If the Lights Go Out, Your AI Goes With Them

What If the Lights Went Out?

There's a specific kind of film that surfaces every few years about civilization without electricity. American Blackout (2013) followed ordinary people through a national power grid collapse. E.M.P.: 333 Days (2018) dramatized what a full electromagnetic pulse event would mean for the United States. Grid Down, Power Up (2022) made the documentary case that this isn't science fiction.

These films are exploring real physics.

A high-altitude nuclear detonation generates an electromagnetic pulse — a burst of energy that induces current surges in electronic systems across enormous areas. A powerful geomagnetic storm from the sun can do the same thing. It's happened at smaller scales. The 1989 Quebec blackout left millions without power for nine hours after a solar storm knocked out transformers across the province. Engineers who design military systems take it seriously. The U.S. congressional EMP Commission has published reports on it.

The scenario is extreme. But it forces a question most people skip: what does total electrical failure tell us about our relationship with any technology that requires it?

For AI, the answer is uncomfortable.

The Dependency Chain Nobody Mentions

When you type a prompt into ChatGPT or Claude, here's what's happening:

• Your request travels over a network to a data center
• Servers and GPUs inside that data center process the inference
• Those GPUs require enormous, sustained power — AI data centers rank among the most electricity-intensive facilities on earth
• That power comes from the grid: generation plants, transformers, distribution infrastructure

Every layer in that chain depends on the layer below it. You cannot skip a step.

An EMP event doesn't just shut down your laptop. It disables the grid control systems. It fries the transformers — the large, custom-manufactured kind that take months or years to replace. It takes down the data centers. Every AI tool you use: gone. Instantly.

Your notebook and pencil still work.

Here's the line that should stop you: GenAI isn't a technology revolution. It's an electricity-dependent application of computing. Electricity is the revolution. AI is what we're running on top of it.

This is a clarity check on what we've built and what it requires.

Electricity Was Already the Greatest Technology

Electricity has been powering human civilization for over a century. We've spent most of that time ignoring it.

It runs hospitals and surgical equipment. It keeps food cold enough to not kill us. It pumps clean water. It heats and cools the spaces where people live and work. When the power fails at large scale (a real regional outage after a hurricane or ice storm, not a thought experiment), people die. Not because their AI tools stopped working. Because they lost heating in January or power to medical devices.

Electricity also enabled every major communication and information technology before AI. Radio. Telephone. Television. The internet. Each of these looked like a revolution when it emerged. Each was electricity finding a new application.

We celebrate the applications and forget the operating system.

There's a version of this dynamic that plays out every time a powerful new tool appears: the tool becomes the story, and the infrastructure that makes it possible becomes invisible. We obsessed over the iPhone and ignored the cell towers. We obsess over AI and ignore the power grid that every data center in the world depends on.

When the power goes out, you don't miss your AI tools first. You miss your lights.

The PM Lesson: Know Your Dependency Chain

Here's where this stops being a physics thought experiment and starts being a professional skill.

If you're thinking strategically about your work, you know every tool rests on infrastructure. Every capability has a prerequisite. The EMP scenario is a dramatic illustration of something that happens at smaller scale in every organization, on every project: tools disappear, systems fail, vendors go under, APIs get deprecated, budgets get cut.

Task-tracking PMs adopt tools without thinking about what those tools depend on. They build workflows around capabilities they don't control and treat availability as a given.

Strategic PMs map dependencies before they become risks.

Your AI workflow depends on more than a subscription:

• Reliable electricity and internet connectivity
• The vendor staying in business and keeping the product available
• Your organization's security and data governance policies
• The API remaining accessible at the pricing you've built around
• Your own ability to evaluate AI output critically

A PM using Copilot for meeting notes should know: what happens if the subscription lapses, the security team blocks it, or the API terms change? That's not paranoia. That's dependency mapping.

The question every PM should be able to answer about any AI tool they rely on: "If this tool disappeared tomorrow, what's my fallback? And what foundational skill does it amplify that I need to keep sharp regardless?"

That second part is the harder question. And it's the one most people skip.

The PM who uses AI as a thinking partner retains the underlying skill. The PM who uses AI to avoid thinking has built a single point of failure into their own cognition. The tool becomes the capability, not the amplifier of it. When the tool goes away (for any reason, not just an EMP), they're left with nothing.

This is the same mistake as outsourcing your judgment to any process, methodology, or system. The tool was never supposed to be the thinking. It was supposed to make the thinking better.

The Lights Can Go Out. Your Judgment Can't.

Return to the scenario. The grid fails. Every AI tool disappears instantly.

But strategic thinking doesn't run on electricity. Stakeholder analysis doesn't require a data center. Clarity about what a project needs to achieve: clear outcomes, documented assumptions, meaningful success measures. None of that lives in a server farm.

The PMs who would still function in that world aren't the ones with the best AI workflows. They're the ones who used AI to get better at thinking, not faster at typing. They developed the underlying skill and used AI to sharpen it. Their judgment survived the power outage because it was never outsourced.

Electricity existed before AI. It will be here after whatever comes next. It is the foundation that everything else is built on. It deserves the same kind of attention we give the things running on top of it.

The lights can go out. Your judgment can't.

What's one thing in your AI workflow that you've never mapped the dependencies for?