How I Built a 57% Return in 11 Months — The Portfolio, The Thesis, The Mistakes

The screenshot tells one story. $157,034.46. A 57.23% annualised return across 11 months. Thirteen positions, a $100,000 starting book, and a final number that beats most professional fund managers for the same period.

But that number is the last thing we want to talk about.

Before we get to the wins, we need to start with the four positions that didn't work, because they're more instructive than anything SOXQ did over the same period. ADBE is down 16.64% on cost, a loss of $1,482. IAU has shed 12.04%, taking $1,171 off the table. SLV follows close behind at -11.94%, costing us $936. MSFT, the trade that felt most intellectually defensible at entry, sits at -7.84%, a $359 drag. That's $3,948 in aggregate losses across four positions, and every single one of them was a thesis error, not a market error.

The Framework Before The Trades

Three rules built this book. They aren't complicated. They're just consistently enforced.

Rule One: size by conviction times time horizon, not by ticker hype. SOXQ sits at 35% of the portfolio. That wasn't an accident, and it wasn't recklessness. It was the result of multiplying a high-conviction secular thesis — semiconductor infrastructure as the operating system of the AI economy — by a time horizon long enough to survive volatility. Smaller positions on lower-conviction names. No exceptions.

Rule Two: concentration is the alpha. The diversification instinct that finance textbooks drill into you is useful for protecting capital from random variance. It is actively harmful when you have a differentiated view. Spreading $100,000 across 30 names at 3% each because it feels safer is how you earn 12% in a year when the index does 22%. Concentration in your highest-conviction idea is where the excess return lives. SOXQ at 35% is not a mistake we need to explain. It is the entire point.

Rule Three: we don't time entries, we time thesis. We didn't buy NVDA at $134.84 because a chart pattern suggested a breakout. We bought it because the thesis was unambiguous — data center GPU demand was structurally underpriced by consensus after the DeepSeek-driven selloff, and we had a variant view on how fast the hyperscalers would re-accelerate capex. The price was a consequence of the thesis, not the reason for the trade.

The Big Win — SOXQ at $38.30

Six hundred shares of SOXQ at $38.30. $22,980 deployed at inception. That single position is now worth $54,708, a gain of $31,728, representing 59% of the entire portfolio's net P&L.

The thesis had two layers. The first layer was structural: the HBM supercycle that NVIDIA's Blackwell architecture was pulling forward was going to lift the entire semiconductor supply chain, not just the chip designers. Picking individual winners inside that chain — ASML, KLA, Lam Research, Broadcom — required a level of inside knowledge about customer concentration and product cycles that we didn't have an edge in. The second layer was mechanical: when a sector-level thesis is correct and the wave is wide enough, owning the index outperforms stock picking because you eliminate single-name risk without sacrificing the directional exposure.

The position hit +138% in 11 months. On the cost basis we put in, that's an annualised return north of 150%. It is the single best trade in the book, and the lesson is simple: when a secular tide is rising, own the tide.

The GOOGL Trade — $164.51 Into $396.78

Alphabet was the position we almost didn't build the right way. The initial entry was at $164.51, 20 shares, $3,290 deployed. The thesis was Google Cloud taking share in AI workloads, YouTube monetisation recovery, and optionality on Waymo. What we underweighted — and this is the honest admission — was Gemini's inflection.

We were wrong in timing, and lucky in outcome. Gemini's performance improvements in the second half of the year, combined with Google Cloud's deepening integration with enterprise AI stacks, hit before the search retention numbers showed meaningful degradation. By the time the market started pricing that in, the position was already 80% in the money. The position is up 141.19%. The lesson: when you own a company with three independent revenue drivers, you don't need all three to work at once.

NVDA +67% and MSFT -7.84% — The Same Thesis, Two Outcomes

Both positions were built on the AI infrastructure thesis. One returned +67.10%. The other lost 7.84%. The difference wasn't the thesis. It was the price the market had already assigned to the thesis at the time we bought.

NVDA entered the book at $134.84 after the post-DeepSeek panic selloff of early 2025, when retail sentiment had collapsed and consensus estimates for Blackwell demand were being revised down. The market was pricing in thesis failure. We disagreed. That gap between what the market believed and what we believed is where the 67% came from.

MSFT entered at $457.81 at peak Copilot euphoria. The multiple already reflected a scenario where everything went right. When Copilot monetisation came in slower than expected, the stock derated. Our thesis wasn't wrong. It was just already consensus. This is the single most important lesson in the book: a correct thesis is worth nothing if the market has already priced it in.

The Four Mistakes

ADBE -16.64%. The core thesis was that Adobe's creative software suite had a defensible moat. What we underweighted was the speed at which Canva, Figma, Runway ML, and OpenAI's image stack would converge on Adobe's core use cases simultaneously. Generative AI didn't just compete with Adobe's output — it commoditised the input. We should have exited at -10%.

IAU -12.04% and SLV -11.94%. These were hedges. The macro thesis was persistent inflation and dollar weakness lifting precious metals. Real yields surprised sharply to the upside and the thesis was wrong, full stop. In a portfolio with this much semiconductor concentration, the meaningful hedge is cash and duration, not metals.

The change we would make: a hard 10% trailing stop on every hedging position, no exceptions. Hedges that don't work should be small and fast.

The Positions That Quietly Compound

Nobody puts CSCO on a fintech TikTok. But the quiet compounders — CSCO at +87.03%, TSM at +113.94%, CVX at +37.02%, JPM at +11.92%, DB at +9.01%, VSS at +21.53% — are the reason this book has never given back a week of gains in a bad session.

CSCO's re-rating was the least-discussed alpha source in the portfolio. The market had written Cisco off as a legacy infrastructure company. What changed was the AI networking buildout: the same capex cycle driving GPU demand was also driving demand for high-speed Ethernet switches and campus AI infrastructure. CSCO was a beneficiary that the AI thematic crowd hadn't put on their list.

TSM was the clearest conviction carry. We own the manufacturer that makes every chip that matters. The 113.94% return reflects a market that spent 18 months pricing in geopolitical discount and then partially re-rated it out.

The Next 12 Months

Three changes are in process. First, we are trimming SOXQ on strength — taking 10% of the position off the table is risk management, not a thesis change. Second, we are looking at adding cybersecurity exposure via PANW, where the AI-driven attack surface expansion creates a structural demand floor that isn't yet consensus. Third, cash moves to 8-10% of the book from its current 0.6%. With the Fed's path uncertain and earnings estimates leaving no room for error, optionality has value again.

Close

A 57.23% return in 11 months is not purely skill. It is a correct thesis, executed with conviction, in a bull market environment that rewarded concentration in exactly the sectors we were concentrated in. The discipline that built this book is not the discipline of picking winners. It is the discipline of sizing correctly when you have conviction, cutting losses before they compound, and staying honest about when the market has already moved to where you thought you were going.

The next 11 months will require more of that discipline than the last 11 months ever asked for.

This research is for educational purposes only. Not investment advice. Past performance does not predict future results.