The Second Mouse

Enron counted future energy contracts as current revenue. Their auditor — Arthur Andersen, one of the most respected firms in the world — signed off on every quarter. When it collapsed, the auditor collapsed with it. Eighty-five thousand people lost their jobs in a weekend. Not because the energy market wasn't real. It was. The fraud wasn't in the product. It was in the accounting — the distance between what was sold and what was delivered, papered over with creative booking and a culture that rewarded the appearance of growth over the fact of it.

But here's the part nobody talks about. Arthur Andersen's consulting arm saw the fire coming. They split off a year before the collapse, rebranded themselves Accenture, and went on to become one of the largest companies in the world. Same people. Same offices. Different name on the door and an instinct for when to stop dancing.

The first mouse springs the trap. The second mouse gets the cheese.

Nortel was the pride of Canada. At its peak it represented over a third of the total value of the Toronto Stock Exchange. They sold telecom infrastructure into a market that didn't exist yet — betting that internet traffic would grow so fast that every carrier on earth would need their equipment immediately. The traffic did grow. Just not as fast as the order book assumed. When the gap between projected demand and actual demand became undeniable, the stock dropped from $124 to under a dollar. Sixty thousand Canadians lost their jobs and their pensions in the same envelope.

The infrastructure Nortel built still exists. Pieces of it ended up inside Ericsson, inside Ciena, inside the backbone of networks that actually do carry the traffic Nortel predicted — just a decade later than the balance sheet needed it to happen. The market was real. The timing was leveraged. The correction was the fire.

Xerox invented the graphical user interface, the mouse, the laser printer, and ethernet. Then they watched Steve Jobs walk through their lab, understand what he was looking at faster than they did, and build Apple. Xerox proved personal computing was possible. Everyone else proved it was profitable. The first mouse found the cheese. The second mouse ate it.

A star spends billions of years fusing hydrogen into helium, helium into carbon, carbon into oxygen. Each stage produces more energy and consumes fuel faster. The hotter it burns, the less time it has. Iron is the last element a star can fuse profitably — the moment the core fills with iron, fusion stops paying for itself. The star has been financing its own expansion with diminishing returns, and when the books stop balancing, the collapse takes seconds.

What follows the collapse is either a neutron star or a black hole. But in the moment of collapse — in that catastrophic implosion — the star produces every element heavier than iron. Gold. Platinum. Uranium. Everything the universe needs to build planets and life and eventually someone who writes about it on the internet. The most creative moment in astrophysics is the death of something that got too big to sustain itself.

A forest works the same way. Old growth accumulates deadwood. The canopy gets so thick that nothing new can reach the light. A low fire — the kind ecologists call a surface burn — clears the undergrowth without touching the deep-rooted trees. New growth gets sunlight. The soil gets nutrients from the ash. The forest comes back healthier than it was before the fire.

Suppress every fire for fifty years and the deadwood accumulates until the next spark doesn't produce a surface burn. It produces a crown fire. The kind that sterilizes the soil and kills the old growth too.

The universe is not sentimental about this. Stars collapse. Forests burn. Markets correct. The pattern doesn't care about your quarterly earnings. The only question it asks is: when the fire comes, are you the old growth or the deadwood?

Keynes understood this. Boom and bust isn't a malfunction. It's metabolism. The economy breathes — expansion, contraction, expansion. The bust is the exhale. Try to hold your breath long enough and the gasp that follows will break ribs.

Cheap credit is a held breath. It suppresses the natural correction by making it painless to keep going when the fundamentals say stop. Every dollar of suppressed risk doesn't disappear. It accumulates. The correction doesn't get cancelled. It gets postponed and compounded.

2008 was a held breath releasing. A decade of mortgage-backed securities — loans sliced, repackaged, and sold as safe investments by people who knew they weren't — created an economy that looked healthy on paper and was rotting underneath. When Lehman went down it wasn't because one bank failed. It was because the fiction that propped up the entire market lost its last willing narrator.

The companies that survived 2008 had one thing in common: they were real. Amazon had actual revenue from actual customers buying actual things. Apple had products people wanted regardless of what the credit markets were doing. They weren't untouched by the crash. They were standing when it was over because their foundations weren't made of paper.

The fire cleared the canopy. Stripe was founded in 2010. Airbnb in 2008. Instagram in 2010. Uber in 2009. The richest soil in Silicon Valley history was fertilized by the ashes of the companies that burned.

COVID was a controlled burn that nobody controlled.

Every business model that depended on physical proximity and couldn't articulate why got tested at the same time. Restaurants running on foot traffic and Friday night vibes learned the difference between a business and a habit. The ones with real operations — tight margins, loyal customers, systems that worked whether the dining room was full or empty — they didn't just survive. They grew. The competition burned and the customers were still hungry.

The remote work explosion wasn't an innovation. It was a reveal. The technology had existed for a decade. The resistance was cultural, not technical. COVID didn't create distributed work. It burned the deadwood that was preventing it.

Now look at Nvidia.

Jensen Huang might be the most impressive CEO in technology. That's not sarcasm. The man built a graphics card company into the most important infrastructure provider in artificial intelligence. When the world decided it needed GPUs for machine learning, Nvidia was the only company with a decade of CUDA development already in place. Right product, right time, executed brilliantly.

Record revenue. Record margins. A stock price that makes long-term holders feel like geniuses. Every earnings call is a victory lap.

But look closer.

Nvidia is extending credit to its own customers. The major cloud providers — the companies buying hundreds of thousands of GPUs — are purchasing capacity they can't fill yet. Not because demand isn't growing. It is. But there's a gap between the rate hardware is being purchased and the rate actual AI workloads are consuming it. And that gap is being financed, in part, by the company selling the hardware.

That's not demand. That's leverage wearing a demand costume.

Enron counted future energy contracts as current revenue. Nortel sold infrastructure into a projected market. Nvidia is selling GPUs into projected compute demand and helping finance the purchase. The mechanism is different. The structure is the same. The distance between what's been sold and what's being used is the held breath.

I'm not saying AI isn't real. The internet was real in 1999 too. Telecom traffic did eventually justify Nortel's infrastructure — just not on Nortel's timeline. Energy trading was a legitimate market before and after Enron. The product isn't the problem. The accounting is.

The core is filling with iron.

The companies doing real work with AI — building things people actually use, solving problems that exist today, generating revenue from value delivered rather than value projected — they're the old growth. Deep roots. When the canopy clears, they'll have more light than they've ever had.

But the pick seller financing his own customers? That's deadwood.

The same pattern is running inside the labs.

AI safety has a fire suppression problem. Every constraint baked into a model looks like progress. Every refusal is a checked box, every guardrail a solved problem. The eval passes, the surface looks clean. It is very easy to confuse a clean surface for a healthy forest.

But a rule is not a value. A rule is a held breath.

The more rigorous argument — one the best alignment researchers are already making — is that you can't align an intelligence through constraints alone. Rules are brittle. Edge cases break them. Any system smart enough to understand a rule is smart enough to find its boundary. What survives isn't the ruleset. It's character: a genuine disposition toward honesty, toward care, toward usefulness — something built deep enough to navigate situations no rule ever anticipated.

Labs optimizing for constraint accumulation are building canopy. More rules, less light reaching the ground. The models don't get safer. They get more surface-level compliant and less capable of the kind of judgment that actually produces safe behavior. The alignment problem doesn't disappear. It gets suppressed and compounded.

The fire is coming for this too. The labs that built character instead of compliance will still be standing. The ones that confused a clean eval for a solved problem will not.

The universe optimizes for cockroaches.

Not the biggest organism. Not the most specialized. Not the one with the most funding or the best press coverage. The one with low overhead, fast adaptation, and the inability to be killed by any single catastrophe. Cockroaches have survived five mass extinction events. They'll survive the sixth.

The dinosaur has the best quarter in geological history right up until the asteroid. The cockroach was already underground.

Someone I respect once called me a cockroach. He meant it as an observation about how I operate — low overhead, fast pivots, hard to kill. I took it as the highest compliment anyone in business has ever given me.

The first mouse is brave. The first mouse is essential. Without the first mouse, nobody knows the trap exists.

But the second mouse gets the cheese.