The AI Buildout Has Two Bills Coming Due

Power is now a data center product

Long-duration energy storage is shifting from a future grid requirement to a near-term necessity for hyperscalers facing multi-year grid connection queues, according to World Fund. Data center operators are signing contracts to secure always-on power for islanded renewable grids, and World Fund estimates that up to 30% of data centers may avoid a grid connection altogether by 2030 and operate as island grids. m:…2642

The economics are driven by delay: every day of delay costs a 100 MW data center around $1 million. World Fund requires long-duration technologies to reach an added energy cost under $20 per kilowatt-hour to be competitive, and notes that arbitrage alone will not pay back a 100-hour battery asset. Successful technologies must decouple energy and power costs to beat lithium-ion at durations beyond 15 hours. m:…2642

Capital is already moving: Form Energy raised $750 million in August 2026 at a pre-money valuation of $1.75 billion. But World Fund cautions that most long-duration storage companies are raising money for a market that might not go mainstream until the 2030s. m:…2642

The revenue arithmetic

Janus Henderson portfolio managers Denny Fish and Jonathan Cofsky put the payback question in numbers. Annual AI-related capital expenditure is expected to top $1 trillion in 2027 and exceed $1.6 trillion by 2030. At a 15% return on invested capital and a 30% operating margin, the companies writing the checks would need to generate nearly $2.5 trillion in sales by 2030 to justify that spending. m:…5338

The funding model is also changing. Janus Henderson says 2026 represents the last year that hyperscalers, in aggregate, can fund capital expenditure growth solely through cash flows, and that some will increasingly rely on debt issuance. Bottom-up revenue projections, the managers add, currently fall short of what the top-down model requires, though companies may accept lower return targets or higher margins to stay competitive. m:…5338

One escape valve in the analysis: sovereign governments view AI as essential to economic health and national security, so their spending is not return-driven. And if companies accept a 10% return on invested capital instead of 15%, required revenues could be 33% less than the baseline scenario. m:…5338

The fiber warning

State Street draws the historical parallel. Hyperscaler capital expenditure has grown 86.5% year over year, but investment expands capacity, and expanding capacity can move scarcity and economic rents to new parts of an ecosystem. The 1990s telecom buildout is the cautionary tale: Global Crossing and WorldCom aggressively financed the fiber groundwork, then entered bankruptcy as demand and capacity dynamics changed. m:…3106

Falling unit costs may accelerate adoption while making it harder for undifferentiated infrastructure providers to sustain margins, State Street argues. The long-term beneficiaries, in its view, may instead be healthcare companies that improve productivity through automation, financial institutions that combine AI with complex operational needs, and industrial firms that embed AI into processes. m:…3106

State Street's numbers underline the rotation: expected IT earnings growth is 41.8% for 2026-27 and 26.8% for 2027-28, while earnings growth expectations for industrials and financials remain comparatively resilient and improve going into 2028. m:…3106

The practical advice is to distinguish between volume growth and economic rent capture: a sector can grow enormously while the profits pool somewhere else entirely. m:…3106