You are mistaken if you think George Soros is still the macro hedge fund of the 1990s. The 13F filed on August 15 reveals a portfolio that looks more like a sector rotation ETF than a global macro play. New positions in Nebius, DigitalBridge, American Electric Power, Taylor Morrison Home, and Apogee Therapeutics, while clearing Salesforce, GlobalFoundries, and others. This is not a bet on a single narrative; it is a structural shift from 'software and subsidized chips' to 'physical infrastructure and AI compute.' The ledger remembers what the mempool forgets—the filing is a snapshot, not a strategy, but the direction is unmistakable.
Context: The 13F filing, required by the SEC for any manager with over $100 million in U.S. equities, offers a 45-day-old peek into portfolio holdings as of June 30, 2025. Soros Fund Management, now under Alex Soros, manages roughly $6.5 billion in public equities, a far cry from its heyday as a macro powerhouse. Still, the rotation carries signal value. The timing is critical: Nebius Group (NBIS) only relisted on Nasdaq in October 2024, and the Q2 2025 snapshot likely reflects a deliberate pivot away from hype-driven narratives toward hard assets. This is not a macro hedge fund making a directional bet on the dollar; it is a family office executing a sector-level reallocation that speaks to the ongoing AI infrastructure buildout, housing shortage, and inflation stickiness.
Core: Let’s tear down the new positions one by one. The most telling is Nebius (NBIS), a GPU cloud provider that combines AI compute with a footprint in Europe and the Middle East. Soros is betting that the AI inference wave will require more than just data center shells—it needs actual compute capacity. Nebius operates its own HPC clusters, and its revenue model is tied to GPU utilization, not just real estate. The risk: founder ties to Russia and dependency on Nvidia supply. Yet the bet is explicit: compute is the new energy. Next, DigitalBridge (DBRG) is a digital infrastructure REIT, owning data centers, cell towers, and fiber networks. This is a direct play on the physical layer of AI—the wiring, cooling, and power. Soros is not buying the AI application layer; he is buying the picks and shovels. American Electric Power (AEP) is a regulated utility, but its growth story is increasingly tied to data center power demand. The U.S. grid is underprepared for the load growth from AI—AEP’s service territory in the Midwest and Appalachia sees new hyperscale projects. The dividend yield is a bonus, but the real thesis is that electricity demand growth will outpace the grid’s capacity expansion, driving up rates and profits. Taylor Morrison Home (TMHC) is a homebuilder. In a high-rate environment, this seems contrarian. But the housing shortage is structural: the U.S. is underbuilt by at least 1.5 million homes. TMHC focuses on entry-level and move-up buyers, and its land-light model reduces risk. Soros is betting that the Fed will cut rates by year-end, or that the housing deficit will overwhelm rate sensitivity. Apogee Therapeutics (APGE) is a biotech focused on antibodies for inflammatory diseases. This is a high-risk, high-reward play on clinical trial data. The clearings are equally instructive. Salesforce (CRM) was dumped—the company’s AI agent pivot (Agentforce) is already priced into a 50x P/E on uncertain TAM. The market is paying for narrative, not delivery. GlobalFoundries (GFS) was cleared despite being a CHIPS Act beneficiary. This tells me that Soros values market competitiveness over government subsidies. GFS is stuck in mature nodes, losing share to TSMC and Samsung. The subsidy is a lifeline, not a moat. Truth is a derivative of transparent data—the 13F shows what Soros sold, not why, but the pattern is clear: old tech, old subsidies, and overhyped software are out. Physical infrastructure, compute, and energy are in.
Contrarian: What the bulls got right? The bulls on Salesforce argue that its enterprise relationships and data moat give it a unique advantage in AI. They might point to the 25% revenue growth from AI-related products in Q1 2025. But Soros’s exit suggests that growth is already fully discounted, and the market is ignoring the risk of disruption from native AI startups. Similarly, the bulls on GlobalFoundries highlight the $1.5 billion in CHIPS Act grants and the national security rationale for keeping fabs in the U.S. Yet Soros is betting that the subsidy will not compensate for the lack of leading-edge process technology. The contrarian angle here is that Soros’s new positions in homebuilders and utilities are actually defensive, not aggressive. They are hedges against a slowdown—utilities provide stable cash flows, homebuilders benefit from rate cuts—while the AI infrastructure bets are the growth engine. The illusion persists until the liquidity dries—the Salesforce narrative may be a form of wash trading in sentiment, and Soros is calling the bluff.
Takeaway: The Q3 13F, due November 14, 2025, will be the real test. If Soros doubles down on Nebius and AEP, this is a multi-year structural shift. If he sells them, it was a tactical trade. The Fed’s September and December FOMC meetings will determine the fate of the homebuilder and utility bets. GPU supply from Nvidia to Nebius is the single most important data point for the infrastructure thesis. 13F is not strategy, it is merely a snapshot—but this snapshot captures a rare moment of capital rotation from the virtual to the physical. Will the market follow Soros into the substations, or will it stay stuck in the software narrative?
