Why the AI Bubble May Not Burst in 2026

A Teneo survey finds 68% of CEOs plan to increase AI spending in 2026 despite most current AI projects not yet being profitable, signaling continued corporate commitment to AI development. The report highlights sustained demand for AI infrastructure—benefitting firms like Nvidia, which the article notes has a ~ $4.6 trillion market cap and a forward P/E near 25 versus the S&P 500 average of 22—while also flagging investor caution around elevated tech valuations (Nvidia was ~11% below its 52-week high).

Analysis

Market structure: Persistent AI capex (68% of CEOs increasing 2026 spend) concentrates demand into infrastructure winners — Nvidia (market cap ~$4.6T, forward P/E ~25) plus semiconductor equipment and hyperscaler GPU procurement win pricing power; smaller software vendors and legacy CPU suppliers risk margin squeeze as spend funnels to specialized silicon. The circular intercompany cash flows referenced increase correlation across mega-cap tech, so an idiosyncratic slowdown at one large buyer could propagate rapidly through supply chains and equity flows.

Risk assessment: Tail risks include rapid GPU commoditization, EU/US regulatory restrictions on model deployment, or a synchronized revenue slowdown from reduced enterprise ROI (many projects unprofitable today) — each could trigger >30% downside in richly valued AI names within months. Immediate (days) risks: earnings surprises and options-driven volatility; short-term (weeks–months): corporate guidance revisions and capex lags; long-term (quarters–years): monetization of AI projects and hardware lifecycle replacement rates. Hidden dependencies: customer concentration, supply-chain bottlenecks, and inter-company deals that mask true cash flow; catalysts to watch: NVDA earnings, hyperscaler capex announcements, EU AI Act milestones over next 60–120 days.

Trade implications: Direct: establish a 2–3% core long in NVDA while forward P/E ≤25 and trim to 1% if NVDA rallies >30% from current levels or P/E >30. Pair: long NVDA (1–2%) vs short a basket (1–2%) of AI software names trading >40x forward P/E to neutralize market beta. Options: buy a 6‑month NVDA call-calendar or buy ATM calls and sell 20% OTM calls to fund exposure; hedge concentrated tech exposure with 3‑month puts sized to 30% of notional long.

Contrarian angles: Consensus underestimates the fatigue/inefficiency in AI projects — continued spending may inflate revenue growth without profit, setting up valuation compressions. The market may be underpricing correlation risk (circular agreements) and overpricing optionality in small-cap AI names; historical parallel: dot‑com capex followed by selective winners (not universal), implying selective concentration in infrastructure and profit-proofed franchises, not broad AI exposure.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA as a core holding while forward P/E ≤25; add up to another 1% on a pullback >15% from current levels, and trim to 1% if NVDA rises >30% or forward P/E >30.
• Construct a market‑neutral pair: long NVDA 1–2% funded by short 1–2% exposure to a basket of speculative AI/software names trading >40x forward P/E (equal-weighted; rebalance monthly) to capture dispersion while hedging macro risk.
• Implement options: buy a 6‑month NVDA call spread (buy ATM call, sell ~20% OTM call) sized to 0.5–1% portfolio risk; separately buy 3‑month puts equal to 30% of tech long notional as tail protection ahead of next 60 days of earnings/guidance windows.
• Rotate 20–30% of overvalued growth exposure into semicap and AI infrastructure (e.g., semiconductor equipment and cloud GPU suppliers) and allocate 3–5% to copper/industrial metals exposure as a thematic hedge to sustained capex; reassess within 90 days.
• Monitor these triggers over the next 60 days: (1) NVDA quarterly guidance vs. consensus (beat >5% = hold/trim, miss ≥5% = reduce to cash), (2) hyperscaler capex announcements (aggregate spend down >10% yr/yr = de-risk), and (3) EU/US AI regulatory actions — if any trigger is hit, reduce speculative AI exposure by at least 50%.