Essential infrastructure: The HALO assets of the AI boom

HALO infrastructure assets, especially electric utilities and transmission networks, are positioned to capture the AI boom with less risk than competing data-centre operators.

• AI token demand is expected to multiply 24 times by 2030, while hyperscaler capex is projected to reach US$740 billion in 2026 and US$910 billion in 2027.
• AI token demand is forecast to grow at a 133% CAGR from 2025 to 2030, driving data-centre electricity demand at a 32% CAGR over the same period.
• US hyperscalers are expected to spend around US$1.9 trillion in capex between 2025 and 2027, versus projected operating cash flows of US$2.2 trillion.
• Meta’s Hyperion campus in Louisiana will cost about US$10 billion, use 5 gigawatts and consume electricity equivalent to half of New York City’s total consumption; Entergy is investing around US$13 billion in supporting generation and transmission assets.
• By 2027, a server rack is expected to increase from 8 GPUs consuming 41 kWh to 567 GPUs consuming 600 kWh, while AI processing should represent about 29% of data-centre electricity demand by 2030.

Takeaway: Utilities, grids and renewable power providers offer investors exposure to structurally rising AI demand while avoiding the need to identify the winning hyperscaler; however, the outlook depends on power demand, renewable deployment, electricity efficiency and the eventual emissions impact.