Bricks Beat Bytes

By Filippo Cagliero · cf0 Chief Strategy Officer
Published 8 May 2026 · cf0research.substack.com

This is a real cf0 research note, produced inside Lab by cf0’s CSO and published on Substack. Filings, macro data, and market series all flowed through one thread; every figure carries a citation back to its source.

Conviction statement. “The physical infrastructure — power, cooling, aerospace, defense, rail — is underowned, undervalued, and entering a multi-year capex cycle with sovereign backing.”

The HALO thesis in one chart

A structural rotation toward companies whose moats rest on heavy physical assets — assets that take decades to replicate and cannot be coded away. The risk-return scatter below frames the case: HALO names cluster top-left (better risk-adjusted return than the AI basket), while a handful of AI names sit further right on the volatility axis.

Risk-return scatter — HALO vs AI vs Market

For portfolio managers, the basket offers valuation relief from stretched AI names while capturing the infrastructure tailwinds that underpin long-term productivity. Goldman Sachs’ asset-heavy basket has outperformed asset-light peers by ~35% since the start of 2025, with the spread accelerating ~25 percentage points since November 2025.

The six themes powering it

The note decomposes the thesis into six themes, each with a near-term momentum score and a 2027 durability read. Concentration risk is flagged where a theme leans heavily on a single name.

HALO themes — momentum / durability / concentration risk table

The macro backdrop favours all six:

900.6B in authorised US defence spending for 2026, NATO's 5%-of-GDP-by-2035 commitment (~

2.7T cumulative), and ~$1.5T of hyperscaler AI infrastructure capex through 2026 — which itself flows to power generation, cooling, land, and steel rather than software.

How each name maps across themes

The heatmap below scores fourteen HALO names against the six themes — useful for spotting where exposure concentrates (defence platforms is heavy across LMT / NOC / KTOS / AVAV / RTX / HII / GEA / LOAR / TDG) and where it doesn’t (energy infra has one name; logistics & rail has three).

HALO thematic exposure matrix — 14 names across 6 themes

What this used

This note used every part of cf0:

SEC + global filings — 10-Ks and 10-Qs for each HALO and AI name, pulled and parsed.
Lab — the surface where the thesis was framed; one thread held the macro setup, the screener output, the per-name tearsheets, the scenario sensitivity, and the chart inputs.
Skills — recurring research moves bound to a slug, so the same scenario / sensitivity passes can be re-run on a new ticker in one command.
Reports — the polished output exported as a branded PDF before publication to Substack.

Every figure cites its source. Macro spending numbers trace to the NDAA appropriations text and NATO Hague Summit Declaration; ETF mechanics to the Roundhill prospectus on SEC EDGAR; per-name returns to public market data.

Read the full note