Central bank AI: FRED becomes power infrastructure

In 1815 the House of Rothschild learned the outcome of Waterloo before the government in London did. The advantage sat in the private channel, rather than in the data.

The debate on central bank AI is about trust in data, before it is about models. On October 1, 2026 a Federal Reserve governor placed that idea at the center of an official speech. It rewards reading as a document about power.

Three times the channel went public, three times the rent vanished

In 1850 Paul Julius Reuter used carrier pigeons between Aachen and Brussels, where the telegraph wire broke off. A few years later the line covered that stretch. The rent vanished within a season.

In 1866 the transatlantic cable tied London and New York together reliably. Cotton prices in Liverpool and on the American market began to converge within hours, instead of weeks.

Three precedents are enough to call it a pattern. The mechanism repeats: once the channel becomes public, the profit moves from owning the data to designing the channel. Whoever builds the road collects the toll, even while giving the passage away.

The uncomfortable part concerns timing. The rent falls fast, the protocol stays for decades. The telegraph set the format of price lists long after the first movers lost their edge.

Patient capital looks to the second effect.

October 1, 2026: FRED called the greatest public good

Governor Christopher J. Waller opened FRED Con 2026 at the Federal Reserve Bank of St. Louis. The conference carries an explicit title: «Navigating Trust, AI and Storytelling in a World of Data».

In his remarks Waller calls FRED «the greatest public good the Federal Reserve has ever created»^[1], after seventeen years spent in different roles inside the central bank. The archive was born in St. Louis and he claims a modest contribution to its growth.

A monetary official opening a conference on trust in data weighs every word. That sentence works as an accounting classification: the free archive joins the institution's strategic assets, alongside the reserves and the tools of monetary policy.

The title of his speech supplies the rest: the data version of Godzilla versus Kong.

The title states the structure: two titans, one terrain

Godzilla versus Kong describes a duel between equal forces, where no referee exists. The nod to the film released in 2024^[2] serves to map a power relationship, rather than to land a laugh.

On one side the verified archive, slow, signed by an institution. On the other the generative machine, fast, fluent, able to produce a plausible and false historical series. The speech grants the second actor the same scale as the first.

Waller recalls that technological leaps have driven every growth phase of FRED. Here sits the remarkable point: the institution reads its own history as a sequence of adaptations to the dominant channel of the moment.

The two creatures in the title share the terrain, and the terrain is public data. The operational question becomes what shape the next adaptation takes, and what it costs whoever arrives later.

The mechanism: whoever reads the series now is an agent

American Banker describes how the Fed is adapting its data tools to AI-powered bots (the report sits here^[3]). The typical reader of a historical series changes nature: from an economist with a spreadsheet to a program querying an endpoint.

The shift carries a direct consequence. The dataset a model finds first, reads best and cites readily becomes the factual base of the economy for millions of generated answers.

Whoever holds that seat sets the starting level of every discussion about rates, employment and inflation. The access price of zero helps: an agent prefers the source that is open, documented and stable. Zero marginal cost becomes a competitive weapon, beyond a civic gesture.

Here free access stops being charity and becomes distribution. A central bank that gives its data away buys the starting position inside the models the world queries every day.

Structural pattern, decade-long horizon

The change is structural, with an arc of decades. The informational advantage of a central bank was for a century a matter of ownership: proprietary data, collected first, released afterwards.

That ownership is losing value. The European Central Bank admits that the natural rate of interest escapes its models, and language agents with web access hold their own against institutional short-term estimates on many indicators.

An advantage of a different nature remains, infrastructural. Whoever supplies the format, the series code, the dated revision and the signature of authenticity governs the way the world measures the economy.

Power moves from the content to the protocol. This is a regime change, rather than a cycle, and the distinction matters for allocators: a cycle gets waited out, a regime gets repriced.

The institutions that define the standard for machine-readable data today will collect authority for twenty years.

This desk's position

The thesis fits in one line: free public data has become strategic infrastructure, and Waller's sentence reads as monetary policy applied to information.

The market has priced the models and the chips. The layer below, meaning whoever produces the verified facts those models repeat, stays outside the prices.

Two pieces of contrary evidence would change the judgment. The first: a Fed retreat from open access, with fees or severe limits on automated use of the series. The second: the rise of a dominant private archive, cited by models more than the public one, a sign that the private sector writes the protocol.

A third path exists, more mundane: a cut to the resources of federal statistical collection, with a public good that degrades on its own. Whoever manages risk keeps this scenario open too.

Three implications for capital

The operational consequences arrive on different horizons, and they deserve to be kept distinct.

1. Family offices and sovereign funds, 36 months: weight on verified data infrastructure and on the operators that sign the provenance of a series.
2. Boards of directors, 24 months: a plan for the case of generated answers that diverge from the public source they cite.
3. Chief risk officers, 18 months: a scenario of corruption of reference data inside internal models.

The common thread stays trust in data, treated as a measurable risk factor instead of a technical detail. A board that discusses the quality of its own sources once a year arrives late.

A chief risk officer should simulate a precise case: the public series stay available, and the answers built on them diverge from the original data. Operational risk models rarely contain a line for this.

The theme for whoever talks to investors is another one. The macro narrative built on third-party estimates loses solidity, once the primary data becomes free and machine-readable.

A CFO who cites a private data vendor today risks an uncomfortable question eighteen months from now: why pay a rent on a public good. The answer exists, and it deserves preparing now.

The forecast

Expected event: by June 30, 2027 the Federal Reserve Bank of St. Louis publishes a FRED access channel declaredly aimed at AI agents, with dedicated official documentation.

Confidence: 70%. Horizon: June 30, 2027, meaning 271 days from today. Verification: the official page of FRED tools and interfaces.

Falsification signal: as of June 30, 2027 the official FRED documentation shows no interface or guide aimed at automated agents. A listed instrument here would be forced: the event concerns a public institution, outside the exchanges.

The forecast stays verifiable by anyone, with two clicks and a date.

What to watch:

• The FRED tools page: the appearance of interfaces or guides for automated agents.
• Fed governors' speeches on trust in data: frequency and rank of whoever speaks.
• The federal budget for statistical collection: changes in the resources of BLS and BEA.

These signals arrive before the headline. Whoever watches them sees the structure while it forms, instead of reading it after the fact.