Without AI spending, U.S. corporate capex would be negative, Pantheon analyst says

Pantheon Macroeconomics analyst Oliver Allen reports that AI-related capex is now a meaningful contributor to U.S. GDP growth: overall corporate capex rose 2.6% in Q4 2025, with intellectual property and software up 7.4% and computer and communications equipment up 61%, while all other equipment investment plunged 17%. Pantheon estimates AI-related sectors accounted for roughly 0.3 percentage points of the 2.5% average GDP growth in the first three quarters of 2025; consumer spending rose 2.4% in Q4 but appears concentrated among wealthier households benefiting from tech-stock gains, suggesting growth is partly driven by an AI-fueled wealth effect.

Analysis

Market structure: The Q4 data (capex +2.6%, computer & communications equipment +61%, IP/software +7.4%, other equipment -17%) signals a lopsided capex cycle concentrated in AI hardware and software. Direct winners are hyperscalers/cloud (AMZN, MSFT, GOOGL), GPU and AI chip leaders (NVDA, AMD, INTC) and semiconductor-equipment suppliers (ASML, LRCX, AMAT); losers are broad industrial capex names (CAT, DE) and legacy equipment OEMs where orders have fallen. Pricing power will tilt to GPU and wafer-capacity owners, keeping margins high for 6–18 months while creating overcapacity risk beyond that horizon.

Risk assessment: Key tail risks include an AI regulatory shock (data/compute limits), semiconductor supply-chain disruption, or a rapid reallocation by hyperscalers that leaves smaller suppliers with stranded inventory. Timeframes: immediate (days) — market sentiment and volatility around earnings; short-term (weeks–months) — inventory and guidance updates from NVDA/TSMC/ASML; long-term (quarters–years) — structural productivity and power/commodities demand. Hidden dependency: the boost is highly concentrated among a handful of firms and wealthy investors, so consumer spillover is fragile and reversible.

Trade implications: Expect upward pressure on US yields and a stronger USD if growth persistence shows; commodities sensitive to power and copper demand should out-perform selectively (FCX, energy grids). Execute concentrated, signal-driven exposure to top-tier AI infrastructure providers and equipment makers while hedging with short positions in industrial capex names; use duration trimming in fixed income to manage repricing risk over the next 30–90 days. Catalysts to watch: NVDA/AMZN/MSFT guidance, TSMC/ASML capacity commentary, Fed messaging and CPI prints.

Contrarian angles: The market is overstating breadth — AI capex is narrow and front-loaded, raising the chance of mean reversion in smaller AI plays within 6–12 months. Historical parallels: cloud capex cycles (2010–15) produced concentrated winners and post-cycle consolidation; expect a two-speed market where top platforms capture the value and peripheral suppliers get squeezed. Unintended consequence: rising electricity demand and localized grid constraints could increase operating costs for data-centers, pressuring smaller colo players first.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA via a staged approach: buy a 3-month call spread (ATM buy / +20% OTM sell) now and add to a 12–18 month LEAP (buy Jan 2027 $X calls) on any >10% dip; target +30–50% on near-term call spread, reassess after next earnings (30–60 days).
• Allocate 1–2% to semiconductor-equipment exposure: buy ASML (1% position) and LRCX (0.5–1%); set a protective trim if either reports capex slowdown or orders fall >20% sequentially over two quarters.
• Initiate a 1% short/hedge vs industrial-capex names: short CAT (CAT) or buy 6-month puts (delta ~0.35) sized to offset ~1–1.5% portfolio risk, or run a pair trade long NVDA / short CAT dollar-neutral to capture relative strength if industrial orders continue falling.
• Reduce portfolio duration: sell 25–50% of 10y+ Treasury exposure within 30 days and redeploy proceeds into cash/IG credit or the above equities; reverse only if CPI prints <0.2% m/m and 10y yield drops >50bps from current levels.