Billionaire fund manager Stanley Druckenmiller, head of Duquesne Capital, made several notable changes to his portfolio in the second quarter, closing positions in Micron (MU) and Intel (INTC) while establishing new stakes in Advanced Micro Devices (AMD) and Alphabet (GOOGL/GOOG). As one of the world’s most closely watched investors, Druckenmiller’s moves often draw significant market attention.
Druckenmiller booked profits in both Micron and Intel during the year. Micron has benefited from a memory supercycle, with supply-demand imbalances driving sharp increases in memory prices and boosting revenue and gross margins. Although the memory market has historically been highly cyclical, the large-scale build-out of AI data center infrastructure is altering market dynamics. Graphics processing units and other AI accelerators require substantial amounts of high-bandwidth memory to perform optimally, and multiple factors are expected to keep HBM supply constrained for years. Meanwhile, as the three major memory makers concentrate on expanding HBM capacity, prices across the broader DRAM market have risen significantly. Micron currently trades at a forward price-to-earnings ratio of just over 6.5.
Intel presents a different picture. The company has seen rising demand for data center CPUs as hyperscalers and neocloud providers prepare for increased use of AI agents. However, that demand is viewed as stemming more from external conditions than from a fundamental improvement in Intel’s business. Its foundry operations continue to post losses, weighing on overall results, and the stock is no longer considered inexpensive.
Although Druckenmiller sold Intel, he remained in the server CPU space by adding a position in AMD. AMD currently leads the CPU market and has been steadily taking share from Intel. Its high-core-count CPUs are designed for agentic AI workloads, and the company expects this market to reach $220 billion in the coming years, with a target of capturing more than half of it. AMD also has an opportunity in AI inference, a market widely expected to grow larger than AI training. The company’s chiplet design allows processors to integrate more memory, which is particularly beneficial for inference workloads. AMD has also partnered with wafer-scale engine company Cerebras to offer a disaggregated system aimed at inference, and its recent acquisitions of chipmaker Taalas and memory optimization firm MEXT have strengthened its position in this segment.
Druckenmiller added Alphabet to his top ten holdings in the second quarter. Alphabet’s cloud business continues to expand, and its custom tensor processing units provide a cost advantage for inference workloads. The chips also lower the cost of training AI models, which are then integrated into products including Google Search. Alphabet benefits from significant distribution advantages through the Chrome browser, the Android operating system, and a search revenue-sharing agreement with Apple that makes Google the default search engine on Apple devices. Its global digital advertising network also supports monetization of consumer AI relative to many large language model developers. Alphabet is seen as having one of the most complete AI technology stacks among major technology companies.
Overall, the second-quarter portfolio changes suggest a partial shift away from traditional memory and CPU names toward companies with deeper exposure to AI inference and agentic AI.
