Artificial Intelligence Investment Will Hit 2% Of U.S. GDP In 2026, Analyst Says—Nearing Defense Spending Levels

AI-related capital spending is highest in the U.S. and next highest in Norway and Saudi Arabia at 0.7% of GDP, according to TS Lombard. China’s data-center spending is about 0.4% of GDP, below Malaysia and Sweden, while the Eurozone is around 0.2% and Canada trails at roughly 0.15%. The piece is a cross-country spending comparison rather than a market-moving event.

Analysis

The key signal is not absolute spend, but relative willingness to subsidize the AI buildout at the sovereign level. Countries committing a materially larger share of GDP to compute infrastructure are effectively choosing to underwrite energy, land, power-plant, and grid vendors before the revenue model is fully proven; that tends to front-load demand for picks-and-shovels while delaying equity value capture for application-layer software. The market should expect a wider dispersion between beneficiaries tied to physical bottlenecks and those reliant on monetization of model usage.

The second-order winner set is likely to be power-adjacent: transformers, switchgear, cables, cooling, gas turbines, nuclear service, and utilities with regulated capex recovery. If AI capex remains a macro priority, the constraint shifts from chips to electrons, which means the most attractive exposure may be in industrials and utility infrastructure rather than semis alone. Conversely, jurisdictions with low spend intensity risk a widening competitiveness gap in productivity and cloud capacity, which can pressure domestic software, telecom, and data-center REIT ecosystems over the next 12-36 months.

The contrarian view is that the spending race may be more cyclical than secular at current valuation levels. If governments or hyperscalers hit power-permit bottlenecks, AI capex can pause abruptly, creating a sharp air pocket in orders for electrical equipment and construction services even while headline AI enthusiasm stays intact. That makes the near-term risk less about demand disappearance and more about timing slippage: orders stay, but revenue recognition moves right by 1-2 quarters.

Catalysts to watch are grid interconnect approvals, utility rate cases, and any sign that data-center power pricing is rising faster than compute utilization. A reversal would likely come from slower enterprise monetization, tighter fiscal conditions, or a policy shift toward export controls and localization rules that fragment supply chains and raise project costs. In that scenario, hardware names with long lead-time backlogs are less vulnerable than land, power, and construction proxies that are priced for uninterrupted build-out.

Key Decisions for Investors

• Long XLU / short QQQ for 3-6 months: express the view that AI spend increasingly accrues to regulated power assets and away from crowded software/mega-cap AI beneficiaries; stop if cloud capex growth re-accelerates sharply.
• Add to industrial infrastructure exposure via ETN, EMR, and HUBB on pullbacks: these names benefit from the bottleneck shift to grid and power distribution, with 12-18 month upside if data-center builds stay above trend.
• Pair long VRT or CARR against short a high-multiple AI software basket: if compute demand remains strong but power is constrained, thermal management and cooling should monetize earlier than application-layer software.
• For higher convexity, buy 6-12 month call spreads on SMR or OKLO only if power-policy headlines turn supportive; upside is large if AI forces nuclear re-rating, but timing risk is high.
• Avoid chasing pure-play data-center REITs at peak sentiment until power availability clears; use them as funding legs against utility/infrastructure longs if cap rates compress further.