F2F #62: Narco AI

In late 2016, my good friend Josh and I decided to go to Colombia for a couple of weeks of remote work. During the summer of 2016, the peace treaty in Colombia, officially called the Final Peace Agreement between the Government of Colombia and the FARC-EP, was announced. It was, officially, a safer country to visit.

Last week, I read an article claiming that AI companies are burning their hardware faster than expected, which might cause a shortage in GPUs in the market, which could, in turn, slow down progress and force big projects to cancel. Ultimately, this means that investors won't see returns on their investment.

As I was reading the article, I immediately remembered my trip to Colombia, and how former drug smugglers changed the motor without stopping the boat when they do tourist trips nowadays.

Bear with me. I'll make it worthwhile.

Changing the motor without stopping the boat

As we were on a boat trip to Isla Grande, the motor suddenly burst into flames. In the middle of the sea, with an impassive tranquillity, the boat captain and his crew proceeded to put out the fire and replace the motor on-the-fly as if it was something they did ten times a day.

Maybe not ten, but not even 30 minutes later it happened again. Same thing. Rinse and repeat. All of them super quiet and tranquil, as opposed to me and the rest of the tourists, who were scared shitless every time that happened.

Turns out that some ex-smugglers from the Colombian narco era started running tourist boat trips after "retiring". They had the boats, they had the expertise, they knew the routes inside out, and they were forced out of business by the peace treaty and impending downfall of the narco industry, so they had to change business. In the startup world, they'd call it a "pivot" 🤨

Same sea, very different cargo. But in the end, it boils down to this idea: when they were being chased by the police or the military, the only way to prevent getting caught was to avoid stopping at all costs, so they forced the engines to run faster than they were engineered to do, at risk of burning them. The downside was way worse than the upside. Getting caught meant game over. Making it to port meant just another day in business.

Nowadays, they don't have to do it. They can take more time to change the motor because they know they can do it and there's no stress.

Or is there? tourists get scared when the boat suddenly stops at sea, with the boat being rocked by the waves, splashing water all over with no clear indications in sight. Also, the more trips you can complete in a day, the more money you make, so there are two clear incentives to hustle hard. That, and old habits don't die.

Those boats weren’t built for that kind of nonstop abuse. If you run an engine long enough at high RPM without breaks, it overheats and burns out. Most sane people would take that as a signal: stop, fix the boat, adjust the business.

Today’s AI market feels a lot like that boat.

The boat is AI, the motor is capital, and nobody dares to stop

Right now, AI isn’t just hot, it's seething hot. Valuations, rounds, GPU orders, headcount, hype cycles… everything is dialed to 11.

On paper, this is about "building the future" and "accelerating AGI". In practice, it often feels like what’s really happening is this:

• The boat is the AI narrative: "We’re on the inevitable path to AGI".
• The captain is Sam Fucking Altman selling us the dream trip to make believe and changing motors without us noticing.
• The motor is capital: billions in VC, corporate budgets, sovereign funds.
• The tourists are everyone enjoying the ride: founders, employees, vendors, early adopters.
• The coast guard is reality: unit economics, regulation, energy costs, physical limits, user fatigue.

The boat can’t stop. My feeling with nowadays state-of-the-art mass-consumer AI is that they're firing on all cylinders, pedal to the metal, because they can't afford to stop. It's an inevitable race to the bottom.

By pushing their engines (GPUs and the like), they're burning them way ahead of their supposed life expectancy, much like the smuggler boats' engines. This will increase the cost of hardware as the supply goes down but demand goes up, brute-forcing smaller players out of the game because they can't afford to keep up with the race, until only two or three players remain. Lo and behold, another oligopoly will have been created right in front of our faces, faster than ever before. Faster than regulation (which isn't a lot to say, to be honest).

Instead of slowing down, the market keeps changing the motor in motion. When one narrative overheats (chatbots, copilots, foundation models, open-source vs closed, MCPs...) another fresh engine appears and is bolted on.

From the outside, it can look irrational that so much money keeps pouring into AI when margins are often terrible, differentiation is thin and moats are mostly "we raised more than you".

The feeling is that this can't stop. If we stop, we lose as society: progress will end, we will go back to stone age, to child mortality, incurable cancers and apocalyptic doom. The downside of not doing AI feels way too expensive.

If you think like those ex-smugglers, it starts to make sense.

1. Stopping looks more dangerous than burning another motor. In smuggling, stopping means inspection or jail. In markets, stopping invites a price correction. If one big player openly pauses or pivots away, it signals: "Maybe this isn’t working out as expected". Nobody wants to be first to blink and/or lose authority.
2. Everyone is optimising for optionality, not present-day sanity. If AGI is even plausible within our lifetime, not being heavily invested looks like a catastrophic miss. So allocators think: "If we keep funding this, and AGI happens, we’re in the boat. If we stop, and AGI happens anyway, we’re watching from the shore". Who remembers NFTs anyways?
3. Narratives compound faster than fundamentals. It’s easier to raise a new fund or a mega-round on "We’re closer to AGI than people think" than on "We have a solid, boring, profitable AI-adjacent business". So markets reward speed, not stability.

But what about the upside? Will our destination justify going faster than we should because the incentives of the former smugglers are selling an adulterated narrative? Wouldn't we reach shore anyways just going a bit slower, without burning boats, draining rivers and spending more power than entire countries generate?

Here’s the wild part: the behaviour might be irrational at the micro level but effective at the macro level.

Constant capital injections mean:

• More research labs, bigger teams, more experiments, more failed models.
• Massive hardware built that otherwise would’ve taken decades.
• Talent reallocated from "normal" software to deeply technical work.
• A cultural acceptance that "of course we’re building AGI; that’s just what this era does". Crossing the chasm vibes.
• Life-changing tangible outcomes like cheaper and better drug discovery, productivity gains, more efficient robots, etc.

The question is not whether this is efficient. It’s whether this is accelerating the inevitable. And from where I’m sitting, yes: this behavior likely pulls timelines forward, even if it destroys a lot of engines (and companies) along the way.

AI is going through its narco-tourism phase right now. Same sea, different cargo. The boats won’t slow down voluntarily: there’s too much at stake. Capitals will keep swapping engines mid-ride. And the sheer brutality of that process will almost certainly drag AGI closer on the calendar.

New engine. Same boat. Still moving fast. But going where? And what will happen if we burn all the motors before we reach the shore?