Work &
Projects

The Backdrop

Two major geopolitical events within four years of each other have now disrupted global oil supply. Russia's invasion of Ukraine sent Brent from around $80 to over $130 in a matter of weeks in 2022. Then the US-Iran conflict closed the Strait of Hormuz, which carries roughly 20% of global oil and gas, pushing Brent from $70 to a peak of around $118 when the Strait first closed earlier this year, and currently sitting in the $85 to $90 range. Each time, the countries that fared best were the ones that had built up strategic stockpiles ahead of the disruption. China is the clearest example. Its reserve capacity provided a meaningful price buffer that countries without storage simply could not replicate.

The pattern is becoming hard to ignore. Geopolitical disruptions to oil supply are not one-off events. They are recurring. And each time one happens, the case for having your own storage infrastructure gets stronger. The thesis is that governments and energy companies around the world will increasingly invest in building out strategic petroleum reserves and oil storage capacity, not as a luxury but as a form of energy security insurance.

I am currently in the process of identifying the right stock to express this view, looking specifically at storage operators and infrastructure companies that would be the direct beneficiaries of that investment cycle. A full stock pitch with valuation and buy recommendation will follow.

What They Do
Infineon is the world's number one power semiconductor company. They make chips that control and convert electricity, the components that sit inside EV powertrains, industrial motors, and the power supply chains of AI data centres alongside every Nvidia chip rack.

Why It's Mispriced
The market still prices Infineon as a slow-moving European car chip company. That is the wrong frame. Management has explicitly guided AI data centre revenues of €1.5 billion this year rising to €2.5 billion next year, 67% growth in that business alone. They upgraded their full year outlook from moderate to significant growth, raised their cash flow target to €1.65 billion, and increased AI manufacturing investment to €2.7 billion because demand is outstripping what they can currently produce. Gartner independently named Infineon the company to beat in AI data centre power semiconductors. The order backlog is €25 billion, 25% higher than a year ago and still growing. Yet the stock is down 28% from its June high. That gap between what the business is doing and where the stock is trading is the entry point.

Catalysts

August 5 Q3 earnings, the main event. Infineon guided Q3 revenue at around €4.1 billion, 8% higher than the €3.812 billion they reported in Q2, with profit margins in the high-teens. If the actual numbers beat those targets, analysts upgrade their forecasts and the stock moves on the day.

AI data centre revenue finally showing up in the numbers. Management has told the market to expect €1.5 billion of AI revenue this year, but that number has never appeared as its own line in the reported results. August 5 is the first real chance to see it broken out explicitly. The moment it does, investors stop thinking of Infineon as a European car chip company and start thinking of it as an AI infrastructure company, and those two types of business are valued very differently.

Guidance could get raised again. Infineon already upgraded their full year outlook to more than €16 billion at around 20% profit margins. If Q3 beats and the second half looks even better than expected, a second upgrade on the same call is possible. The last time they raised guidance, Berenberg moved their target to €100, JPMorgan to €96, and Susquehanna to €100.

The order backlog shows the demand is real. The backlog stands at €25 billion, 25% higher than a year ago and still growing. These are confirmed orders from customers who have already committed, not management optimism. August 5 is the chance to show those orders are converting into actual revenue.

Key Risks

Automotive weakness. Infineon's biggest segment is still struggling. Car manufacturers and their suppliers are placing orders cautiously and only committing to short-term volumes rather than building up stock. In Q2, the gross margin actually fell because demand for high-voltage electric vehicle components was weaker than expected and the company had to restructure that part of the business. If that weakness drags on, it eats into the overall recovery story.

Over-reliance on AI. AI data centre revenues are growing fast but they are still a relatively small part of the business today, around 9% of total sales in FY2026, rising to around 15% by FY2027. As that grows, if hyperscalers like Microsoft or Amazon slow down their data centre spending, Infineon starts to feel it in a way it would not have a few years ago.

Japanese SiC competition. There is a potential alliance between Rohm, Toshiba, and Mitsubishi that would directly target Infineon's position in silicon carbide chips for cars. If that comes together it creates a well-funded domestic competitor in Japan that could take market share in one of Infineon's most strategically important product areas.

US tariffs. Infineon makes its chips in Germany and does not have large manufacturing plants in the US. Under the current Trump administration, if tariffs are extended to cover more types of semiconductors, American-made chips would have a cost advantage over Infineon's European-made products in the US market.

Conclusion

Markowitz Portfolio Optimisation Tool

Developed a Markowitz portfolio optimisation engine in Python across 15 stocks and 5 sectors. The engine derives expected returns and identifies optimal portfolio allocations under multiple constraint scenarios.

How I built and refined it

Monte Carlo to SLSQP — I started by running 50,000 Monte Carlo simulations, generating random portfolio weight combinations and picking the best one. I quickly realised the problem. No matter how many simulations I ran, I was still sampling randomly and could easily be missing thousands of better portfolios I never landed on. This is why I introduced SLSQP optimisation, which solves the problem mathematically rather than guessing, guaranteeing the true optimal portfolio.

Historical returns to CAPM blend — My initial expected return estimates were based purely on historical data, meaning the model was entirely backwards looking. I wanted something more forward looking, which is why I implemented CAPM. However, CAPM relies on assumptions that don't always hold in practice, so I decided to take a 50/50 blend of historical and CAPM estimates to balance both approaches.

Weight and sector caps — Looking at the output from an investor's perspective, I noticed the optimiser was concentrating too heavily into single assets and sectors. To make the portfolio more realistic and properly diversified, I introduced a 30% weight cap per asset and a 40% sector cap, preventing any one position or sector from dominating the allocation.

Short selling — My model was initially long only, meaning every asset had to have a positive weight. I extended it to allow short positions of up to 15% per asset, which meant the optimiser could take a negative position in a stock it expected to underperform. This made the model more realistic and gave the optimiser more flexibility to improve the Sharpe ratio.

Ledoit-Wolf covariance shrinkage — When I looked at the covariance matrix the model was producing, some of the correlation estimates between stocks looked unrealistically extreme. These extreme estimates were feeding into the optimiser and affecting the weight allocations in ways that did not make practical sense. I researched solutions and implemented Ledoit-Wolf shrinkage, a technique that pulls extreme correlation estimates back toward a more realistic average, producing a more stable and reliable covariance matrix.