Effective date: 31/07/2026In July, global equity markets moved in a more volatile environment, marked by escalating tensions between the United States and Iran, with second-round effects on energy prices, inflation expectations, and market sentiment. The Fed’s new communication policy, namely its intention to no longer provide forward guidance, also slightly fueled volatility. At the same time, the AI theme underwent a significant adjustment, as valuation levels and positioning reached extreme levels. The slightest negative news then triggered profit-taking. The market was particularly concerned about the emergence of a new Chinese open-weight model with 2.8 trillion parameters, Kimi K3 from Moonshot AI, thus reviving questions raised during the “DeepSeek” episode: if high-performing, lower-cost open-weight models are quickly adopted, the market could challenge current assumptions regarding monetization, return on invested capital, and continued massive AI infrastructure spending, especially since OpenAI and Anthropic occupy a central place in current capex expectations. These factors fueled a rotation from momentum positions and AI-related stocks to value and defensive sectors for most of the month, until the publication of Microsoft and Amazon’s results on Thursday, July 30. These releases, which confirmed not only good profitability from AI infrastructure spending but also continued robust future demand, marked a true inflection point for the technology sector. Initially, the beneficiaries of these expenditures, notably semiconductors, rebounded strongly; subsequently, cloud providers began to outperform and catch up on their year-to-date lag. Furthermore, half-yearly earnings releases proved extremely solid, with upward revisions to expected results, and resilient in the face of disruptions related to the Middle East conflict and inflationary pressures. Nevertheless, over the month, the AI theme underperformed via semiconductors and electrification companies, while energy (supported by the rise in oil prices amid renewed tensions between the United States and Iran) and financials (driven by a favorable market and economic environment as well as rising long-term rates) outperformed.
Correction phases are often the best way to test our convictions. The strategy posted, at the month’s low point, a decline of around ten percent, before rebounding significantly during the last two sessions. The profit-taking carried out in June helped preserve part of the outperformance accumulated since the beginning of the year, even though this effect was partly neutralized by a strong divergence in behavior between US-listed semiconductor stocks and their Asian counterparts. The relative underperformance compared to the index was concentrated on two elements. On the one hand, TSMC ADR underperformed the Taiwan-listed line, due to a compression of the valuation premium associated with the US listing. On the other hand, Micron, on which the exposure was more limited, outperformed SK Hynix and Samsung, overweighted in the portfolio, in a context of marked correction in the KOSPI, accentuated by positioning that became too consensual and high use of leverage, which amplified the decline of Korean sector stocks. At the individual position level, Amkor came under pressure after third-quarter revenue guidance fell short of expectations, against a backdrop of lower-than-anticipated Android volumes. At the same time, the company nevertheless announced a $1.5 billion financing agreement with Nvidia related to its Arizona site. SK Hynix also reported a quarter below expectations, mainly due to a delay in the HBM4 delivery schedule and a larger share of memory volumes covered by long-term contracts. At the end of the month, we began to gradually redeploy capital into certain targeted pockets of semiconductors, after a marked correction had brought some stocks down nearly -30%.
Management conviction remains unchanged: even if the value chain of models may evolve in favor of application layers, agents, and vertical software, the structural needs for computing, memory, and networks remain intact.