Notes

Google now pays more than 400 news publishers for AI use of their journalism. That is not a market. It is a panel.

On 15 October Google said its News AI pilot had passed 400 publications, with CNN, the FT, Axios and Forbes among the new names. No price is public.

It does not need to be. The buyer is also the company that decides how much search traffic each of those 400 publishers gets tomorrow. Nobody negotiates hard with the hand that ranks them. Google does not have to dictate the price. It is the only party at the table that could.

Nine days earlier, six scientific publishers went the other way through Redpine, a Swedish start-up: each publisher sets what an AI agent may use and what it pays per transaction.

My bet: when the pilot cheques stop, the publishers still earning from AI will be the ones who set their own price.

"Data product" is the most expensive phrase in the enterprise.

It sounds like a business. Too often it is a table with a roadmap.

The data-mesh playbook says start with the consumer. In practice, too many programmes start with the data they already have and go looking for someone to use it. That's how a catalogue fills up with products nobody uses.

Internal or external, direct or three hops downstream, the test is the same: which decision, made by whom, does this change?

If your team can't answer that in one sentence, you haven't got a product. You've got expensive inventory.

We charged a few hundred euros for years. The customer was willing to pay two thousand.

The directory business I worked in sold a listing for a few hundred euros a year. Then we built a website product on the same data and sold it for two thousand.

Nothing underneath had changed. Same advertisers, same categories, same records the company had kept for decades.

Here is the part that took me a while to understand. Two thousand euros was roughly what the average Belgian advertiser spent with us in a year, on everything. You would assume the money moved across, that they bought this instead of something else. A lot of them did not. Much of it was new spend, sitting on top of what they were already paying.

So a few hundred euros was never what the listing was worth. It was the most that shape could reach. What the customer was willing to spend sat somewhere else entirely, and the company had read one as the other for years before I got there.

That is what under-capture looks like from the inside. It does not arrive as a loss. The revenue line is fine, nobody complains, and you never find out what the customer would have paid because you never put anything in front of them that could carry it.

The company eventually filed for Chapter 11 anyway. An underpriced asset and a broken balance sheet were never the same problem, and fixing one does nothing for the other.

The headline prices held for seventeen years. What could be charged for changed.

NYSE charges the same three numbers for OpenBook today as it did in 2009. Five thousand a month for access, sixty for a professional user, fifteen for a qualifying nonprofessional. They are in the March 2009 SEC order and they are on the pricing guide dated 14 May 2026.

The title of that 2009 order tells you what was actually going on. It approves a nonprofessional fee, and it revises the unit of count that determines the device fees payable by data recipients. Even then, the pricing decision included deciding what counts.

The order defined subscribers to include both individual people and devices. On today's schedule there are three non-display categories at six thousand dollars each, and a redistribution fee of three thousand, sitting on the same page as the sixty dollar user.

None of it is buried. It is enumerated, with product codes, on a public document. This is a deliberate pricing decision and the seller is the one who makes it.

In the negotiations I have sat through, the rate gets the argument. What counts as a chargeable use gets almost none.

Before you negotiate the price, find out what you can be charged for.