AI Compute Shift: Druckenmiller Buys Amazon & Alphabet Over Broadcom

For AI builders and platform strategists, the Duquesne Family Office rotation is a signal about where value is accruing in the AI stack. Broadcom's custom silicon business helped it reach a 40 P/E, but Druckenmiller sold and bought Amazon and Alphabet, the hyperscalers that are increasingly designing their own accelerators and controlling model training and inference demand. The trade suggests the next AI returns may flow to those who own the workloads, not just the components.

Stanley Druckenmiller's Duquesne Family Office executed a consequential AI portfolio rotation during the second quarter of 2026, fully liquidating its Broadcom position while establishing what it described as relatively large stakes in Amazon and Alphabet. The moves, disclosed in the latest 13F filing and reported on August 29, 2026, mark a shift away from a key AI component supplier and toward the hyperscale cloud platforms that increasingly set the terms of AI demand. The decision carries particular weight because Druckenmiller has a record of spotting AI-driven market shifts early, though he has also acknowledged mistiming his exit from Nvidia.

“

Broadcom's custom silicon business helped it reach a 40 P/E, but Druckenmiller sold and bought Amazon and Alphabet, the hyperscalers that are increasingly designing their own accelerators and controlling model training and inference demand.

Druckenmiller first bought Nvidia for Duquesne Family Office in late 2022, an early bet on the AI chip boom that ultimately generated hundreds of millions of dollars in gains. However, he sold the entire Nvidia position in 2024, later admitting he exited too soon. That history frames the latest 13F as an attempt to reposition before the next phase of AI infrastructure spending matures. Rather than simply owning the dominant merchant AI chip supplier, Duquesne is now betting on the platforms that own the workloads, the cloud distribution, and increasingly the custom silicon that powers model training and inference.

Broadcom's exit is notable for its timing. The shares floated up to a price-to-earnings multiple of 40 at one point last quarter, compared with a forward P/E below 30 when Druckenmiller purchased the stock in the prior quarter. While valuation is not typically the primary driver of his decisions—he has said he invests based on major shifts in the economy and markets—the re-rating may have made the risk-reward less attractive. More importantly, the hyperscalers are taking more control of their own destinies in AI compute. Amazon and Alphabet are not merely customers for merchant chips; they operate massive cloud platforms, design custom accelerators, and decide how aggressively to deploy AI capacity.

The rotation into Amazon and Alphabet reflects a structural view of the AI supply chain. Broadcom is a critical custom ASIC partner for large technology companies, but its role remains that of a supplier negotiating with increasingly powerful buyers. Amazon, through AWS, and Alphabet, through Google Cloud, are closer to the end customer and have more direct exposure to AI revenue generation. They can bundle AI services, attract developers, and benefit from both training and inference demand. The sources indicate that Amazon and Alphabet were trading at historically low valuations at the time of Duquesne's purchases, which aligns with Druckenmiller's willingness to lean into securities when the market structure supports it.