The Two-Tier Web

"Now what they would like to do is use my pipes free, but I ain't going to let them do that," said Ed Whitacre, then running SBC, the company that was about to buy AT&T and take its name. "For a Google or Yahoo! or Vonage or anybody to expect to use these pipes for free is nuts!"

Strip the bluster and the sentiment bumper-stickers itself: they use my pipes free, and that's bull. It's worth sitting with how strange that claim was. Google was not using SBC's pipes for free. Google paid enormous sums for its own bandwidth. SBC's customers paid SBC every month for internet access. What Whitacre was describing was not a company freeloading on his network — it was two parties who had both already paid, meeting in the middle, exactly as the internet was designed to let them. What he wanted was to charge both ends for the same pipe. He wanted to be paid by the customer for the connection, and paid again by whatever the customer chose to connect to. And he was annoyed that the architecture of the internet didn't let him do that yet.

That annoyance is the origin of what we now call the net neutrality debate. Not a principle, not a movement — a phone executive's irritation that the network he'd inherited treated every packet the same, and his stated intention to change that. Everything since has been an argument about whether he got to.

i · the fight nobody was having

The remarkable thing, which MIT Technology Review noted at the time, is that a year earlier this was not a debate at all. In 2005 net neutrality was a non-issue — an obscure preoccupation of a handful of legal academics. There was no fight because everyone assumed the answer. The internet moved data without asking what the data was or who sent it, and that was so obviously how it worked that nobody thought to defend it. You don't pass a law protecting gravity.

Whitacre's interview turned an assumption into a demand, and a demand into a fight. By June 2006, it was in front of the Senate Commerce Committee, which considered — and rejected, on June 28 — amendments that would have banned tiered pricing for internet access. That rejection is the moment the two-tier web stopped being unthinkable and became a live policy option, one committee vote away from law in either direction.

The sides assembled fast, and they're the same sides we have now. On one side, the carriers — SBC/AT&T, the cable companies — plus Cisco, which built the routers that would do the actual prioritizing and stood to sell a great many of them if prioritizing became the business. Their argument was infrastructural and, on its surface, reasonable: we paid to lay the fiber, the heavy users are the ones clogging it, and we should be able to charge them for the load they put on what we built. Someone has to pay for the network, and it might as well be the companies making billions on top of it.

On the other side, the companies that lived on top of the network — Google, Yahoo, Vonage, Microsoft — plus the academics and activists who understood what was actually at stake. Their argument was about permission. The internet had produced an unbroken run of things nobody asked the carriers about — the web, email, search, voice calls over data — precisely because you didn't need the carriers' blessing to launch them. A neutral network is one where a kid with a good idea reaches the same audience as an incumbent with a lobbying budget. Introduce tiers, and the next Vonage has to negotiate with the pipe before it can compete with the pipe's own phone service. Which, not incidentally, is exactly the competition Vonage represented to the phone companies.

ii · what two tiers actually means

Strip away the jargon and the two-tier web is simple. Today, when you request something online, the network delivers it at whatever speed the physics allow, indifferent to what it is. A two-tier network introduces a fast lane and a slow lane, and lets the carrier decide who rides in which — usually by who paid. The premium services you're willing to fund get the fast lane. Everyone who can't or won't pay gets relegated, quietly, to worse.

The carriers always insisted this was about adding a fast lane, not degrading the slow one. But a fast lane is only worth paying for if the alternative is slow, which means the incentive to let the ordinary lane rot is built into the model from the first day. You do not sell premium unless standard hurts. The whole value of the toll road is the traffic on the free road.

And notice who chooses. In a two-tier web, the carrier — a company you already pay, that faces little or no competition in most American markets — becomes the arbiter of which services thrive. Not the market, not the users. The pipe. This is the part the infrastructure argument always skated past: the carriers weren't asking to be compensated for bandwidth. They were asking to become the gatekeepers of everything that traveled over it. The money was the pretext. The control was the product.

iii · they lost the argument and won the war

Here's where the disappointed idealist in me has to tell you how it went, because the 2006 fight reads very differently once you know the ending.

Net neutrality's defenders appeared to win. After a decade of trench warfare, the FCC under Tom Wheeler enacted the Open Internet Order in 2015, reclassifying broadband so that the neutrality rules had real legal teeth. It was a genuine victory, hard-fought, and it lasted about two years. In 2017 the FCC under Ajit Pai repealed it, and the great predicted catastrophe — the throttling, the shakedowns, the extortion of Netflix — largely did not arrive in the crude form everyone had braced for.

But it didn't need to, because the literal two-tier web arrived anyway — through a different door, and nobody stopped it, because it didn't look like the thing everyone was watching for. It came dressed as a favor to the consumer. It was called "zero-rating."

Zero-rating is the carrier saying: use this app and it won't count against your data cap. Stream from our partner, free; stream from a competitor, and it eats your allowance. No throttling, no blocking — just a thumb on the scale, gentle enough to feel like a gift. It is Whitacre's dream implemented as a loyalty perk. The carrier chooses winners. The services that can pay for, or negotiate, zero-rated status get the fast lane; everyone else gets the meter running. Same two tiers. Nicer branding. That much is a straight line from the 2006 fight to now: the exact mechanism Whitacre wanted — the carrier picking winners at the pipe — got built, just wearing a friendlier face.

But something larger happened alongside it, and it's worth being careful not to collapse the two, because they're different animals. While everyone watched the pipe, the winners got picked anyway — one floor up, at the application layer, by the handful of platforms that became the actual internet for most people. That enclosure wasn't a consequence of the neutrality loss; Apple's App Store cut wasn't downstream of Pai's 2017 repeal. It was a separate enclosure, at a separate layer, running on its own clock and its own logic. When the vast majority of your online life happens inside a few apps owned by a few companies, the neutrality of the pipe beneath them is close to irrelevant. So there are really two endings here: the pipe-level tier Whitacre wanted, which arrived as zero-rating — and a second, broader enclosure that the whole neutrality fight never even addressed, because in 2006 the layer it happened on barely existed.

iv · the pattern under the pattern

What 2006 actually shows, read from here, is that the neutrality fight was a fight about layers — about which layer of the stack gets to charge rent and pick winners. The activists defended neutrality at the pipe and, on paper, mostly held it. Nobody was defending neutrality at the app store, the platform, the ad exchange, because in 2006 those barely existed. So that's where the enclosure happened instead. The water finds the level nobody's watching.

This is the thing worth taking from the whole saga, and it's not a comforting one. You cannot win a war over openness by defending a single layer, because the impulse Whitacre voiced — they use my thing for free, and that's bull, I should be the one who charges for access to it — is not a phone-company impulse. It's the default impulse of anyone who finds themselves owning a layer that everyone else has to pass through. Defeat it at the network and it reappears at the platform. Beat it at the platform and it'll be waiting one floor up again.

And that's not a hypothetical to close on — it's the fight already underway. The indispensable middle now is the AI layer: the handful of companies that own the large models everyone else is starting to build on, route through, and pay to reach. Whitacre's sentence rewrites itself almost cleanly — they use my model for free, and that's bull — except this time the thing being enclosed is closer to a shared inheritance than a phone line. The models are trained on the collected writing and knowledge of nearly everyone, gathered from the commons and metered back to us through a few private gates. The 2006 lesson isn't history; it's the operating manual for that enclosure. You cannot defend openness by guarding a single floor, because the impulse to own the middle simply takes the elevator to whichever floor nobody is watching. In 2006 it was heading for the platform. Today it's arriving at the model.

The 2006 debate wasn't really about pipes. It was the first skirmish in the permanent question of who owns the layers we all have to route through, and whether they get to charge us twice for the privilege. Whitacre asked that question out loud, twenty years ago, and lost the round. The question never went anywhere. It just changed floors — and it's stopped saying the quiet part into microphones.