How Our Portfolio Captures the AI Boom — While Staying Balanced

Two-thirds of our portfolio is plugged directly into the AI build-out. Ten of our 15 equal-weighted holdings — names like TSM, MU, STX, LITE — earn revenue from the same megatrend driving NVIDIA past a four-trillion-dollar market cap. The other five are deliberately not in the AI lane. That mix is the point: full participation in the boom, with ballast in case the boom takes a breather.

This post explains how our quant process arrives at that mix — and why we will not, on principle, let the AI weight drift toward 100%.

Where the Portfolio Touches AI

AI is not one trade. It is a stack — silicon at the bottom, then memory, then the optical fabric that connects everything, then power, then the application layer on top. Each layer has its own economics, its own competitive structure, and its own way of monetising the same demand curve. Our portfolio currently has names sitting across most of that stack:

Ten holdings, five layers, one demand curve. Each name was selected independently by the same systematic process — none was added because "we need AI exposure." They earned their slot on the same quality, momentum, and earnings-revision metrics every other position has to clear. The AI weighting is an emergent property of the screen, not a thematic decision.

Why Not 100% AI?

The temptation is obvious. NVIDIA is up multi-fold over the last two years. The semiconductor index has crushed the broad market. If you knew with certainty that the trend would continue, the right answer would be to lever up and concentrate. But that is not how risk-adjusted compounding works.

AI-exposed names carry meaningfully higher Beta than the average S&P 500 constituent. That is wonderful when sentiment runs in your favour. It is brutal when it doesn't. The 2022 semiconductor drawdown took the SOX index down roughly 35% in a year. Concentrated AI portfolios fared worse. A single negative earnings revision from one hyperscaler — a slower CapEx cadence, a shift in chip supplier, a regulatory headline — can clip 10% off the entire group in a session.

Our goal is not to maximise the upside of one theme. It is to outperform SPY and QQQ with equal or less Maximum drawdown. That objective forces us to keep diversifiers in the book.

The Five Non-AI Holdings, And Why They Are There

These five names give the portfolio different return drivers. When AI-exposed semis sell off on a hyperscaler CapEx scare, the miner, the bank, the defense name, and the refiner are usually doing something else — sometimes nothing, sometimes the opposite. That is the entire point of Diversification: lower portfolio volatility for the same expected return.

How the Quant System Picks the AI Names

We do not have an AI shortlist. We have a process. The fact that it currently surfaces ten AI-adjacent names is what the data is telling us right now — not what we set out to find.

Every position in the portfolio passes through our three-pillar process: Stock Universe → Optimizer → Risk Overlay. AI-beneficiary names are appearing at the top of pillar one because they are simultaneously posting strong earnings revisions, expanding margins, accelerating revenue growth, and exhibiting the price momentum that the screen rewards. They earn their place — they are not granted it.

Pillar 1 — Stock Universe

A wide net of large-cap U.S. equities filtered on quality and momentum: stable margins, positive earnings revisions, accelerating fundamentals, persistent price strength. AI build-out is currently the cleanest source of those factors anywhere in the market. So semiconductors, memory, optical, and power names dominate the qualifier list. If the next quarter belongs to industrials, healthcare, or financials, the universe will tilt that way instead. The screen does not care about the narrative.

Pillar 2 — Optimizer (Selection, Not Sizing)

From the qualifier list, the Optimizer chooses which 15 to 30 names actually enter the portfolio. Crucially: the Optimizer selects, it does not size. Every position is held at the same weight — currently 6.67% per name across 15 holdings. We do not let any single AI thesis become 15% or 20% of the book on the way up. That is what blows portfolios up when the cycle turns. See our piece on portfolio construction for why 15-30 equal-weighted names is the sweet spot.

Pillar 3 — Risk Overlay

Whatever the underlying portfolio looks like, the daily risk overlay decides how much of it we run. When 23 named indicators flag deterioration, the overlay scales equity exposure down — independent of whether the underlying names are AI darlings or defensive utilities. The overlay does not know it is "AI exposure" it is dialling down. It is dialling down market exposure. That is exactly the point: know when to accelerate, know when to brake.

The System Adapts — That Is Its Job

A natural worry: "What if AI cools off, or the leadership rotates? Are you stuck holding old winners?"

No. The portfolio is rebuilt monthly. At each rebalance, every name has to re-qualify on the same factors. Names whose earnings revisions decelerate, whose price momentum fades, or whose margins compress get dropped. New qualifiers — from any sector — replace them. The portfolio that exits the year will look different from the one that entered it, and the AI weighting will move with the data, not with the headlines.

This is also why the current AI tilt is not a bet. It is a snapshot. If the next twelve months reward industrials over semis, the screen will rotate — without anyone having to "call the top." Our most recent monthly review shows what that rotation discipline looks like in practice.

Two Holdings, Two Deep Dives

For the curious, two of our current AI-adjacent positions have full deep dives in the cockpit:

• Seagate (STX) Deep Dive — why HDDs are not "dying tech" but the storage backbone of every hyperscaler's training-data layer.
• SanDisk (SNDK) Deep Dive — high-capacity NAND as the unsung beneficiary of inference workloads.

Both names cleared our screen on standalone fundamentals. The AI tailwind is the bonus, not the thesis.

“We participate in the AI boom because the data tells us to — not because we want to. And we keep five non-AI names in the book because the same discipline that put us in semis also tells us to never bet the whole portfolio on one theme.”

*Past performance is not an indication of future results. Your capital is at risk. Holdings are accurate as of the publication date and rotate at each monthly rebalance.*

Weekly investment newsletter, every Sunday — positioning, performance vs SPY/QQQ, and what we're watching next. No spam.