AI as Infrastructure — Lessons from Electricity, Water, and Telecom

1/ Every time a powerful technology becomes infrastructure, the same fight happens: Who controls it? Who gets access? Who profits? AI is replaying a pattern we've seen at least three times before. 🧵

2/ ELECTRICITY, 1880s: Edison wanted private power grids for rich neighborhoods. Insull built centralized utilities. The fight took 50 years. Result: regulated monopolies with universal service obligations. Access became a right.

3/ Before regulation, electric companies could charge whatever they wanted and serve only profitable areas. Rural America got electricity decades late — until the Rural Electrification Act of 1936 forced the issue.

4/ WATER, 1800s-1900s: Cholera epidemics forced cities to treat water as public infrastructure. Private water companies had served the wealthy; public systems served everyone. The pattern: crisis, then public demand, then universal access.

5/ TELECOM, 1900s: AT&T's monopoly was tolerated because of "universal service" — the deal was: you get the monopoly, everyone gets a phone line. When that bargain broke down, we got the 1984 breakup.

6/ The lesson from all three: technologies that become essential don't stay unregulated. The question is never IF governance arrives, but WHEN — and whether it serves incumbents or the public.

7/ Langdon Winner asked: "Do artifacts have politics?" Yes. Robert Moses built bridges too low for buses. Edison designed DC power to lock in customers. Infrastructure choices are political choices, made durable.

8/ Now look at AI: A few companies control the frontier models. API pricing determines who can build what. Training data is enclosed. Compute is concentrated. This is the "Edison phase" — private control, selective access.

9/ Ursula Franklin distinguished "holistic" tech (augments human judgment) from "prescriptive" tech (replaces it with machine outputs). Most deployed AI is prescriptive: follow the algorithm. That's a political choice.

10/ The infrastructure metaphor isn't perfect. AI isn't a natural monopoly like water pipes. But the pattern holds: essential capability + concentrated control = redistribution crisis. Every time.

11/ History says this resolves one of three ways: regulated private monopoly (electricity), public provision (water), or managed competition (telecom). Each has tradeoffs. None is inevitable.

12/ The window for shaping AI's infrastructure future is open now. Once the pipes are laid, they're hard to re-route. Ask your policymakers: what's the universal service obligation for AI? /end

---

LinkedIn version:

Every time a powerful technology becomes infrastructure, the same fight plays out: Who controls it? Who gets access? Who profits?

We've seen this at least three times. Electricity in the 1880s: Edison wanted private grids for wealthy neighborhoods. It took 50 years of political struggle before the Rural Electrification Act brought power to all Americans. Water in the 1800s: cholera epidemics forced cities to make clean water a public service, not a private luxury. Telecom in the 1900s: AT&T got its monopoly in exchange for universal service — and lost it when the bargain broke down.

The pattern is consistent. Essential technologies don't stay unregulated. The question is never IF governance arrives, but when — and whether it serves incumbents or the public.

Now look at AI. A handful of companies control frontier models. API pricing determines who can build. Training data is enclosed. Compute is concentrated. As Langdon Winner taught us, infrastructure choices are political choices made durable. We're in the "Edison phase" of AI — private control, selective access.

History says this resolves through regulated monopoly (electricity), public provision (water), or managed competition (telecom). Each path has tradeoffs. None is inevitable.

The window for shaping AI's infrastructure future is open right now. Once the pipes are laid, they're extraordinarily hard to re-route.