The best AI investment might be in energy tech

Venture investors poured over $500B into AI startups in the last five years, but Sightline Climate warns up to 50% of announced data center projects may be delayed; of 190 GW tracked only 5 GW are under construction, 6 GW came online last year, and ~36% of projects slipped in 2025. Power constraints and rising electricity prices are driving big tech and startups into on-site/hybrid power, renewables and batteries (U.S. battery storage ~65 GW by year-end), exemplified by Google's 30 GWh Form Energy deal and Form Energy's reported $500M fundraising target ahead of IPO. Implication for portfolios: potential sector-level upside in grid-scale batteries, power-conversion/solid-state transformer companies, and energy-management software as a structural hedge against AI demand volatility.

Analysis

Grid scarcity will reprice the economics of compute delivery rather than the compute itself: companies that can internalize or pre-purchase large blocks of dispatchable capacity will take margin from third‑party colocation providers and wholesale markets. Expect a rerating over 12–36 months where ownership or long-term contracts for generation and storage become a de‑facto part of data‑center unit economics, compressing returns for players who rely on spot power or short-duration PPA contracts.

A second‑order supply shock will propagate into the power‑electronics and transformer supply chain: capacity limits in legacy transformer manufacturing and slow ramp in gas turbines accelerate adoption of solid‑state alternatives and integrated power stacks, creating a multi‑year demand stream for high‑margin, low‑volume suppliers. This structural transition favors firms that can monetize systems integration and software orchestration of distributed assets, not pure hardware commodity suppliers.

Catalysts arrive on different cadences: expect visible effects in power pricing and utility rate filings over the next 6–12 months, with project timelines and balance‑sheet commitments resolving over 12–36 months. Tail risks that could unwind this trade include a rapid drop in AI compute intensity per model (efficiency gains), an accelerated manufacturing scale‑up for conventional generation, or swift regulatory moves forcing utilities to prioritize grid access for data centers — any of which could flatten the premium for vertically integrated power solutions.

Key Decisions for Investors

• Long XEL (12–36 months): buy Xcel Energy for regulated upside from new rate designs and utility‑backed project lift; target +25–35% total return if incremental rate base approvals occur, downside -20% on adverse regulatory rulings or lower demand.
• Long GOOGL (GOOGL/GOOG) (12–24 months): overweight Alphabet to capture moat from vertically integrated energy deals and preferential access to capacity for Google Cloud; consider 12–18 month call spreads to limit capital with asymmetric upside (~2:1 reward:risk if AI growth normalizes).
• Long ORCL (6–18 months): accumulate Oracle as an underappreciated vendor for the software/hardware orchestration layer in hybrid on‑site/hybrid data centers; trade as a tactical 6–12 month earnings play—target +30–40% if enterprise capex reaccelerates, risk from cloud migration remains.
• Thematic exposure to power‑electronics/storage (6–36 months): rotate capital from pure AI hardware plays into battery/storage and solid‑state transformer suppliers via selective names or ETFs; implement via options or selective small caps for asymmetric upside while capping drawdown from tech execution risk.