I built the first iPhone with Steve Jobs. The AI industry is at risk of repeating an early smartphone mistake

U.S. electricity demand is forecast to rise by 128 GW over the next five years while data centers under construction and planning will add roughly 93 GW, raising the prospect of underutilized infrastructure unless investment shifts from pure CapEx to AI-driven real-world applications. The author urges reallocating capital toward an ‘AI app store’ and builders—drawing analogies to AWS and the iOS App Store—to monetize data-center capacity across sectors such as energy, waste, healthcare and transportation, and cites Goldman Sachs’ $8 trillion AI opportunity as the upside if apps are developed at scale.

Analysis

Market structure: Hyperscalers (AMZN, GOOGL, GOOG) and chip leaders (NVDA) are the asymmetric winners because they capture both CapEx and the scarce inputs (GPU compute, colocated power). Data centers alone represent ~93 GW of planned incremental draw vs projected national demand growth of 128 GW over five years (~73% of incremental demand), which boosts pricing power for power producers and GPU suppliers but risks oversupply of raw rack capacity absent application demand.

Risk assessment: Tail risks include regulatory limits on AI power use or model exports, large-scale grid curtailments during summer peaks, and stranded data-center assets if commercial apps don’t materialize — each could cause >20–40% repricing for exposed names over 12–24 months. Immediate catalysts (days–weeks) are earnings/CapEx guidance from NVDA/AMZN; medium-term (3–12 months) are federal/state permitting and grid upgrades; long-term (2–5 years) is realization of an AI app ecosystem and monetization, which determines ROI on current builds.

Trade implications: Prefer owning software/app-enablers and AI compute suppliers over pure real-estate/data-center plays; NVDA (core compute) and AMZN/GOOGL (cloud + developer platforms) should outperform DLR/EQIX if application adoption accelerates. Cross-asset: higher power demand lifts natural gas/clean-power names; municipal utility bond spreads widen if capital needs spike, creating selective muni underperformance.

Contrarian view: Consensus underestimates timing and capital intensity of the app layer — building apps that monetize compute will likely take 18–36 months and favor vertically integrated hyperscalers and enterprise SaaS, not standalone colo REITs. Historical parallel: early cloud cycle where infrastructure led but value migrated to software (post-2008); unintended consequence: a rapid buildout could compress utilization rates below 60% and force mark-to-market impairments for weaker operators.

Key Decisions for Investors

• Establish a 2–3% portfolio long position in NVDA via 9–12 month call spreads (e.g., buy 1x 12‑month ATM call, sell 1x 20–30% OTM) to capture secular GPU demand while capping cost; increase to 4% if NVDA reports >20% QoQ revenue upside or gross margin expansion in next two earnings.
• Add 1–2% long in AMZN (AWS exposure) and 1% long in GOOGL (Cloud/Vertex) outright equities; overweight cloud/integration exposure and plan to trim if cloud growth falls below consensus by >300bps on next two quarter prints.
• Initiate a 1% short position in data‑center REITs (e.g., DLR, EQIX) funded by cutting 1% from traditional REIT exposure; add to shorts if national data‑center utilization drops below 60% or guided CapEx ramps exceed revenue-backed utilization assumptions within 6 months.
• Rotate 3–5% from pure infrastructure CapEx names into industrial/healthcare AI software winners (AAPL services ecosystem analogs, selected enterprise SaaS) over the next 3–12 months; prioritize companies showing >30% YoY software revenue growth and positive gross margins.