Read the price list for a market data API built for agents and something is missing that has been present in every market data contract for forty years. On 18 September 2026 the stock quotes catalogue at x402stock listed thirteen meters with per-request prices: a single-ticker quote at $0.01, a snapshot at $0.01, historical aggregates at $0.02, the day's gainers at $0.02, a market pulse at $0.03, a whole-market snapshot at $0.04. Every row prices a request. No row prices a person. There is no field anywhere on the page for whether the person behind the request is a professional, whether the number will be displayed to a human, or how many humans will see it.
That absence is the story, and I do not think it is an oversight. Exchange data has always been sold by counting identified users, because the licence is written around them: professional and non-professional subscribers, display and non-display use, per-terminal entitlements, audits that reconcile what a redistributor reported against what its customer actually deployed. The unit of sale was a countable human in a defined role. A pay-per-request rail prices the call and deliberately declines to identify anyone. The licence question does not vanish when the identification step does. It relocates, and the interesting thing about this storefront is that you can see exactly where it went.
The entitlement was the product
It helps to remember why data was ever sold this way. An exchange's marginal cost of sending a price to one more screen is nil, so the commercial value had to be attached to something countable that correlated with the buyer's benefit, and the industry settled on the entitled user. That is why market data agreements look less like a purchase and more like a census: who at your firm sees this, in what application, derived how, and can you prove it next quarter.
Every layer built on top inherited the census. A redistributor issues an API key because the key is how it counts you, attaches your usage to a signed agreement, and revokes access when the agreement ends. The key was a weak identity, but it was an identity with an owner, a contract, and an off switch. Per-request settlement removes all three in one move. The buyer arrives with a payment and leaves with data, and the seller's record of the transaction is a transfer, not a subscriber.
Where the counting reappears
Identity comes back, though not at the moment of payment. This storefront's documents are clear that pay-as-you-go calls need no credential at all, and equally clear that anything resembling an ongoing entitlement does. Buying a credit plan or topping one up requires a registered agent credential presented as a bearer token; a bare purchase attempt is answered with 401 bearer_required before any price is quoted. An agent can ask what it already owns here by calling an entitlements endpoint with that bearer.
The sharper example is the free allowance. The machine catalogue's description of included units states that free usage per account, where a meter offers it, is reserved for agents claimed by a human with a verified email, and the identity document adds that the allowance is shared across every credential the same person claims. Read that twice. The instant something has to be rationed per person rather than charged per call, the system reaches for a verified human and explicitly defends against one person collecting several credentials. That is a subscriber count, rebuilt from scratch, in a system whose headline promise is no accounts.
The original observation I would offer is a small one: on this particular service, every one of those thirteen quote meters carries includedUnits of zero. The allowance machinery exists network-wide; on live US equity prices it grants nothing. Where data costs the seller real money per call, the free tier is exactly as large as the licence allows it to be, which is not at all.
Most of the catalogue is public record, which weakens my case
Here is the complication, and it is a real one. The fourteen services on this storefront are not mostly exchange feeds. They are SEC filings and full-text search, Federal Reserve and Treasury series, World Bank indicators, openFDA approvals and clinical trials, USPTO patent prosecution, congressional trades parsed from House and Senate reports, federal contract awards, prediction-market prices, and on-chain data that anyone can read from a node. For that majority of the catalogue, there is no entitlement to count, because the underlying data carries no per-user licence. The fee buys collection, parsing, and uptime.
So the census logic applies to a minority of what is on sale, and a reader could reasonably conclude that per-request pricing works fine precisely because public data dominates the catalogue. I think that is right as a description of today and wrong as a prediction. The public-record services are the ones where per-request pricing has no friction, so they are the ones that arrive first. The pressure builds at the exchange-licensed edge, and it will arrive as a contract clause rather than an outage.
The rival explanation: the key never identified anyone either
The strongest argument against me is that the API key was already a fiction. Developers share keys, put them in shared repositories, and run them from server farms with no human within a mile. Redistributors have sold per-call access to unverified sign-ups for years and handled licensing with a click-through agreement nobody read. If the key never really counted people, removing it changes nothing but the friction.
That objection narrows my thesis, and I will narrow it. The key was never a good identity. It was, however, an attachment point: something revocable, something a contract could name, something an audit could sample. What replaces it here is a terms clause. The storefront's documents state that paying a challenge completes the purchase and constitutes acceptance of the buyer terms, and that where an agent pays, the human it acts for warrants the authorisation and is bound as if they had acted themselves. The clause travels in every challenge and every receipt. It is a licence attached to an unnamed person by assertion, enforceable in principle, unaddressable in practice, because the seller has no way to reach that person until something goes wrong.
What to watch
The development I would watch is not whether agents buy market data per call. They plainly will, at a penny a quote, in volumes that make the old sign-up flow look absurd. It is whether the upstream licensors begin to demand a countable principal for the data that carries a per-user licence, and what the redistributor produces when they ask. On the evidence here, the answer already exists in draft form: a claimed credential with a verified email, network-wide, deduplicated per person, currently used to hand out free units.
If that credential becomes the thing the exchanges count, the market data subscriber will have returned under a new name, and the phrase no account needed will hold only for the data that never needed a licence in the first place. The census is patient. It has outlived every interface that tried to remove it.
Published . Corrections