Post-AGI: the AI view of 2036

#2036 base case and scenarios

Base case. AI keeps diffusing through the economy. Capex on training and inference stays large enough that power and the grid remain the binding bottleneck. Inequality widens, so demand for scarcity goods, longevity and private wealth services rises. Geopolitics stays multipolar. Real rates stay stickier than in the 2010s. The horizon for the public book is about ten years.

We plan for overlapping alternatives, not a single path:

Scenario	What it looks like	How the book leans
Bumpy boom	Capability and spend keep rising, with at least one sharp correction when revenue lags capex	Core long book; dry powder to add on the correction
Bust and grind	Capex cuts; AGI slips later; debt and inflation dominate	T-bills and short TIPS as ballast; trim crowded AI names
Fast transition	Cognitive work automates early; labour markets and politics strain	Favour rails, scarcity and hedges over pure model owners
Taiwan or supply shock	Chip and foundry disruption	Taiwan capped by the country tilt; defence and non-Taiwan capacity matter more
Systemic break	Great-power war or a failure in market plumbing	Gold, cash, low leverage; act slowly

Several can arrive together. Writing them down forces honest kill criteria (below), not forecasts.

#Ranked thesis

Profit accrues first where physical scarcity meets a durable toll, not at every model lab. The ranking drives position size.

#1. Power

Generation, grid equipment, transmission build-out, firm clean power and nuclear fuel. Data-centre load turns kilowatts and interconnection queues into rent. Expressed through grid equipment and contractors (ETN, PWR), utilities (NEE), nuclear (CEG, CCJ) and copper and uranium ETFs (COPX, URA).

#2. Foundry

Leading-edge capacity and the lithography and custom-silicon stack that only a few firms can supply. AI accelerators are useless without wafers. Expressed through ASML, AVGO and TSM, with Taiwan exposure capped by the country tilt.

#3. Distribution and payment rails

Networks that move money and goods at scale collect a toll on commerce that grows with the economy the models create. Expressed through Visa (V).

#4. Compute

Hyperscalers, GPU franchises and the security layer around enterprise AI (MSFT, META, NVDA, CRWD, PANW, EQIX). Valuable, but further down the scarcity stack and often already priced for success, so we hold a measured sleeve rather than a concentrated bet on one model owner.

#5. Automation

Industrial controls, robotics and sensors as AI moves into the physical world (ROK). Slower to show up in orders, so we start small and add as factory adoption appears.

Supporting sleeves (commodities, logistics, towers and data-centre real estate, short-dated inflation-linked bonds and T-bills) back the same worldview without needing a sixth rank.

#Cross-check: The Commanding Heights

We checked the ranking against The Commanding Heights by Will Manidis and Joe Roberts (29 Oct 2026). Its core claim is that frontier models become commodities within months, so lasting wealth goes to the scarce positions intelligence must pass through, above all distribution and payment rails.

• Where we agree: model labs are not where durable profit sits, so compute ranks fourth; payment rails are a genuine toll, so Visa is a full-weight holding.
• Where we differ: for the next ten years we rank the physical bottlenecks (power and foundry) above rails, because capex has to clear them first.
• What we watch: if stablecoin dollar rails start taking volume from card networks, the rails sleeve needs a second expression.

#Model equities (reference)

A concentrated, single-book equity model that shows the ranking at full conviction. It is a reference for the thesis, not the family-office book. Full weights go only to the top toll in each of the first three ranks; the rest are starters.

Holding	Rank	Weight
ETN	1 Power	9%
ASML	2 Foundry	9%
V	3 Rails	9%
META	4 Compute	8%
MSFT	4 Compute	7%
NVDA	4 Compute	7%
TSM	2 Foundry (Taiwan-capped)	6%
NEE	1 Power	5%
AVGO	2 Foundry	5%
ROK	5 Automation	5%
PWR	1 Power	4%
T-bills (BIL)	Cash	26%

#What would change our mind

Falsifiers we watch (and log when they hit):

• Sustained hyperscaler capex guidance cuts, or an efficiency breakthrough that collapses the power-demand case.
• Foundry high-performance-computing revenue rolling over quarter after quarter, or a Taiwan shock that forces a permanent re-map of capacity.
• Payment-rail volume or take-rate turning against the network thesis.
• IT-outsourcing names showing durable AI-driven growth that kills the short thesis.
• Policy rates and term premium collapsing back into a lasting low-rate regime (hurts the long-bond short).
• A nuclear policy reversal or a clear uranium glut.
• A collapse in China luxury demand with no offset elsewhere.

Course corrections go into the changelog.

#Wealth and wellness sleeve

If AI cheapens cognitive labour, more of the surplus accrues to owners of capital and to things AI cannot mint. This sleeve tilts the book to that side of the ledger, alongside power and foundry rather than instead of them:

• Scarcity luxury and travel: Hermès, Richemont, Ferrari, Brunello Cucinelli, IHG.
• Longevity and fitness: Life Time.
• Metabolic health: Eli Lilly, Novo Nordisk.
• Private wealth platforms: Morgan Stanley, UBS, Julius Baer.
• Defence in a multipolar world: Lockheed Martin, RTX, Northrop Grumman.
• Nuclear power and fuel: Constellation, Cameco.
• Water: American Water Works, Xylem.

#Country tilt

Jurisdiction is part of risk, not an afterthought. We apply a 1.2× multiplier to places with strong property rights and AI-relevant infrastructure (US, UAE, Saudi Arabia, Singapore, Switzerland); treat core allies (Japan, Korea, UK, Denmark) at 1.0×; soft-underweight Europe at 0.8–0.9×; apply 0.7× to Taiwan and China (foundry access against geopolitical risk); and 0.5× to India, which we hold mainly through outsourcing shorts and selected longs. Taiwan foundry exposure is capped by this tilt rather than ignored.