Infinite Compute
Most work has a price ceiling. Two industries don't, and they'll set the price of intelligence for everyone else.
September 4, 2026
The short version:
- Most knowledge work has a price ceiling: what a competent human charges. Extra intelligence earns no premium.
- Investment management and cybersecurity have no ceiling. The budget is set by the prize, and being 1% smarter than the opponent is the whole game.
- That makes them the marginal buyers of frontier AI. The marginal buyer sets the price for everyone.
The price ceiling on knowledge work
The price of a task is set by the cheapest competent person who can do it. Intelligence beyond competent earns nothing, because the task has a finish line. A tax return is filed or it isn't. Nobody can sell a service ticket that's ten times more resolved.
That's most desk work: contract review, slide decks, support tickets, most software maintenance. The price of the task is bounded below by the cost of adequate work and above by what a competent person charges. The whole market sits in that band.
When a model can do the job, the price of the job falls toward the cost of running the model. That's good for customers, and it's a real market on volume. But the premium for extra intelligence stays at zero.
Two exceptions
The logic runs in reverse wherever a contest has a scoreboard. An NBA team will pay to raise its three-point shooting by one percent. A Formula 1 team will pay to cut a tenth of a second from every lap. Those prizes are real but small. Two markets are big enough to move the price of intelligence, and in each the only limit on the spend is the size of the prize.
1. Investment management
An investor's profit comes from being right about something the market has priced wrong, before the market catches up. The industry calls that edge alpha. Whoever finds it first captures it. Whoever finds it second gets nothing.
That holds for every kind of edge, from a millisecond signal to a ten-year thesis on a private company. The question a fund asks is “what is the edge worth,” and the cost of an analyst falls out of the picture entirely. If an edge is worth $40M, spending $39M in compute to find it is a fine trade. The budget is set by the size of the edge, and the edge can be very large.
The quant end of the business shows it most plainly. Quant firms have been the quiet whales of the hardware market for decades. They bought FPGAs long before anyone else cared. One firm ran a private fiber line through the Allegheny Mountains to shave three milliseconds off the Chicago-to-New York route. Every one of those purchases was a ticket to a race, and the purse set the price of the ticket.
2. Cybersecurity
A working zero-day against a major operating system sells for millions. That number reflects what's behind the door: every device running that software, until a patch ships. The hours it took to write are irrelevant to the price.
Defense mirrors it. A bank deciding how much to spend finding a hole before an attacker does is pricing against the cost of the breach, and for a large institution a single breach runs into the billions once you count losses, fines, and the customers who leave. Every large company makes the same calculation, which is why defense alone is a market of hundreds of billions of dollars a year.
What the two have in common
The shape is the same in both.
- They're search problems. You're hunting for something that exists whether or not you find it, and that other well-funded people are hunting for too.
- They're adversarial. My edge is someone else's loss. My exploit is someone else's breach. Value gets captured, and the winner keeps nearly all of it.
- The marginal unit of intelligence is worth a disproportionate amount. In normal work, being 10% smarter than the job requires is waste. Here, being 1% smarter than your opponent is the whole game. The last increment of capability is worth more than everything before it.
Put those together and you get a demand curve that looks like nothing else in the economy. Most industries want intelligence the way they want office furniture: enough of it, at a fair price. These two want it the way a person underwater wants air.
Why this matters for compute
We're in the middle of the largest buildout of computing power in history, and the open question is who uses it all. The usual answer is copilots for lawyers and agents for call centers. Those are real, but they're price-ceiling markets. They'll adopt AI, the cost of the work will drop, and the compute they consume will be bounded by how cheap the task becomes.
The price of frontier intelligence gets set by whoever will pay the most for the next unit of it. Only two buyers are structurally willing to pay whatever it takes: the fund chasing an edge and the team chasing, or defending against, the exploit. The best model in the world will find its most valuable customers there, because those are the only customers who need it to beat the other guy's model, and who will pay for the difference.
There's a darker corollary. In a normal market, spending stops when the job is done. In these two, spending accelerates when the opponent spends. A rival fund's better model erodes your edge, so you buy more compute. An attacker's autonomous bug-finder means every defender needs a faster one. Each side's intelligence is valued against the other's, and the budget feeds on itself.
A caveat: “infinite” is a figure of speech. An edge depletes as more compute chases it. Exploits die the day they're disclosed. At any moment the prize is finite. But new markets open, new code ships every day with new bugs in it, and the prize grows back. Finite in any given quarter and unbounded over time.
How I see it
If we want to push the frontier, build for the people who will pay the most to push it.
Nearly everywhere, intelligence is becoming a commodity, and commodities get cheap. In two places it's a weapon in a duel, and people in duels don't comparison shop. Investment management and cybersecurity will pay all the way up to the value of the prize for one more increment of thinking. That makes them the marginal buyers of the most powerful technology ever built, and the marginal buyer sets the price for everyone.