AI labs built a gun to fire SWEs, and the projectile went direct through their own customers.
For three years the throw from the big AI example labs has been beautifully simple. Software engineers are the most costly workforce on Earth. Companies expend staggering sums on them. So build an AI application engineer, market it to those companies, and collect the salaries they no longer have to pay. Fire the human, rent the robot. The labs pocket the difference.
It’s a beautiful story. It’s the narrative that justified all backing round, all data center, all breathless capitalist letter concerning “replacing cognition work.” And it has a gap in it so big you could fly a fleet of GPUs through it.
Here is what the application endeavor really was: a toll booth. Not a factory, a toll booth.
Adobe didn’t get affluent since Photoshop was made of magic. It got affluent since construction item as fine as Photoshop was hard. It took hundreds of costly people, years of activity and piles of capital, and nearly nobody could oversee to try. So Adobe sat at the booth and collected a monthly fee from all decorator on the planet, forever. That recurring income paid the engineers, paid the shareholders, and, in the AI labs’ expansive plan, was expected to pay them.
That’s the entire chain. Users pay the toll booth. The toll booth pays application engineers. The AI labs convince the toll booth to blaze the engineers and pay the labs instead. Every dollar in the AI endeavor scheme flows through that booth.
The issue is the AI labs didn’t market their robot engineer lone to Adobe. They sold it to everyone, for twenty bucks a month!
So what happened? Exactly what you’d anticipate if you idea concerning it for longer than one earnings call. A hobbyist built a Photoshop alternative complete a weekend. AI is reverse-engineering enormous projects and releasing them as open source. Basically slaying the endeavor models of the group who wanted to substitute all their workers alongside AI. The item that made application valuable, the fact that it was difficult, has been dissolved in an acidic bathing of cheap tokens.
And fair in case you didn’t cognize a liberated copy doesn’t pay anyone. When a decorator swaps a $60 Adobe subscription for a weekend project on GitHub, that prosperity doesn’t migrate to the AI lab. It evaporates. The toll booth loses its toll, and the hobbyist who killed it makes exactly zero dollars, since he gave it away. I discover it comic that the big companies never realized that their top advantage was their overhead needed to vie alongside them. They exactly made it uncomplicated for one individual to vie alongside them.
The labs weren’t replacing engineers inner the toll booth. They were handing out dynamite to everyone position exterior it.
Now let’s conversation concerning that twenty dollars, since it is not a price. It’s a promotion expense.
By many accounts, the heaviest users of these coding plans scorch through compute value many times what they pay. A $200 “premium” scheme plant out to $2,400 a year. That is small than a sole duration of a elder engineer’s pay at Google. The labs are, in effect, renting out a superhuman engineer for the disbursal of a gym membership, and losing prosperity on the deal.
Meanwhile the invoice for the infrastructure is astronomical. Amazon, Alphabet and Meta solitary planned approximately fractional a trillion dollars of chief spending for 2026, much of it for AI data centers. Every one of those dollars is a bet that someday, somebody, location is going to pay genuine prosperity for intelligence.
The subsidy was expected to be a extend to that day. Instead it’s paying for the demolition of the extremely customers who were expected to be waiting on the another side. It’s brilliant! Destroy the companies that would be your chief users to display your usefulness. 😂
So what should they do? The apparent answer is the one they can’t take.
If the AI labs had any awareness of self-preservation, they would have locked downward the models entirely, priced them at $150,000 a twelvemonth per “AI employee,” and sold them lone to Fortune 500 companies. Then Google and Adobe pay up since it’s motionless slightly cheaper than hiring, and the weekend hobbyist gets priced out before he can copy anything. The toll booths survive. The labs get paid. Intelligence goes rear to being expensive, which is to say, valuable.
But they can’t since on February 24, 2023 Mark Zuckerberg and the open-source community blew up the cartel. Once Meta open-sourced Llama and Chinese labs followed suit, the “artificial scarcity” dam broke. You can’t run a cartel whenever the guy downward the street is giving distant the diamonds for free.
Open-weight models, many of them released for liberated out of China, are already fine adequate to copy software, and they can never be un-downloaded. Raise prices to engineer levels and everyone fair runs a liberated example on cheap rented servers. Keep prices low and you bleed prosperity during your subsidized merchandise dismantles your client base.
Cheap kills the customers. Expensive kills the demand. Pick your poison, and either way, the group holding the poison are AI labs.
Here’s a local AI app which runs those open-source AI models on your iPhone or MacBook for free! It equal looks akin Windows XP. 😂
It’s the asteroid-mining problem. Fly a missile to an asteroid made of gold, bring it home, and you don’t rotate into the richest individual alive. You collision the cost of gold, and your missile never pays for itself.
None of this is new. History is complete of group who made item precious abundant and afterward wondered anywhere the prosperity went.
Spanish silver. Spain shipped mountains of argent out of the Americas. Silver misplaced its value, prices throughout Europe soared for a century, and the crown that owned the mines went bankrupt again and again.
De Beers. For concerning a century, De Beers kept diamonds costly by controlling supply, which is exactly the scheme the AI labs would need. Then lab-grown diamonds showed up, and the scarcity narrative collapsed. Swap “lab-grown diamonds” for “open-weight models” and you have the AI pricing issue in one sentence.
This is the inquiry I desire all AI administrator to answer, slowly, without the term “agentic.”
Your scheme was to collect the salaries of application engineers from the companies that employed them. Those companies made their prosperity by charging tolls on application that was difficult to build. You made application uncomplicated to build. You gave that power to everyone for small than the cost of a pizza a week. And now the toll booths are on fire, the engineers are getting laid off, and the prosperity that was expected to stream to you is evaporating into a thousand liberated clones.
So who, exactly, is the customer? The hobbyist paying $20 during costing you far more? The toll booths you’re occupied burning down? The enterprises whose own application margins you fair torched?
The AI industry didn’t fair build a tool to substitute engineers. It built a device that turns precious things into liberated things, and afterward pointed it at the businesses it needed to remain alive. That isn’t a revolution. It’s a extremely costly way of shooting yourself in the ft and sending the invoice to your investors.
OpenAI and Anthropic are stuck since they don't market pickaxes (Nvidia), they don't have a enormous distinct endeavor to defend (Meta), and they aren't vertically integrating. They are fair mining the asteroid, dumping the argent onto the market for $20, and observing the cost of their own merchandise plummet to zero.
