U.S. IT-GDP Correlation Breaks at 2014Q3, Six Years Before AI Era, With Irish Data-Centres Tied to Multinational Accounting

The analysis applies a sup-Wald test to U.S. IT investment and real GDP, confirming the 2014Q3 structural shift survives Bonferroni correction while an imposed 2020 AI boundary does not. Irish metered data-centre electricity shows r = +0.41 with real GDP over the AI-era subsample yet collapses to near zero when real GNP replaces GDP, indicating the apparent energy-GDP nexus is an artifact of multinational tax footprints rather than domestic output. Cross-correlation profiles further reveal GDP leading data-centre load, positioning compute capacity as coincident or lagging rather than leading.

This timing aligns with the consolidation of hyperscale cloud platforms that converted IT capital expenditure from a growth accelerator into a mature cost center, a pattern visible in earlier shifts from hardware to software-as-a-service around 2013-2015. The Irish result extends long-standing critiques of FDI-distorted national accounts, where data-centre electricity registers in GDP via tax residency without commensurate local income effects, echoing earlier findings on pharmaceutical and software multinationals.

Subsequent work should test whether GNI*-adjusted aggregates or high-frequency grid telemetry restore predictive power for the next Kitchin trough. Energy-constrained investment cycles may now require separate monitoring from headline output measures in both policy and portfolio models.