Signal Scanner · ARTIFICIAL INTELLIGENCE & AUTOMATION · 8 August 2026

Steering by Drifting Instruments: AI Capex Has Decalibrated the Macro Data Boards Plan On

AI capital spending has grown large enough to bend the national accounts: deflator choice, terms of trade and one concentrated spending line now drive headline readings, leaving central banks, treasuries and boards setting 2026 to 2029 decisions against drifting instruments.

The argument about AI capital spending has settled into two camps, bubble or breakthrough, and both assume the numbers describing the boom are sound. The quieter development of this quarter is that they are not. The Bank for International Settlements puts AI-related infrastructure spending at around 1% of GDP in the most exposed economies and says the boom blurs the cyclical signals central banks rely on (BIS, 28/07/2026). The Peterson Institute has shown how much of measured real growth turns on the deflator you choose (PIIE, 18/05/2026). Anyone setting rates, tax forecasts or multi-year capital plans against those readings is working from a drifting dial.

Signal Identification

An emerging inflection in measurement rather than in capability. The signal is not that AI investment is large. It is that its scale, price behaviour and concentration now degrade the aggregates used to infer the cycle: real growth, potential output, terms of trade and the natural rates. Resolution turns on statistical decisions over the next two years, not on whether the boom continues.

Time horizon: 1-4 years (deflator and account decisions 2026-2027; distortion widest 2027-2029; resolved once AI output is measured separately)
distortion widest2026202720292030
Plausibility band: Medium–High
LowMediumHigh
Geographic / Jurisdictional Scope: Primary: the United States, Korea, Taiwan, Japan, Singapore, Malaysia, Thailand, Australia and the euro area. Spillover: AI-equipment importers and economies benchmarking policy to US readings.
PrimaryUSKorea, TaiwanJapan, SingaporeMalaysia, ThailandAustraliaEuro area
SpilloverAI-equipment importersEM commodity exporters
Sectors exposed:
Central banks and finance ministriesNational statistical officesCorporate treasury and planningAsset allocation and pensionsSovereign and corporate creditSemiconductor exporters

What's Changing

Three readings that used to move together have come apart. On scale, data centres and IT manufacturing facilities alone reached 0.8% of GDP in the United States (BIS, 28/07/2026), taking US computing infrastructure as a whole to about 1.5% of GDP in the first quarter of 2026 against a 2015-2022 average of about 0.7% (Epoch AI, 05/06/2026).

On prices, the deflator is doing visible work. A quality-adjusted AI price index in place of the conventional one would have lifted measured US real growth by roughly 2 percentage points in 2024 and roughly 4 percentage points in 2025, an upper bound on its authors' own account (PIIE, 18/05/2026). Those GDP shares are nominal: real compute growth runs faster, real construction growth slower (Epoch AI, 05/06/2026).

On geography, the gains are narrow. Terms-of-trade gains lifted upstream exporters' national income and purchasing power by substantially more than real GDP; in Korea, five firms took 43% of export earnings in the first quarter of 2026, up from 27% two years earlier (BIS, 28/07/2026). The top four net exporters of AI hardware beat first-quarter forecasts by an average of 4.4 percentage points; the rest of the world came in 0.3 percentage point below (IMF, 08/07/2026).

A narrow base: where the AI economy actually sits

0% 20% 40% Korea, five firms' export share two years earlier 27% Korea, five firms' export share Q1 2026 43% US business AI use, national 19.8% US AI use, Information sector 39.7% US AI use, Finance and Insurance 33.9% Indigo: concentration of export earnings. Orange: breadth of business adoption. Two ratios that a headline growth print does not separate.

Source basis: BIS (28/07/2026); US Census Bureau (26/05/2026).

Disruption Pathway

Stage one is running now: the distortion is acknowledged and uncorrected. There is no dedicated national-accounts line for AI investment and the equipment carries high import content, so the net contribution to GDP is smaller and far more volatile than gross investment implies (Federal Reserve, 17/07/2026). Stage two, across 2027 and 2028, is corrective: separate AI accounts and quality-adjusted price indices arrive and history is restated. Stage three resolves it, as a payoff visible in total factor productivity or as an investment turn running the same composition effect in reverse.

Stresses concentrate at three seams. Monetary calibration is the first: overestimating supply or underestimating demand leaves policy too accommodative, with the mirror risk on the other side, and the boom may be moving the natural rates themselves (BIS, 28/07/2026); the ECB agrees that the net effect on R* remains uncertain (ECB, 06/07/2026). Fiscal treatment is the second, with windfall economies told to avoid procyclical spending (IMF, 08/07/2026). Corporate planning is the third, because national series flatter a narrow adoption base: US business AI use sat at 19.8% nationally in early May against 39.7% in Information (US Census Bureau, 26/05/2026). Two adaptations follow: separate AI accounts, and finance functions that decompose aggregates first.

Why This Matters Now

For boards, CFOs and sovereign asset owners the exposure is not to AI itself but to the macro assumptions under the plan. A growth print that is mostly one capital-spending line says little about the demand a given business faces; a windfall that lifts national income above real output has not widened the tax base permanently; a rate set against a natural rate nobody can pin down carries wider error bars than the forecast admits. The revision needed is to the inputs, not the strategy: strip the AI capital-spending contribution out of demand forecasts, treat export windfalls as cyclical, and hold wider ranges on the rate path. The corrections will land after this year's plans are signed.

Decision-action posture for this signal: Prepare — the distorted readings are already inside this planning cycle, and the corrections land over the next 12 to 24 months, repricing positions taken now.

Counter-Argument

The strongest objection is that this is a future problem dressed as a present one. Federal Reserve staff, on the same public data, find labour productivity trends across high, medium and low AI-exposure sectors relatively consistent over time, and read the evidence as a buildout phase (Federal Reserve, 17/07/2026). The Peterson Institute concedes as much, writing that its argument is not that AI mismeasurement today is large enough to materially shift headline numbers, but that it may become so within a small number of years (PIIE, 18/05/2026).

The concession is about today, and the decisions are not. Rate paths and multi-year capital plans fixed this year run to 2029, the window in which both sets of authors expect the gap to widen. And the cross-sectional evidence has arrived: a 4.4 percentage point forecast surprise for four AI-hardware exporters against a shortfall for everyone else is not a rounding error (IMF, 08/07/2026).

Implications

This is a durable change in how aggregates should be read rather than a transient wobble. The inflection window is 2026 to 2028, when statistical offices decide whether AI output gets its own account and price index; once it does, history is restated. Positioned to gain: forecasters and allocators who decompose national data first, and exporters that save windfalls as cyclical. Positioned to lose: planners who read a concentrated capital cycle as broad demand, and governments that spend a terms-of-trade gain as a permanent base (IMF, 08/07/2026).

Early Indicators to Monitor

Disconfirming Signals

Strategic Questions

Keywords

AI capital expenditure; national accounts; GDP deflators; quality adjustment; terms of trade; potential output; natural rate of interest; monetary policy calibration; AI satellite accounts; macroeconomic measurement; semiconductor exports; corporate planning assumptions

Bibliography

Source tiers: Tier 1, governments, regulators and intergovernmental bodies. Tier 2, think-tanks, academic institutes, major consultancies and quality data providers. Tier 3, quality journalism and specialist trade press. Tier 4, vendor, company and practitioner sources, used only as directional corroboration.

This Signal Scan is one component of a Shaping Tomorrow intelligence cycle on Artificial Intelligence & Automation. Every monitored topic also includes monthly Change Trackers that assess what is materially shifting in momentum, and a quarterly Decision Intelligence briefing that converts the cycle into board-level strategic implications. Cycles may also close with Action Triggers, an end-of-cycle synthesis that reads across the three formats and surfaces patterns no single deliverable carries on its own. Continuous intelligence on this topic is available as a Single-Topic or Multi-Topic programme.

Discuss a Tailored Programme

Prepared by Shaping Tomorrow: 8 August 2026