive