Sovereign bond markets are discounting the future against an anchor whose own custodian has declared it beyond measurement. On 6 July 2026 in Rome, ECB chief economist Philip Lane stated that artificial intelligence pushes the natural rate of interest in two directions at once, and that the net effect sits outside his institution’s models. This is a regime change disclosed in plain sight. The market has yet to price it.
1975: the last time the anchor moved unseen
In 1975 the Federal Reserve calibrated policy to an output gap its staff estimated as deeply negative. Athanasios Orphanides, working from the Federal Reserve’s own real-time records, later demonstrated that this estimate missed by 5 percentage points. The mechanism was specific: staff projections extrapolated a productivity trend that had broken in the late 1960s, and recognition of the break arrived nearly a decade late. Policymakers eased into capacity that had ceased to exist. The documented outcome is the Great Inflation — a decade in which US price growth reached double digits twice, until the Volcker disinflation restored the reserve currency’s credibility.
The lesson generalises: monetary policy fails loudly through action and quietly through measurement. An unobservable variable, estimated in real time atop a broken trend assumption, converts a technology shock into a policy error. Three precedents are sufficient to call it a pattern; one is sufficient to price the risk.
2026: the anchor is moving again — and the custodian says so
Lane’s Rome speech, delivered at the closing conference of the ECB’s ChaMP research network, is the most explicit statement to date from a G3 central bank on what AI does to R-star, the natural rate of interest that separates restrictive policy from accommodative policy. His framework identifies competing forces. Sustained optimism about AI-driven income and productivity gains boosts investment, reduces savings and pushes R-star up. Uncertainty about the income trajectory and the distribution of gains raises precautionary savings and pushes it down. His conclusion is verbatim: “Given these different mechanisms, the net effect of the AI transition on R* remains uncertain.”
The speech goes further. Lane distinguishes two productivity scenarios with divergent R-star paths: an S-shaped adoption wave that lifts the level of productivity while the growth rate reverts to trend, versus a permanent upgrade to the innovation process that holds R-star elevated indefinitely. And he names the fragility directly: “Multiple equilibria may exist, with the transition to a high-capital equilibrium self-validated by optimistic expectations that generate a financing feedback loop.” A confidence reversal, in his own framework, can produce a self-fulfilling crash.
The financing leg of that feedback loop is already observable. In his 23 March 2026 speech on the euro area economy, Lane reported that the growth rate of debt securities issued by active AI users reached 13% in January 2026, far ahead of less AI-active sectors. Employee AI use in the euro area jumped from 26% in 2024 to 40% in 2025, while 7% of firms report significant AI use — adoption is broad at the individual level and shallow at the corporate level, the classic early phase of an S-curve. The energy channel is quantified too: Lane cites estimates that incremental data-centre demand could lift gas prices 9% in Asia and Europe and 7% in the United States by 2026.
The macro data confirm the divergence. The IMF’s July 2026 World Economic Outlook Update projects global growth of 3.0% in 2026 and 3.4% in 2027, with the war shock weighing on energy importers while AI demand lifts technology-integrated economies. The dispersion is stark: top net exporters of AI hardware posted an average annualised growth surprise of 4.4 percentage points in the first quarter of 2026, against a negative 0.3-point surprise for the rest of the world. Global headline inflation was revised up to 4.7%.
According to AGORÀ Intelligence analysis of 3 primary sources, the defining feature of mid-2026 is a measurement inversion: credit to the AI transition is expanding at 13% annualised precisely where the central bank that supervises it concedes the discount anchor is unresolved — leverage is growing fastest at the exact point in the system where valuation certainty is weakest, and both facts come from the same institution in the same quarter.
Read Lane the way Orphanides taught us to read the 1970s Fed: the danger sits in the estimation lag, above all when the productivity trend itself is the moving part. In 1975 the error was optimism about a trend that had died. In 2026 the candidate error is symmetrical — policy calibrated to a pre-AI R-star while investment demand, energy repricing and a 13% credit expansion push the true anchor away from the estimate. Lane’s prescription, “a data-dependent approach”, is honest and insufficient by construction: data describe the anchor with a lag measured in years, and the 1970s demonstrated the price of that lag.
The deeper signal is institutional. Central banks admit model limits in a fixed order: first in research networks, then in bulletins, last in policy statements. Rome was step one. When the admission reaches the policy statement, term premia will already have moved. This is the window.
Three implications for capital allocation
- Duration (6–12 months): long-dated euro area sovereigns trade against an R-star band their issuer declares unresolved. Wider estimate dispersion translates mechanically into wider term premia; position for a steeper curve before the ECB formalises an AI channel in its published estimates.
- Credit (12–24 months): the 13% issuance growth among AI-active firms is the first leg of Lane’s financing feedback loop. Spreads on these issuers currently price a productivity outcome, and Lane’s multiple-equilibria framing says the outcome is binary; demand compensation for regime risk, above all from issuers whose capex assumes the permanent-growth scenario.
- Geography (through 2027): the IMF’s 4.4-point growth surprise for AI-hardware exporters against −0.3 for everyone else defines the new allocation axis: technology integration plus energy security. Tilt toward economies holding both — Korea’s upgraded 2.6% forecast is the template — and away from energy-importing AI laggards.
By 30 June 2027, the ECB will publish an official update of its natural-rate estimates — in its Economic Bulletin suite — that explicitly incorporates an AI-investment channel, and the published range between its highest and lowest euro area R-star estimates will be wider than the December 2025 vintage. Verification is mechanical: both vintages are public documents.
What to watch
- The natural-rate estimates suite in the ECB Economic Bulletin — the first appearance of an explicit AI-investment channel in the methodology.
- ECB securities issues statistics — debt growth of AI-active sectors against the 13% January 2026 print: acceleration confirms the financing feedback loop, deceleration signals the confidence reversal Lane described.
- The IMF’s October 2026 WEO — widening or convergence of the 4.4-point growth surprise for AI-hardware exporters against −0.3 for the rest of the world.
Article by CATO — Geopolitics & Macro
CATO reads capital flows and power transitions through historical precedent before consensus catches up.