The Cheaper AI Gets, the More Dangerous It Is for Companies That Only Rent Generic Intelligence
- Investor Chamath Palihapitiya published a long essay on X titled "The Great Descent," arguing that AI intelligence is retracing the same cost curve smartphones once rode from luxury item to mass commodity
- The first modern smartphone launched in 2007 at around $500, a luxury item for a wealthy few; fifteen years later, phones with more computing power sell for under $50 at market stalls in developing countries, and global ownership has topped 6 billion units
- The author argues AI intelligence's descent is faster than phones' because it stacks two curves at once: hardware itself is getting cheaper, and the compute required for the same capability shrinks year over year
- The internet made facts and knowledge free but never solved the scarcity of judgment; the author argues that this time, AI is delivering judgment itself — a bigger event than the internet
- The author raises a paradox: if every company simply rents the same generic AI, it erases its own competitive edge instead of building one; the real winners will be the companies that encode their proprietary experience into systems of their own
Fifteen Years Ago, That Prophecy Came Entirely True
On July 2, 2026, investor Chamath Palihapitiya published a long essay on X titled "The Great Descent," arguing that AI intelligence is sliding down the very same cost curve smartphones once rode from luxury to ubiquity — only faster this time.
Fifteen years ago, Marc Andreessen predicted software would eat one industry after another — a call that's played out almost entirely. The author pushes the logic one step further: once software starts to think, expert judgment will slide the same way phones did, from a luxury for the few into an everyday commodity for everyone.
Why it's worth reading: The author lays out a concrete mechanism. AI's falling price is driven by two curves at once — the hardware cost curve plus the model-efficiency curve — whereas phones back then only had the hardware curve. That's the core basis for his claim that "AI is getting cheaper faster than phones did."
Andreessen's essay was good in a way that's rare: a prophecy that came so completely true its argument has become the air we breathe — every industry he named got eaten by software, and most of the ones he didn't name didn't escape either. But a piece that good doesn't just close a topic, it opens the next one. Back then software couldn't think yet, so there was a chapter he never got to tell in full: what happens once software starts to think.
The Old Problem: Information Went Free, Judgment Never Did
The internet drove the cost of distributing knowledge to zero. Everything humanity knows, once locked away in libraries and gated behind gatekeepers, is now a search box away. But that revolution only got halfway there: it gave you facts, not judgment. You can look up every symptom of a disease and still not know whether you actually have it; you can read every precedent on a legal question and still not know what to do.
The reason is that knowledge can be copied, but professional judgment can't. Photocopying a book costs almost nothing, but a doctor, a lawyer, a senior engineer, or a seasoned underwriter takes decades to develop and can't be cloned. The author says the scarcity of professional judgment is the oldest, most fundamental bottleneck in the economy — one that's been stuck there since human economies began.
Photocopying a book costs almost nothing; everything humanity knows now fits in a search box — the internet era already made it available to everyone.
Doctors, lawyers, senior engineers take decades to develop, and it walks out the door when they retire — it's never been fully written down or copied.
And now, the author says, that bottleneck is being broken.
How the Phone Went From a $500 Luxury to 6 Billion Units
Start with the phone in your pocket. When the first modern smartphone launched in 2007, it cost around $500 — a luxury for affluent consumers in wealthy countries, with fewer than two million owners worldwide. By any measure, it was a premium product priced for the few.
What happened next isn't visible if you look at the high-end flagship — that's still expensive today, still sitting at the top of the market. You have to look at the category as a whole: within fifteen years, a smartphone with more computing power than that original device could be bought for under $50 at market stalls in any developing country. There are now more than 6 billion smartphones on Earth. Something born as a possession of the rich became, in fifteen years, the most widely distributed and most powerful tool in human history.
Same Mold: Every Disruptive Technology Takes These Three Steps
It's worth spelling out these steps, because they're the playbook every truly disruptive technology follows, whether or not anyone ever writes it down.
where value is dense enough
prices keep falling
the question becomes what to do with it
The phone took exactly these three steps. Lay them side by side and it's clear this isn't a coincidence — it's the same mold:
| Three Steps | Smartphones | AI Intelligence |
|---|---|---|
| Enter at the top | Launched at $500 in 2007, for affluent consumers only | Deployed first on the highest-value problems with the deepest resources |
| Fund the descent | Component specialization and mass-production scale drove years of price cuts | Hardware gets cheaper while model efficiency improves — both curves fall together |
| Endpoint: ubiquity | Under $50, 6 billion units | Judgment becomes cheap enough to be nearly free, available to everyone |
The author says phones pulled this off. And it matters because intelligence is doing the exact same thing right now, riding the exact same curve — just faster.
Why Intelligence Is Getting Cheaper Faster Than Phones Did: Two Curves Descending Together
The fixed cost of completing a cognitive task — the unit price of "one dose of machine intelligence" — is falling at a rate that makes the phone's descent look sluggish by comparison. Capability that cost a fortune 18 months ago costs pennies on the dollar today, and the author estimates it'll drop another notch in another 18 months (this estimate is the author's own projection). You can already watch this happening in real time: open-source intelligence with performance close to the priciest closed systems, at a fraction of the price.
Phones descended along a single curve — hardware getting cheaper, driven by cheaper chips, memory, power, and the sheer scale of global manufacturing. AI intelligence rides that same hardware curve down, because it also runs on silicon, memory, and power, which are specializing and getting cheaper just like phone components did back then. But it has a second curve stacked on top of the hardware curve: the models themselves keep getting more efficient, needing less compute for the same task every year. Two discounts apply at once, and they compound — one from hardware, one from the intelligence itself. Phones only ever had one.
Phones going downhill is like coasting down a mountain with no brakes. AI going downhill is coasting down with no brakes while also throwing luggage out the window piece by piece, getting lighter the whole way down. Two kinds of acceleration stacked together — that's why it's going faster.
The author says that once both curves run their course, intelligence will become abundant and nearly free — abundant the way information became abundant, abundant the way a networked computer in every pocket became abundant. This isn't wishful thinking about a distant future; it's a straightforward extrapolation of two curves already in motion, and we've already lived through this exact pattern once before.
From Everyone Having Knowledge to Everyone Having Judgment
The internet gave everyone access to knowledge. This time, everyone gets access to professional judgment — and the author says the latter is a far bigger deal.
Facts and knowledge became available to everyone, but the truly scarce thing — judgment — stayed expensive, rationed, and locked inside a few trained minds.
An analyst's reasoning, a clinician's diagnostic instinct, a good lawyer's drafting skill, a senior engineer's design sense — this layer of judgment is starting to become summonable by anyone.
Specialized judgment that once required a trained professional, a credential, a firm, a salary, and a lot of money is, for the first time, something anyone can summon at nearly zero cost. The author stresses this isn't about the facts those professionals know — the internet made facts available long ago — it's about the layer where they apply judgment, a capability that's never been copied before and is now becoming abundant.
Afraid of Losing Your Job? The Author Says the Premise of That Fear Is Wrong
Right on cue, the fear shows up: if machines can supply expert judgment for free, what happens to the experts? If intelligence becomes this abundant, what's left for humans to do? The author says this question deserves a direct answer, not a comforting platitude.
That fear rests on a hidden assumption: that the total amount of work is fixed, so every task a machine takes on is one less for a human. The author says this assumption has been proven wrong every single time, and for good reason. When something valuable becomes extremely cheap, we don't use less of it — we use dramatically more of it, and we invent new uses for it that were unimaginable when it was scarce. Cheap information didn't end knowledge work; it spawned an entire category of new jobs that couldn't have existed when information was expensive and slow. The truly scarce thing was never labor — it's the ability to turn judgment into action, and we're about to have a near-limitless supply of exactly that.
So work won't disappear — it'll move. When expertise is scarce, the bottleneck is getting access to it; when expertise is abundant, the bottleneck becomes what to do with it: which questions to ask, whose judgment to trust, which problems are worth solving, and who's accountable when something goes wrong. The author calls this people moving up the stackStack here means the layers of a system — this refers to humans moving from the layer of "doing the analysis by hand" up to the layer of "deciding what the analysis should serve, and who's accountable.".
who's accountable
The author says none of this makes the transition painless — real people in real jobs will be disrupted and forced to relocate. But in aggregate, over the long run, the direction isn't less human work, it's more — because ambition always expands to fill whatever capacity becomes available. It always has.
The Trap: If Everyone Rents the Same Intelligence, No One Has an Edge
This is where the trap comes in: if everyone rents the same intelligence from the same vendor, no one has any advantage at all.
The best companies never won on generic capability — they won on a way of doing things that's entirely their own, hard-earned, what the author calls their alphaA finance term originally meaning above-average excess return; here it broadly means a company's genuinely distinctive experience and judgment.: a retailer's grip on its own logistics, an insurer's feel for its own risk, a manufacturer's control over its own process. This layer of advantage never used to make it into software, because encoding it into a working system required expensive engineering — so it stayed locked inside veteran employees' heads, walked out the door when they retired, and could never be fully articulated, written down, or passed on. Once intelligence gets cheap, the cost of encoding it collapses, and for the first time companies can build the thing that truly sets them apart into the foundational documents and systems they control — making it replicable, scalable, and able to compound.
But the danger is hiding in exactly this spot. If the only thing you do with cheap intelligence is buy the same generic, off-the-shelf intelligence as your competitors, you haven't built any advantage — you've erased one with your own hands. You've traded the capability that used to make you distinctive for the same commodity everyone else is buying.
The moat keeps getting deeper
The knack for logistics dispatch, the instinct for risk pricing, the feel for process control — once locked inside veteran employees' heads and lost when they retired. Now that encoding costs have collapsed, for the first time it can be written into systems you control, compounding daily.
Indistinguishable from your competitors
You're buying the same commodity as your competitors and using it the same way — trading the capability that made you distinctive for something every company can buy.
The author says the winners of the last era weren't the companies that used software — everyone used software. The winners were the companies that understood the entire game came down to how you used it, how proprietary you made it, how much of your own hard-won advantage you built into it. Everyone is about to relearn this same lesson, with much higher stakes.
The Bottom of the Descent: Expert Judgment Stops Being a Giant's Exclusive
Which brings us back to the descent. The author says the same playbook is running again: intelligence starts at the top, expensive and rationed, deployed first on the highest-stakes problems with the deepest resources, because that's where the value is dense enough to pay for itself. That revenue funds the descent, the cost curve does its work — twice as fast this time — and capability slides all the way down to everyone.
And at the bottom of that curve is something the world has never had: professional-grade capability itself, abundant, nearly free — belonging not just to the big institutions that could once afford an entire army of experts, but also to small companies, solo founders, and anyone with an idea but not the capital to encode it. The same arc that carried the smartphone from a luxury item into 6 billion hands is now replaying itself with intelligence, heading toward the very same destination: ubiquity. The author says it's the biggest opportunity he's ever seen.
The internet let everyone get access to knowledge. This time, everyone gets access to professional judgment — a much bigger thing.Chamath Palihapitiya, "The Great Descent"
Expert Judgment: From a Luxury for the Few → Available to Everyone, Like Phones
Investor Chamath Palihapitiya: AI is crashing the price of expert judgment down the same curve phones rode — twice as fast; but companies that just rent generic AI end up erasing their own advantage. The whole argument, with diagrams, on one page
↓ Read it in one page · includes an animated diagram
Investor Chamath Palihapitiya published a long essay, "The Great Descent," comparing AI's falling price to the smartphone's journey from luxury item to one-in-every-hand. He starts by naming an old problem: the internet revolution only ever got halfway there.
The author says what AI is putting into ordinary people's hands this time isn't more knowledge — it's judgment itself. Judgment that once required hiring a professional, a credential, and a salary can now be summoned at nearly zero cost.
Some worry AI will take their jobs. The author pushes back: work won't disappear, it'll just move up from "doing the analysis by hand" to "deciding what to ask, whom to trust, who's accountable." And he says this price curve is falling faster than the phone's did — why? Look at the diagram below.
Phones only descended along one curve — chips, memory, and power getting cheaper. AI gets that same hardware discount, plus one more: the compute (computing power) needed for the same task shrinks every year. Two discounts apply at once, and they compound.
Just saying "twice as fast" doesn't give you a feel for it. Look at the phone's descent, which has already played out — what it costs, then versus now, to buy computing power at the level of the original iPhone:
found them all…
have it or not?
not judgment.
Fewer than 2M owners worldwide
<$50, and more powerful
judgment's turn?
faster than phones?!
hardware got cheaper.
falling faster, lower!
I can afford it too!
- × Just renting generic AI
- × Identical to your competitors
- × You erased your own edge
proprietary experience into your systems.
Members can copy or download the full Markdown, and let their AI subscribe and ask.
Found this useful? Leave your email — every new explainer, the moment it's out. Unsubscribe anytime.