AI's Trillion-Dollar Question: IMF Sees Growth Support, BIS Sees a Reckoning

Start with scale, where the numbers are not in dispute. AI-related spending is projected near $800 billion in 2026 and close to $3 trillion cumulatively through 2028, according to Morgan Stanley Research, which estimates more than 80% of that total spending is still ahead of us - meaning the investment cycle is still in its early-to-middle innings, not its end.

The composition matters as much as the total: this is spending on servers, memory, power infrastructure, data centres, software platforms, and R&D - much of it long-lived physical capital rather than software that can be repriced or abandoned cheaply if demand disappoints. That's why the BIS is treating this as a systemic question rather than a sector-specific one.

On the productivity side, the evidence is real but narrow: task-level studies show 20-50% time savings on specific work, and one estimate puts the annual US consumer surplus from generative AI at roughly $172 billion. The IMF's own framing carefully separates that from economy-wide productivity statistics, which haven't yet shown the broad acceleration that would justify capex at this scale.

That gap - real micro-level gains that haven't aggregated into macro-level productivity growth - is the crux of the BIS's caution. General-purpose technology cycles historically show exactly this lag: investment runs well ahead of measured productivity for years before the aggregate numbers catch up, if they do.

There is a real-economy channel worth separating from the financial-markets one: higher demand for AI hardware has already strengthened export activity across parts of Asia's supply chain. Morgan Stanley estimates AI-related spending alone will add roughly 3.3 percentage points to true capital expenditure growth in 2026.

The financing structure is where the BIS's 'reckoning' language earns its weight. A large share of hyperscaler data-centre buildout is funded through retained earnings, corporate debt, and increasingly complex off-balance-sheet arrangements - structures that concentrate risk in ways harder to see from outside, and harder to unwind cleanly if demand growth disappoints.