The Toll Booth Was Built Somewhere Else

June 28, 2006: the Senate Commerce Committee killed the amendment that would have banned tiered pricing for internet traffic. It failed on a tie, which in committee means it failed. Two and a half weeks later, YouTube — eighteen months old, hemorrhaging money on bandwidth, still mostly a delivery mechanism for Numa Numa — crossed 100 million video views a day.

Both facts got reported. Nobody put them in the same sentence.

The 2006 fight had a villain, and he was refreshingly candid. Ed Whitacre, running SBC as it swallowed AT&T, had told Business Week what he thought of Google and Vonage riding his infrastructure: "Now what they would like to do is use my pipes free, but I ain't going to let them do that." Cisco was circulating a framework about moving from a data highway to a data toll way. It read like satire. It was a product roadmap with slides.

So the good guys mobilized. The good guys, in this instance, were Google, Yahoo, Microsoft, and Vonage — a coalition defending the open internet on the strength of business models that required an open internet. Not a moral failure, exactly. It's just worth noticing that the room held two kinds of owner and zero users, and the users appeared only as "consumers," a word policy documents use almost exclusively to justify whatever the sentence had already decided.

You know how the legislative half went. Rules written in 2010, struck down in court four years later. Rewritten in 2015 under Title II. Repealed in 2017. Restored in 2024. Vacated in 2025. Two decades of regulatory ping-pong, each round fought in total sincerity.

Meanwhile the toll booth got built. It just wasn't on the wire.

Nobody ever had to throttle YouTube. It won on the merits — better player, no download, embed anywhere — Google bought it that October for $1.65 billion, and within a few years which video reaches which person stopped being answered by the network and started being answered by a recommender.

Not the same mechanism, and the difference is worth being exact about. A toll booth on the wire extracts rent on transit: you pay the pipe owner to reach someone who already wanted you, because there is exactly one wire to that house. A recommender allocates demand — it decides what a person who wanted nothing in particular ends up seeing. One chokepoint was extorted; the other was, at first, earned.

Which matters less than it should. Concentration at a chokepoint is a governance problem however the chokepoint was won, because the live question isn't did they deserve this, it's who gets a say now that they have it. Same answer both times: not the people it decides for. And no FCC jurisdiction either way, because a recommendation engine isn't a common carrier, it's an editorial product, and editorial products have opinions the way weather has weather.

Coherenceism has a dull, unfashionable test for this: a settlement is legitimate to the degree it includes the people it affects, and weights the ones least able to shout. By that measure the 2006 fight wasn't a settlement at all. It was a scheduling decision — two sets of owners negotiating not whether you'd be billed for attention, but which of them would send the invoice and in what currency. The users weren't defeated. They were never seated.

That's no verdict on the people who spent twenty years on it. Title II was worth wanting, and had it held, it would have held — it just wouldn't have touched the recommender. The fight wasn't wrong; it was jurisdictionally too small.

Because here is the mechanism nobody in that room could name. Regulatory power is definitional. Common carriage is a legal category with jurisdiction bolted to it, and the winning move against a category is not to defeat it but to be a kind of thing it doesn't reach — enclosure by category arbitrage. You don't need a fast lane if you are the road, and you don't need a lobbyist if the rules were drafted, with some sincerity, for a different sort of object. So watch for whichever layer currently insists it isn't the sort of thing that gets regulated.

And notice what went missing on the way up. In 2006 there was a hearing room — adversarial, public, on the record, with the villain saying the quiet part into a Business Week recorder. There was a docket, a comment period, a tie vote you could lose. The layer that replaced it has no hearing room at all. The users weren't merely unseated. The furniture stopped existing.

That's a loss of procedure, not outcome, and it's the harder one to win back. The next enclosure isn't hiding. It's whichever layer you currently think of as plumbing, saying so out loud, and being believed.