Barings paper argues the AI buildout's limits are physical

On Aug. 17, Barings published a research paper identifying physical constraints as the source of investment risk in AI infrastructure. The paper, distributed on IPE Infra as asset-manager content, says those constraints carry "important implications" for inflation, corporate margins, and infrastructure completion — the three variables that drive any data-center underwriting.

The public text names no specific constraint; it sits behind a registration wall. The likely candidates — transformer lead times, substation capacity, water rights, labor availability, land — are familiar limits across the sector. They are the paper's argument only by inference. Enough to frame the argument, not enough to verify it.

The warning lands mid-stampede. This week Private Infrastructure Daily logged a hyperscaler signing a 20-year lease at Goodman's Tsukuba campus. The lease covers 50 MW. Mubadala is weighing a ¥1 trillion bet on a site in Akita. That site would run 500 MW. AWS lifted its Louisiana program to $18 billion. Every one of those commitments prices in construction finishing roughly on schedule.

The schedule is the failure point. When physical constraints delay completion, committed capital does not disappear; it compounds. Longer construction periods run up provisional financing costs, force escalation clauses, and squeeze margins on assets priced as if cash flows arrive on a fixed date. Yield-driven infrastructure investors should read schedule risk as inflation risk.

The paper is thought leadership, not disclosure, and the extract offers no numbers. But it names the terms on which every announced project will be judged. Construction schedules, not power forecasts, are now the binding constraint. The supply chain has not absorbed this much demand this quickly.