For a Solution to AI’s Energy Crisis, Look at the Human Brain

OpenAI reportedly targets building 250 GW of AI data‑centre capacity by 2033 (versus India’s 476 GW installed capacity in June 2025), underscoring the rapidly rising electricity demand of large AI models. The piece highlights neuromorphic hardware — atom‑thin 2D materials such as graphene, MoS2 and hBN enabling devices that co‑locate memory and computation (including single‑device 'neuristors') — as a potential solution to slash energy use, enable on‑device learning and privacy, and offer a post‑Moore’s‑Law growth path for computing performance.

Analysis

Market structure: Near-term winners are semiconductor-equipment and materials suppliers that enable new device physics (AMAT, LRCX, ASML), power-grid/electric utilities and grid-equipment makers (NEE, ETN, ABB) and base-metal producers (FCX, SCCO) because AI-capex implies large, multi-year electricity and copper demand. Losers over the long run could be undifferentiated cloud compute providers if on-device/neuromorphic chips meaningfully replace some datacenter workloads; pricing power will shift toward foundries and IP owners that can industrialize 2D-material processes.

Risk assessment: Key tail risks include a software/algorithmic breakthrough that slashes compute needs (reducing energy demand), regulatory limits on AI datacenters, or failure to scale 2D materials from lab to fab (yield/contamination). Timeframes: negligible market moves in days, order-book and utility-stock re-ratings in 3–12 months, and meaningful device replacement cycles in 3–10 years. Hidden dependencies: access to advanced fabs, precursor-chemical supply chains, and cross-licensing/IP barriers.

Trade implications: Direct plays — establish 2–3% core long positions in AMAT and LRCX and a 2% position in NEE and a 1–2% exposure to FCX (scale into 3–6 months). Use 12–36 month LEAPS call spreads on AMAT/LRCX (buy 2027 Jan 20–40% OTM call spreads) to profit from multi-year retooling without high theta. Pair trade: long AMAT/LRCX (0.5–1.5% each) vs a small (0.5%) hedge short on NVDA to express equipment demand outpacing GPU obsolescence risk.

Contrarian angles: Markets may underprice the multi-year scaling risk — neuromorphic is complementary not immediately substitutive of GPUs, so shorting NVDA aggressively is a high-risk contrarian. Also, edge-AI winners could be mobile SoC players (QCOM) and analog/sensor specialists rather than exotic graphene microcaps; avoid microcap graphene stocks >1% portfolio allocation until commercial fab wins are documented. Historical parallel: CPU→GPU shift took years; treat neuromorphic as a multi-year sector rotation into industrials, materials, and utilities rather than a quick tech trade.

Key Decisions for Investors

• Establish a 2–3% core long position in Applied Materials (AMAT) and a 2% position in LAM Research (LRCX), scaling into any pullbacks over the next 3–6 months to capture retooling orders tied to 2D-material integration.
• Allocate 2% to regulated utilities/grid-equipment: 1% NextEra Energy (NEE) and 1% Eaton (ETN) or ABB (ABB) to capture electricity demand and grid-upgrade capex; consider 3–5% XLU ETF if seeking broader exposure.
• Buy 12–36 month LEAPS call spreads on AMAT and LRCX (e.g., buy 2027 Jan 30% OTM calls and sell 2027 Jan 60% OTM calls) sized to 0.5–1% notional each to express upside while limiting premium risk.
• Take a 1–2% exposure to copper miners—start with Freeport-McMoRan (FCX) or Southern Copper (SCCO), building to target over 6–12 months as data-center and grid capex announcements (threshold: >$20bn incremental hyperscaler capex) materialize.
• Avoid or limit microcap/OTC graphene stocks to <1% portfolio weight until they announce commercial foundry partnerships or revenue (catalyst window: 6–18 months); replace speculative exposure with Qualcomm (QCOM) at 1–2% to play edge-AI SoC adoption.