For three decades the industry's pitch has been simple, and it isn't a lie. Better targeting means less wasted ad spend. Less waste means lower prices on the things you buy. Free content gets to stay free. Advertisers like it, publishers thank it, the user gets a feed that costs nothing at the register. In its own telling, it's a win for everybody.
We just want to know who's keeping score — and on whose books the gain is actually showing up. Five questions worth asking before signing off on a deal nobody bothered to write down. Five charts, plain English, no spooky music.
Behavioral data isn't expensive because it's scarce. It's expensive because prediction beats persuasion — and the more a buyer knows about you before the bid, the higher the bid lands.
In 1995 an ad cost a brand roughly four dollars per adult per year, give or take, and the buyer had no idea who you were. In 2024 it costs them three hundred times that — because for the first time in commercial history, the buyer can model what you'll do next with enough confidence to underwrite real revenue against that prediction.
The dollars are not paying for the ad. They are paying for the discount on guessing.
Markets emerge wherever supply is cheap and demand is hungry. You supply behavior at near-zero marginal cost. Buyers will pay roughly $0.18 per ad impression to act on it. The arithmetic took care of itself.
App opens. GPS pings. Background pings. Search queries. Ad impressions seen. Clicks. Dwell time. Volume up. Volume down. Each one of these is a tradable signal. A typical US adult emits about 7,200 of them in a single day, and at $0.0008 each, the implied annual value comes out close to what advertisers actually pay.
This is why the marketplace exists: not because anybody designed one, but because every byte you generate is, in aggregate, worth more than it costs to capture.
When Congress passed the Telecommunications Act of 1996, the legislative record made a series of explicit bets about how the internet would self-regulate. It is a useful exercise to grade them at three decades.
The single bet that kept its promise was access — most US adults are online, and the digital economy did democratize information distribution. The other four bets — self-regulation, consumer choice, anti-monopoly, and anti-discrimination — did not.
Worth saying plainly: no Congress since 1996 has passed a federal privacy law. The default regulatory regime is still a 30-year-old framework written before behavioral targeting existed.
What gets bought, watched, read, believed, where someone goes, who they meet, and what they think next — every one of those decision categories is now mediated by something running on substrate owned by four companies.
This is the part of the picture that doesn't fit on a single screen. AWS hosts your shopping decisions. Google indexes your reading. Microsoft owns the productivity layer where you decide what work matters. Meta calibrates your social context. The four cloud providers are also the four largest ad-platform owners, which makes them also the four largest behavioral-prediction shops.
The end game is not a single company controlling decision. It's four companies sharing the substrate underneath all decision — and renting back, at retail, the AI tools that learn from those same decisions to make the next prediction sharper.
For every dollar of US digital ad spend in 2024, roughly thirty-three cents came from the top ten advertisers. Roughly seventy cents from the top one hundred. The "long tail" people imagine — many small advertisers, fairly distributed — describes the last thirty cents only.
Amazon alone accounts for roughly one tenth of US digital ad spend. P&G operates sixty-plus consumer brands and routes them through the same four ad platforms. Disney, Comcast, AT&T, and Charter each own a constellation of media properties and a buying desk that talks to the same exchanges as everybody else.
This matters because the consolidation isn't only on the supply side. The buy side is consolidated too. Four substrate companies, ~12 consumer platforms, ~100 brand conglomerates — and roughly 260 million US adults at the bottom of the stack, paying via product markup.
In less time than it takes you to blink, your profile is queried, a dozen advertisers bid, money changes hands across five companies, and a single ad is rendered. The whole machine runs in under 100 milliseconds.
The attention economy wasn't designed — it was added to. Every few years a new layer quietly slid into the middle of the page-load and started taking a fee. Below, each box appears the year it actually became standard practice. Tap a box for the institutional file on that company. Tap any line between boxes for what literally moves across it — and what slips out of your hands when it does.
The polite industry term is "the programmatic stack." That's the technical name for the ten companies you've never heard of who quietly share fees on every page you load. Tap a box for what each one does, who runs it, and what the public record shows. Tap a line for what literally moves between two parties — and what gets handed over when it does.
Programmatic ad-tech runs ~10.5 trillion impression decisions per year in the US alone. That works out to roughly 333,000 per second, every second, day and night. Every page load is a new auction. Every scroll is a new decision. The counter below started at 0 when you opened this slide.
Press play. 1991 to 2026 in seven seconds. Watch the tech rail tear ahead while the law rail waits.
Brings privacy and ad-tech cases under "unfair or deceptive" authority. Major actions: Facebook $5B (2019), Spokeo (2012), Acxiom (ongoing inquiries).
Strongest privacy regime in force. Requires lawful basis for processing, granular consent, data-subject rights to access/deletion. Major fines: Meta €1.2B (2023), Amazon €746M (2021).
Modeled on GDPR. Requires opt-out from sale of personal info, right to access/delete. CPRA (2023) strengthened with sensitive-data category.
2024 in force. Requires "very large online platforms" to disclose algorithmic ranking, allow non-personalized feed options, conduct risk assessments, and let researchers access platform data.
1996 · grants platforms immunity for user-generated content. Has been interpreted to cover algorithmic amplification, narrowing platform accountability for what their recommenders surface.
Four US states require data brokers to register. Provides a public list. Does not constrain what they collect or sell — only that you can find out who exists.
If your news, ads, prices, and social proof are all different from the person next to you — your opportunities are already diverging. The next layer takes those signals and turns them into scores attached to your name. You don't get told. You just stop getting the call back.
Six sectors already score you. Four companies own most of the picture. Different gatekeepers see different slices — together they see everything. That's Episode 03.
We go inside the systems above. How the composite gets built, who insures the model when it's wrong, what happens to the score when you push back, and the four companies betting their next decade on owning your reputation token outright.
Same network. Same news day. Two different worlds. The system did not ask either person what they wanted to see. It inferred it. And then it billed the difference.
CONTINUE · EPISODE 03 · THE TRUST MARKET →A market-defender would push back: concentration in digital advertising reflects efficiency, not pathology. The platform layer is concentrated because the underlying technology has high fixed costs and low marginal costs — natural monopoly territory. The advertiser side is concentrated because the largest consumer brands have the largest budgets. Neither side is conspiring. Both are responding to scale economics.
This is the strongest defense and it is partly right. Concentration is not by itself proof of anti-competitive behavior. What's pathological is what concentration on both sides does to the third party — you — who appears in neither auction. Settling that requires antitrust analysis we do not do here.
If you find the "settlement system" frame too strong, swap it for "natural-monopoly equilibrium with high externalities." The numbers stand.