The $600 billion wave of AI ‘capex’ growth boosting tech stocks will slow down this year or next, analysts warn

Major tech firms signaled a large surge in AI-related capital expenditures, with Meta guiding capex of roughly $135 billion for the year, Microsoft reporting $72.4 billion spent in the first half of its fiscal year (with sequentially higher quarterly capex but slowing Azure growth), and Tesla planning to double capex in 2026 and invest $2 billion in xAI. Wall Street estimates point to a multi-hundred-billion-dollar wave—Goldman Sachs sees AI capex at $539 billion (+36% YoY), Bank of America $641 billion (+36%) this year and $739 billion next year, while TSMC’s ~$54 billion CY26 guide and Wells Fargo/BofA notes support continued hefty spending—backing forecasts that data-center driven capex could meaningfully boost GDP and related sectors; markets reacted with Nasdaq futures and S&P futures up and mixed overnight moves in Meta (+7.85%), Tesla (+3.29%) and Microsoft (-6.53%).

Analysis

Market structure: The immediate beneficiaries are chip designers and foundries (NVDA, TSM), hyperscalers/data-center builders (META, AMZN) and suppliers of power/real estate for data centers; Wall Street estimates imply $540–740bn of AI/cloud capex in 2026–27 (consensus +34–36% in 2026, decelerating to ~15–17% in 2027), which suggests strong near-term demand but slowing incremental growth thereafter. Pricing power will favor constrained capacity owners (TSM, NVDA GPUs, memory leaders) while commoditized GPU and memory suppliers risk margin erosion once capacity catches up.

Risk assessment: Tail risks include rapid regulatory restrictions on generative AI or export controls on advanced nodes, a GPU/memory supply glut in 12–24 months, and grid/permitting bottlenecks that delay projects—each could produce >30% downside in suppliers’ equity. Time horizons: days (earnings-driven volatility), weeks–months (capex booking and supplier order flow), years (ROI on capex and demand elasticity). Hidden dependencies: wafer-equipment lead times, regional power availability, and labor for buildouts. Key catalysts: TSMC CY26 guide, hyperscaler quarterly capex disclosures, and Fed rate moves.

Trade implications: Direct plays—overweight NVDA and TSM (capture hardware scarcity) and selective hyperscalers (META, AMZN) that publicly commit cash; underweight or hedge MSFT-Azure exposure where growth is slowing. Pair trades: long TSM / short MSFT to isolate foundry demand vs. cloud-margin risk. Options: defined-risk call spreads on NVDA/TSM for upside capture and 3-month protective puts on a tech basket if capex signals decelerate. Rotate portfolio into semiconductors, data-center infra and utilities for 6–18 months; reduce exposure to cyclical software names if capex confirms slowing.

Contrarian angles: The market underestimates diminishing marginal returns—consensus capex growth may be front-loaded and create a 2027–28 supply surplus that compresses hardware ASPs by 10–25%. Historical parallel: cloud-capex booms in the 2010s produced both infrastructure winners and long tails of commoditized suppliers that underperformed for years. Unintended consequences include local political pushback on new data centers (permits, power rates) and higher effective costs that shorten payback periods; trim positions into rallies and size for mean reversion.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA within 0–4 weeks, add on pullbacks >5%; target +25–35% in 6–12 months, stop-loss at -12%.
• Initiate a 2% long position in TSM (TSM) and simultaneously short 1.5% MSFT to create a 1:0.75 pair exposing foundry demand vs. cloud-margin risk; rebalance after TSMC CY26 capex guide (expected in next 30–60 days) and trim if consensus 2027 AI capex growth <20%.
• Buy a defined-risk NVDA call spread (3-month, buy ATM, sell ~20% OTM) sized to risk 0.5–1% of portfolio to capture near-term hardware tightness while limiting premium exposure; roll or take profits if NVDA rallies >30%.
• Allocate 1.5–2% to META (long) focused on data-center rollout exposure; only add if shares correct >10% or if company provides incremental ROI metrics on capex within next two earnings.
• If two consecutive quarterly hyperscaler capex prints come in below consensus growth by >10ppt, reduce AI-capex exposure (NVDA/TSM/META/AMZN) by 30% and switch proceeds to defensive yield names (utilities/IG bonds) within 30 days.