Balancing AI Risks With the Race to Stay Ahead of China

Eclipse CEO Lior Susan argued that the AI industry should collaborate rather than slow development, warning that a retreat could weaken the US technological position relative to China. He highlighted rapid data-center expansion and physical AI as potential drivers of US economic growth, while urging the sector to better communicate data-center benefits to local communities.

Analysis

The investable AI bottleneck is shifting from GPUs to deliverable power, interconnection and cooling. That favors VRT, ETN and PWR more directly than hyperscalers over the next 12-24 months: their revenue is tied to physical deployment, while cloud platforms must absorb higher power and depreciation costs before monetization catches up. CEG and VST retain scarcity value where long-term power contracts are repriced, but the upside is highly regional and vulnerable to state-level ratepayer backlash.

The underappreciated second-order effect is that local opposition can turn an apparent capacity shortage into a timing problem for AI revenue recognition. Delayed permits or transmission connections do not eliminate hyperscaler capex; they concentrate it into fewer approved markets, increasing bargaining power for utilities, landowners, electrical-equipment vendors and gas-turbine suppliers such as GEV. Over 1-3 months, the relevant catalysts are announced power-purchase agreements, utility interconnection queues and hyperscaler capex guidance; over 6-18 months, regulatory treatment of data-center cost allocation will determine whether power producers or customers bear incremental grid costs.

Consensus remains too focused on aggregate data-center spending and insufficiently discounts execution risk. A softer-than-expected AI workload monetization curve would hurt META, MSFT, GOOGL and AMZN through lower returns on capital, yet could leave contracted electrical-infrastructure backlog intact. Conversely, a federal or state fast-track for generation and transmission would expand the addressable market for PWR, ETN and GEV faster than it expands GPU demand, because grid projects have longer lead times and currently constrain campus energization.

This interview alone is not a trading catalyst. Treat any policy-driven "AI infrastructure" rally skeptically until it is accompanied by disclosed megawatt commitments, signed PPAs, backlog conversion and permitted generation capacity.

Key Decisions for Investors

• Maintain a 6-12 month basket long VRT and ETN versus a short equal-weight cloud-capex basket (META, MSFT, GOOGL) only after confirming continued 2026 capex guidance; the trade captures physical-infrastructure scarcity while hedging broad AI sentiment. Exit if VRT/ETN book-to-bill falls below 1x or hyperscalers materially slow data-center spend.
• Watch PWR for entry following evidence that transmission awards and backlog conversion are accelerating; target a 12-18 month holding period. The falsifier is utility capex guidance shifting from transmission expansion toward maintenance or adverse regulatory disallowances.
• Use GEV as a selective power-shortage expression rather than a broad AI proxy, preferably on pullbacks and only if gas-turbine order backlog and delivery-slot pricing remain firm. Key risk is a faster-than-expected shift to nuclear, renewables-plus-storage, or project cancellations driven by local permitting.
• Avoid chasing merchant-power names CEG and VST solely on AI demand headlines. Add only where disclosed long-dated contracts lock in attractive economics; monitor state regulatory actions on data-center tariffs and cost allocation, which could compress realized returns within months.