Stanley Druckenmiller fully exited Broadcom last quarter and bought Amazon and Alphabet, shifting his AI bets toward hyperscalers. Amazon trades at 20.5x forward earnings while Alphabet trades at 16.4x, both historically low multiples. The move reflects hyperscalers taking greater control of their chip supply chains.
Stanley Druckenmiller fully exited Broadcom and bought Amazon and Alphabet, shifting AI bets to hyperscalers at 20.5x and 16.4x forward earnings.
Amazon CEO Andy Jassy said the largest number of new chips entering its data centers this year will be its custom Trainium processors, not Nvidia GPUs or other off-the-shelf solutions. AlChip designed the newest Trainium generation and won the next-generation Trainium 4 contract.
Broadcom shares floated to a P/E of 40 at one point last quarter, after Druckenmiller bought them below a forward P/E of 30 in the prior quarter. Amazon's free cash flow turned negative at $7.6 billion over the trailing four quarters, while Alphabet posted negative free cash flow of $5.9 billion last quarter — its first as a public company.
Both companies hold large contracted revenue backlogs — Alphabet at $514 billion and Amazon at $496 billion — that they can convert as new data centers come online. The rotation shows the AI compute value chain is consolidating toward the hyperscalers that control chip design and deployment.
Druckenmiller, who bought Nvidia in late 2022 and made hundreds of millions before selling in 2024, has a history of rotating within the AI trade. His Duquesne Family Office added to Amazon and reestablished Alphabet as the hyperscalers raise capital through debt and new equity to fund data center expansion.
Alphabet said it is seeing strong demand for its TPUs and is selling TPU systems to select external customers. The company has long partnered with Broadcom for TPUs but recently signed a deal with Marvell for specialized inference chips, a sign that hyperscalers are diversifying their silicon suppliers.
AI customers are increasingly finding value in custom silicon for both training and inference, and the hyperscalers are showing greater willingness to source chips from multiple vendors. That trend favors the companies with the most control over which chips enter their data centers — a position Amazon and Alphabet now hold.
While other providers have stepped in to offer compute during hyperscaler capacity shortages, most AI inference will eventually run on hyperscaler servers where data and applications already live. Amazon and Alphabet should maintain high utilization rates on their custom silicon, producing strong operating results in the long run.
The depressed multiples reflect investor concern about the heavy capital spending. Amazon's forward P/E of 20.5 and Alphabet's 16.4 sit well below their historical averages, even as both companies hold contracted revenue that can be recognized once new data centers come online.
For holders, the move shows Druckenmiller sees more upside in the hyperscalers than in chip suppliers as AI demand matures. Investors will watch the next 13F filing for confirmation of the new positions and any further rotation within the AI trade.
This article is for informational purposes only and does not constitute investment advice.
