December ended the year with volatility in the markets, particularly in the technology sector. Most of the price movements in the first two weeks appeared to be risk aversion \u2013 mainly linked to AI, with concerns around funding and project delays \u2013 rather than a deterioration in fundamentals.<\/p>\n\n\n\n
On the geopolitical side, the US administration announced that Nvidia will be allowed to sell its H200 chips to China. Despite extensive discussions around US\/Chinese chip restrictions, we see this as positive news for Nvidia and AMD (both holdings in the portfolio).<\/p>\n\n\n\n
Among individual stocks, Broadcom (one of the portfolio\u2019s top three positions) delivered strong results and guidance. More importantly, the company\u2019s AI-related semiconductor business, including custom AI chips and networking silicon for AI data centers \u2013 critical as workloads scale across larger GPU clusters, was a clear vision for the future. Broadcom announced that Anthropic has increased its orders from $10 billion to $21 billion and that a fifth customer will be added with revenue from fiscal 2026. The company now has a $73 billion order backlog for AI semiconductors to be delivered within 18 months. We believe this implies upwardly revised revenue guidance for 2026 and further upside for 2027 as Google TPU, Anthropic and OpenAI volumes increase. Broadcom remains a leader in both custom ASIC solutions (still in the early stages) and networking, with high visibility.<\/p>\n\n\n\n
Taiwan\u2019s November numbers confirmed the strength of the AI supply chain. TSMC (also a top-ranked company) reported a 25 % increase in sales year-over-year (29 % in USD), clearly beating its Q4 forecast. Supplier audits indicate that TSMC\u2019s advanced packaging (CoWoS) capacity will increase significantly through 2026 to meet growing demand \u2013 crucial for both Nvidia\u2019s GPUs and Google\u2019s TPUs. Taiwan\u2019s AI server ODMs hit new highs in November, with accelerated growth of 86 % year-over-year, driven by shipments of Nvidia Blackwell and Blackwell Ultra-based servers.<\/p>\n\n\n\n
Finally, GPT 5.2 was launched in December. Initial response \u2013 particularly around coding capabilities and improved logical reasoning \u2013 has been positive. Demand for tokens continues to grow as new use cases are discovered and more complex problems can be solved.<\/p>\n\n\n\n