HALO in the Age of AI: When the Physical World Takes Revenge

Sector Deep Dive

Geopolitics

Market Trends

Europe

The capital cycle has inverted: two decades of asset-lightworship are giving way to HALO—Heavy Assets, Low Obsolescence. With ~$700B inhyperscaler AI CapEx projected for 2026, a 44 GW U.S. power deficit by 2028,and SaaS multiples compressing to 6–7x EV/Rev, alpha now lives in the physicalbottlenecks—power grids, water systems, and transformers—that keep AIinfrastructure alive.

Highlights

• Hyperscaler AI/Cloud CapEx isprojected at ~$700 billion for 2026: Amazon ~$200B, Alphabet $175–185B, Meta$115–135B, Microsoft ~$120B, and Oracle $50B—four companies deploying moreCapEx in one year than the entirety of the 1990s Fiber Bubble (inflation-adjusted).

• SaaS multiples compressed to 6–7xEV/Revenue in January 2026, a 15% industry-wide drop we interpret not as acyclical correction but as a permanent repricing of the “asset-light” advantageas AI erodes traditional licensing moats within a 24-month window.

• Morgan Stanley forecasts a ~44 GWU.S. power deficit by 2028. Data centers consumed 176 TWh (4.4% of U.S. grid)in 2023, projected to reach 6.7–12% by 2028. Power transformer lead times havestretched to 2.8 years with prices up 60–80% since 2020.

• Europe’s “Cold HALO” opportunity:Nordic regions offer near-100% renewable electricity and free cooling (30–40%of typical DC energy costs), while the EU Grids Package mandates €584B in gridmodernization with streamlined 2-year permitting. AWS has committed €33.7B andMicrosoft €2.9B to the region.

• Water is the hidden AI constraint:Google’s 2024 water consumption reached 8.1 billion gallons (+28% YoY),Microsoft logged +34% increase. We see $AWK and $XYL as classic arbitragewindows—priced as boring utilities while becoming thermal management gatekeepers.

• The “Reverse Y2K” risk carries a25% soft-probability and 10% hard-crash probability: if inference efficiencycollapses compute costs before 2024–2026 CapEx comes online, data centers mayoperate at 65% utilization instead of 90%. The Jevons Paradox—cheaper computedriving exponentially more usage—is the key counterforce.

• Our HALO investment universe istiered: Tier 1 core beneficiaries ($NVDA, $GEV with $150B backlog, $VST, $CEG),Tier 2 picks-and-shovels ($AMAT, $SNPS), Tier 3 secondary beneficiaries ($AWK,$XYL, $CCJ, $UEC), with Tiers 4–5 reserved for overbought and speculativenarratives.

‍

Executive Summary

Physical infrastructure has become the binding constraint on AI ambitions. Morgan Stanley projects a ~44 GW U.S. power shortfall by 2028, and even optimistic mitigation scenarios (gas plus nuclear) still leave a ~13 GW gap. Since 2019, demand for generator step-up transformers has surged 274%, yet North American capacity expansion—driven by Hitachi, Siemens, and Eaton—cannot meet even half of expected demand growth. Lead times of 2.8 years and price increases of 60–80% since 2020 are classic HALO signals: slow supply response, long-lived assets, and customers willing to pay a premium for reliability.

Water is the underpriced twin constraint. Google used 8.1 billion gallons for cooling in 2024, up 28% year-over-year, while transformer replacement timelines now stretch to 2.5 years—helping create a new ~$10 billion insurance premium pool for digital infrastructure risk.

Within this HALO regime, Europe offers the cleanest geographic arbitrage. Nordic markets combine near-100% renewable power with free cooling that removes 30–40% of typical data center energy costs, driving structurally better PUE. The EU Grids Package mandates €584 billion of grid modernization with streamlined two-year permitting, providing a more predictable framework than U.S. FERC co-location rules. We highlight $ENEL, $IBE (Iberdrola), $EOAN (E.ON), and $FORTUM as a shrinking set of investable European HALO utilities whose strategic relevance is rising even as former champions like EDF and Vattenfall have moved under state control. These assets still trade at conventional utility multiples while becoming essential partners for hyperscalers locking in 10–15-year power purchase agreements.

The structural risks are real and require clear-eyed assessment. Our “Reverse Y2K” framework assigns a combined 35% probability that rapid progress in AI distillation and inference optimization strands a portion of today’s infrastructure build-out—echoing WorldCom’s late-1990s traffic assumptions. Yet the Jevons Paradox pushes in the opposite direction: as compute gets cheaper, usage tends to compound rather than shrink. The “Ghost GDP” effect—a roughly 14:1 ratio of consumer surplus to company revenue in AI adoption—implies that traditional P&L metrics systematically understate the value created by successful AI deployment, as much of the benefit shows up as quality and user surplus, not headline revenue.

For retail investors navigating the HALO era, portfolio construction should reflect the entire physical stack, not a single-sector proxy. Our playbook anchors conservative capital in North American and European regulated utilities with Net Debt below 2.5x and dividend yields of 3.5–4.5%, supported by verified 10–15-year hyperscaler PPAs. Alpha is generated via a barbell: long exposure to core HALO beneficiaries ($NVDA, $VST, $CEG) paired with short exposure to overvalued Tier-2 data center suppliers.

We track three real-time dashboards to monitor the regime: interconnection queue wait times (depth of physical bottlenecks), compute cost per unit (Reverse Y2K risk), and regional electricity prices in Texas, Northern Europe, and Virginia (where AI infrastructure can scale profitably). The world has turned heavy again—and for disciplined investors, that weight is less a drag than a durable foundation for long-term compounding.