AI investment surge continues as CEOs commit to spending more in 2026

A Teneo survey shows 68% of CEOs plan to increase AI spending in 2026, despite most current AI projects not yet being profitable, indicating sustained corporate capex into AI development. That demand underpins continued strength for AI infrastructure plays—exemplified by Nvidia, which has a market cap near $4.6 trillion and trades at a forward P/E of ~25 versus the S&P 500 average of 22—creating a tension between strong demand drivers and elevated valuations that could cap further upside.

Analysis

Market structure: The immediate winners are high-end AI infrastructure suppliers (NVDA, TSMC suppliers, hyperscalers AMZN/MSFT/GOOGL) who capture a disproportionate share of incremental spend; losers are mid‑tier silicon/SaaS vendors that lack differentiated offerings. Pricing power will remain concentrated—Nvidia can sustain >20% gross margin premium as long as demand for H100/A100 class GPUs outstrips supply and lead times stay >12 weeks. Cross-asset: equity vols on NVDA and semiconductors will stay elevated (implied vol > historical vol), supporting options premia; higher capex raises industrial commodity demand (copper, power) and can modestly steepen rates if capex drives long-duration investment.

Risk assessment: Tail risks include stricter US/China export controls or a sudden de-rating if large customers pause orders (low-probability, high-impact). Time horizons: days — NVDA print/guide and order flow notices; weeks — hyperscaler capex budgets and TSMC capacity updates; quarters/years — model monetization and datacenter build cycles. Hidden dependencies: concentrated demand (top 5 customers >40% of incremental orders) and TSMC capacity are single points of failure. Catalysts: earnings guides, TSMC wafer allocations, and regulatory announcements can accelerate or reverse the trend.

Trade implications: Tactical allocation: overweight NVDA and cloud (AMZN, MSFT, GOOGL) while underweight speculative small-cap AI equities and unprofitable AI apps. Pair trades: favor long NVDA vs short ARKK-style baskets to capture quality premium. Options: favor defined-risk bullish call spreads (3–6 month) and buy tail protection (6–12 month OTM puts) ahead of major catalysts. Entry: accumulate on 8–20% pullbacks or after positive earnings guides; trim if forward P/E >30 or 12-month gain >40%.

Contrarian angles: Consensus underestimates commercialization risk — 68% of CEOs increasing spend while many projects are unprofitable implies follow-through revenue is uncertain and could produce post-build commoditization pressure. Historical parallel: infrastructure booms (cloud in 2010s) saw an overshoot then consolidation; if hyperscalers internalize chips or TSMC expands capacity faster than demand, pricing power and margins could compress rapidly. Unintended consequences include higher power costs and a mid-cycle oversupply of inference hardware; position sizes should be defensive and event-driven rather than all-weather buys.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVDA via dollar-cost averaging: deploy 1% immediately and add 0.5–1% on any 8–15% pullback within the next 8 weeks; set quantitative sell rule to trim half if forward P/E reaches >30 or position gains >40% within 12 months.
• Implement a relative-value pair: long NVDA (1–2%) vs short ARKK (1–2%) equal-dollar for 3–6 months to express quality infra over speculative AI apps; close if spread narrows by >15% in 30 days or widens >30% in your favor.
• Use options to limit cash outlay: buy a 3–6 month NVDA call spread (buy 10% ITM, sell 25% OTM) sized to 0.5–1% portfolio risk and simultaneously buy a 6–12 month NVDA 20% OTM put sized to 0.5% as tail insurance ahead of earnings/capex announcements.
• Reduce exposure to small-cap/unprofitable AI equities or sector ETFs with >40% concentration in early-stage AI names by 30–50% within 30 days; redeploy proceeds to cloud leaders (AMZN, MSFT, GOOGL) and select semis (NVDA) where monetization visibility is higher.
• Before increasing exposure beyond these sizes, require two confirmations within 60 days: (1) Nvidia quarterly guide showing sequential order growth, and (2) TSMC/wafer-allocation confirmation of sustained 3nm/5nm capacity for AI dies — otherwise cap position additions.