I Tested 2,000 Years of Empire Collapse. Nobody Saw This Coming
In this article
> **Bottom line:** Historians Joseph Tainter and Peter Turchin have mapped the collapse mechanics of more than 20 complex societies — Rome, the Ottomans, the Soviet Union, Ming China — and invasion is rarely the real cause.
The pattern is diminishing returns on complexity combined with elite overproduction: too many credentialed people competing for too few real positions of power.
Silicon Valley in 2026 is showing both signals simultaneously — hundreds of billions in annual AI infrastructure spend chasing shrinking marginal returns, and a computer science graduate glut competing for a shrinking number of entry-level engineering jobs.
This isn't a metaphor. It's the same math, run on a different substrate.
Stop Calling It "Disruption"
I spent three weeks going down a YouTube rabbit hole of collapse historians instead of doing my actual job. I'm not proud of it.
But somewhere around hour twelve, watching a lecture on why the Western Roman Empire didn't fall to barbarians so much as it fell to its own administrative weight, I had the kind of uncomfortable realization that ruins a Tuesday.
We keep describing what's happening in tech right now — the AI capex arms race, the layoffs next to record profits, the credential inflation — as "disruption." Something new.
Something that requires new vocabulary.
It doesn't.
I've been building software companies for over a decade, and I'm telling you: what we're watching is the oldest pattern in recorded civilization, and the people running our biggest companies either haven't read the history or are betting you haven't.
The Sacred Cow: "Tech Moves Too Fast for History to Apply"
Here's the belief everyone in this industry holds, usually without examining it: technology changes so fast that historical precedent is basically useless. Rome didn't have cloud compute.
The Ottomans didn't have venture capital.
Comparing empire collapse to a software industry cycle feels, at best, like a cute metaphor for a conference keynote — not something with actual predictive power.
I understood this belief for a long time.
Tech culture is built on the premise of the S-curve reset — every collapse in tech history (dot-com, crypto, the 2022 correction) got framed as a "correction" followed by a bigger boom, and mostly, that's been true.
Software really has compounded in ways feudal agriculture never could.
The tools change. The talent pool globalizes. The cycle times compress from centuries to years.
But here's what changed my mind: the people who study collapse professionally — not pop-history YouTubers, but actual quantitative historians like Peter Turchin, who founded the field of cliodynamics specifically to model societal breakdown mathematically — aren't describing a technology problem.
They're describing a **structural** one. Complexity, resource allocation, and elite competition follow the same shape whether the underlying tech is bronze swords or GPUs.
The substrate is irrelevant. The math isn't.
That's the part nobody in tech wants to sit with.
The Evidence: Three Patterns, One Script
Diminishing Returns on Complexity
Joseph Tainter's 1988 book *The Collapse of Complex Societies* remains the single best framework for this, and it's disturbingly simple: societies add complexity (bureaucracy, infrastructure, specialization) to solve problems, and it works — until each additional unit of complexity produces less benefit than the last one did.
Rome kept adding legions, provinces, and administrative layers to manage an empire that was already the most complex political structure the ancient world had seen.
Each addition cost more and returned less, until the empire was spending most of its surplus just maintaining itself.
Now look at AI infrastructure spend.
The major hyperscalers committed to well over $200 billion in AI capital expenditure for 2025 alone, and OpenAI's compute commitments — through deals with Oracle, AMD, Broadcom, and others — have been reported in the trillion-dollar range over the coming years.
Meanwhile, the actual marginal capability gains between frontier model generations have visibly narrowed since 2024.
**We are spending exponentially more to get linearly better.** That is Tainter's curve, drawn in GPUs instead of aqueducts.
Elite Overproduction
Turchin's second pillar is elite overproduction: a society trains far more people for elite status — priests, nobles, officers, credentialed professionals — than it has elite positions to give them.
The surplus doesn't quietly accept downward mobility. It gets angry, forms factions, and destabilizes the institutions that rejected it.
He traced this exact mechanism through the fall of Rome, the French Revolution, and the U.S. Civil War.
Computer science enrollment in the U.S. roughly tripled between 2013 and 2023.
Entry-level software engineering postings, meanwhile, contracted sharply after 2022 as AI coding tools ate the bottom rungs of the ladder those graduates were climbing toward.
**We built a pipeline for a job market that no longer exists at the volume we promised.** That's not a talent shortage story.
That's a textbook elite-overproduction curve, and it's why every "learn to code" bootcamp grad from 2024 is currently furious on LinkedIn.
Currency and Compensation Debasement
Every collapsing empire eventually debases its currency to paper over the gap between obligations and resources — Rome cut the silver content of the denarius from over 90% to under 5% across two centuries.
It works for a while. Then trust collapses faster than the debasement did.
Tech's version is stock-based compensation.
It lets companies pay top talent without touching cash reserves, right up until dilution and stagnant stock prices make the "compensation" worth less than the offer letter promised.
Several major tech employers have quietly reduced new-hire equity grants since 2023 while keeping headline salary numbers flat — the same coin, thinner silver.
The Real Problem Nobody Talks About
Here's where it gets uncomfortable, and where I think most tech commentary chickens out. The real problem isn't AI, and it isn't any single company's spending decisions.
It's that **we built an entire industry culture on the assumption that growth solves complexity instead of creating it.**
Every time tech hits diminishing returns, the reflex isn't to consolidate or simplify — it's to add another layer. Another platform. Another framework.
Another funding round. Another department of people whose entire job is managing the complexity created by the last round of growth.
Tainter's research is explicit about this: societies rarely choose to simplify voluntarily, even when the data clearly shows they should.
They collapse instead, because collapse is actually a rational economizing response — it's the fastest way back to a sustainable complexity-to-return ratio.
That should terrify anyone running a company right now, because it means the "correction" tech has been predicting for two years might not be a dip before the next boom.
It might be the system doing exactly what collapsing systems always do: shedding complexity it can no longer afford, whether leadership authorizes it or not.
What You Should Do Instead
I'm not telling you to short Nvidia or move to a cabin. Here's what the historical pattern actually suggests, practically:
**Stop optimizing for elite-track credentials and start optimizing for irreplaceable, narrow expertise.** Turchin's data shows the people who survive elite overproduction cycles are rarely the most credentialed — they're the ones with skills too specific and too load-bearing to be part of the surplus.
**Treat any organization's complexity as a cost, not a status symbol.** If your team, stack, or org chart has grown mostly to manage problems created by its own last growth spurt, that's Tainter's curve arriving at your desk.
Simplify before you're forced to.
**Diversify your bets across time horizons, not just asset classes.** Empires that survived transitions best — Byzantium outlasted Rome by a thousand years — did it by shrinking deliberately and defensibly rather than expanding until they broke.
Build your career and your company the same way: room to contract without dying.
The Uncomfortable Truth
Every empire in the historical record believed its scale made it exempt from the pattern. Rome believed it. The British believed it.
I'd bet a decent chunk of my own net worth that half the executives greenlighting nine-figure data center builds this year believe it too.
The pattern doesn't care about your market cap.
So here's the question I can't stop asking myself, and I'd genuinely like to know how you'd answer it: if you found out tomorrow that the industry you've built your career on was three years into a Tainter-style diminishing-returns curve, would you actually change what you're doing — or would you, like every empire before us, keep adding complexity and call it progress?
---
