Fed officials have started blaming the AI boom for amplifying inflation. Goldman Sachs economists ran the numbers and answered: the case is overstated, and the effect is already peaking. The number they put on it is 45 basis points — 0.45%.
The core personal-consumption-expenditures index, the Fed's preferred inflation gauge, printed 3% year-over-year for August. Goldman isolated the AI contribution and found it worth 0.45%. Do the arithmetic: AI explains 15% of the inflation everyone is arguing about. The other 85% is housing, wages, supply chains — none of it touched by AI.
To see why the number is so small, it helps to ask the question the speeches skip: through which channels, exactly, would AI push up prices?
There are three, and none of them is big enough to matter.
The first is capital spending. Data centers, chips, electricity — AI buys all three at a pace that moves those specific markets. The second is labor. AI engineers command salaries that pull up local wages wherever the labs cluster. The third is a wealth effect. AI companies trade at valuations that make their holders spend more freely.
Add all three together, and Goldman gets 45 basis points. The reason is scale. The American economy is roughly thirty trillion dollars. AI's capex, even growing at an alarming rate, is still a rounding error against that denominator — a sector that is less than a single percentage point of GDP cannot manufacture a 3% inflation rate by itself. It can heat the corner of the economy it touches. It cannot heat the economy.
Then the second finding, the one that actually flips the story: the effect is peaking. The first wave of data-center construction — the burst that drove the price spikes in chips and power — has crested. The marginal contribution of AI to inflation is now falling, not rising.
45 basis points. Hold it up against the narrative. The technology being blamed for overheating the entire economy contributes, by the most careful accounting available, less than half a percentage point — and that contribution is shrinking.
Fifteen percent. Measured, quantified, fading. The inflation panic that names AI as its culprit is, in dollar terms, arguing about a rounding error.
So why does the blame keep landing on AI? Because AI is a convenient explanation. It is new, it is loud, it is the story every screen is already showing. A convenient explanation is not the same as a correct one — and the correct one is boring: housing, which is the single largest weight in the index; wages, which lag the labor market; and supply chains still settling after their shock. AI is the headline. It is not the cause.
But there is a second inflation, and it is not in any price index, because no statistician thought to count it.
Open a short-video platform. AI clips, AI animation, AI images, AI voice, AI music — stacking up at an exponential rate. The cost of producing a moving image has collapsed. Not to zero — tokens still cost money — but by orders of magnitude. A piece of film that once required actors, a set, a camera crew, and a post-production team now requires a prompt and a batch of compute.
That is the mechanism. And it matters because of an old rule: rare is precious. Scarcity is what gave content its value. A thing that took a hundred people a month to make was, by definition, scarce — there could only ever be so much of it. AI did not just lower that cost. It removed the constraint that made scarcity possible in the first place.
When a good-looking video is everywhere, being good-looking is no longer scarce. So the price of "good-looking" falls — not the price in dollars, but the price in the only currency content ever really traded in: attention.
This is not prices rising. This is content deflating.
And the deflation is not a free-fall. Every AI video still burns tokens, and tokens are bought with compute — the same compute this review argued, two weeks ago, is AI's real bottleneck. The cost of making content is not zero; it is the price of an atom's work, and that price is falling but never disappears. So content inflation has a floor, and the floor is set by physics.
Content inflation becomes money through exactly one channel: advertising. And here the chain is mechanical.
The middle term is the whole story, so it is worth making it airtight. Attention is rigid. A human being has twenty-four hours a day, a fixed fraction of it awake, and a hard upper bound on how much of that can be spent watching anything. No quantity of AI content adds a single minute to that budget. This is not a matter of taste or habit; it is a matter of physiology. The supply of attention is fixed, and everything that competes for it is bidding against that fixed supply.
Now follow the advertising money. An advertiser is not buying content. An advertiser is buying a chance at attention — a thousand chances, priced as the CPM. When the supply of content explodes, the supply of ad slots explodes with it, because every new piece of content carries new places to put an ad. But the attention those slots are selling has not grown at all. The same audience, split across far more slots, means each slot captures less attention. And an ad slot with less attention in it is worth less — the advertiser bids lower, and the CPM falls.
That is the quiet tax of content inflation. It does not show up in the CPI. It shows up in the unit economics of every publisher on earth.
But here is the counter-force hiding inside the word "deflation," and it is the part the doomsayers miss.
When AI content is everywhere and no one trusts it, trust itself becomes scarce again. This is the old economics of a lemon market — when buyers cannot tell good from bad, they price everything as if it were bad. But that same logic cuts the other way: whoever can prove they are not a lemon, gets to charge a premium for it.
So the CPM does not fall uniformly. It splits.
Mediocre inventory — the AI slop, the content farms, the endless reshuffle of the same idea — races toward the floor. There is nothing there to believe, so no advertiser pays anything to stand next to it.
Premium inventory — a real judgment, a real reputation, a real name — does the opposite. It appreciates. Advertisers do not fear cheap AI content. They fear AI content that nobody believes. When fakes flood the market, the genuine article rises in price, because the one thing AI cannot manufacture is the reason to trust what is on the screen.
The same AI has manufactured two inflations, and they are mirror images of each other.
One is monetary: 45 basis points — so small that reasonable people still argue about whether it exists, and shrinking as it is argued. It has a number, a central bank, and two banks fighting over it on the front page.
The other is exponential: it has flooded every screen, broken "rare is precious," and repriced the one asset that was never abundant to begin with — attention. It has no number, which is exactly why it is reshaping the economics of every publisher, every creator, and every advertiser while the argument rages on about half a percentage point.
One inflation is being measured to death. The other is being lived in, every time a screen loads — and it is the second one that will decide who still has an audience when the first one is settled.
Content is for informational and analytical purposes only — not investment, financial, or legal advice.