Soros Fund Management’s Q2 2025 13F hit the SEC filing system on August 15. The dataset is clean: five new positions, five full exits.
The metadata is the story. New buys: Nebius Group (NBIS), DigitalBridge (DBRG), Apogee Therapeutics (APGE), Taylor Morrison Home (TMHC), and American Electric Power (AEP). Exits: Salesforce (CRM), GlobalFoundries (GFS), and three others. The net effect is a sector rotation from legacy software and mature chip manufacturing into AI infrastructure, digital real estate, biotech, and housing.
Data doesn’t care about your timeline. The filing is a snapshot as of June 30, 2025. By the time you read it, Soros may have already flipped. But the directional signal is still informative. Over the past 16 years of watching institutional flows, I’ve learned that 13F reports are noisy, but they become clean when you isolate the sector-level shifts. This is one of those clean signals.
Context: The 13F Limitations and the Soros Shift
Before diving into the on-chain (or off-chain, in this case) evidence, let’s level-set. The 13F covers only U.S. long equity positions above $200 million. It excludes derivatives, shorts, and bonds. Soros Fund Management is now a family office under Alex Soros, but the name still carries market weight. The filing is a 45-day-old snapshot.
I’ve run similar analyses on institutional flows for years. The typical mistake is treating the 13F as a trading signal. It’s not. It’s a positioning statement. The real value lies in the
Core: The Evidence Chain
Let’s walk through each new position and the implied bet.
Nebius Group (NBIS) – This is the anchor. NBIS re-listed on Nasdaq in October 2024 after a complex restructuring. The company operates a massive GPU cloud for AI training and inference. Soros built a new position here. The logic: AI compute demand is still in its early innings. NVIDIA’s H200 and B200 supply constraints are well known, but the real bottleneck is cloud capacity. NBIS, along with CoreWeave, is a pure-play AI infrastructure provider. The filing shows a 100% new position, meaning it wasn’t in the prior quarter. This is a conviction bet.
DigitalBridge (DBRG) – A digital infrastructure REIT focused on data centers, cell towers, and fiber. Soros adding DBRG alongside NBIS shows a thematic bet on the physical layer of the AI stack. Not just compute, but the real estate that powers it. DBRG’s AUM has grown 40% year-over-year as hyperscalers lease data center capacity. The 13F confirms this is a deliberate infrastructure play, not a speculative tech bet.
Apogee Therapeutics (APGE) – A biotech with a pipeline in respiratory and inflammatory diseases. This is a classic Soros move: buying a high-risk, high-reward biotech with near-term catalysts. The Q2 2025 filing shows APGE as a new position at a time when the stock was trading at a discount to its cash value. The contrarian angle here is that biotech IPOs have been underperforming, but Soros is picking a specific clinical-stage compound. This is a stock-picking signal, not a sector call.
Taylor Morrison Home (TMHC) – A top-10 U.S. homebuilder. Soros buying TMHC signals a bet on housing supply shortage and a soft landing. The U.S. is underbuilt by 1.5 million units. Even with mortgage rates at 6.5%, demand remains strong due to demographics. TMHC’s backlog-to-revenue ratio suggests pricing power. The 13F shows a new position at a time when homebuilder stocks were off their highs. This is a macro bet on rates not rising further.
American Electric Power (AEP) – A utility that generates and distributes electricity. Soros buying AEP is a two-for-one bet: First, AI data centers will drive a structural increase in electricity demand. The EIA projects U.S. power demand growth will double from 0.5% to 1% annually over the next decade. AEP serves regions with high data center concentration. Second, utilities are bond proxies. If rates fall, AEP’s valuation expands. The filing shows a new position at a time when AEP was trading at 18x earnings, a reasonable entry point.
Now the exits: Salesforce (CRM) and GlobalFoundries (GFS). CRM is a legacy enterprise software giant. Soros exiting CRM is a statement that AI-native tools are eating CRM’s lunch. Salesforce’s Agentforce launch hasn’t moved the needle on revenue growth. The 13F shows a 100% reduction, meaning Soros sold all shares. GFS is a mature chip foundry that benefited from the CHIPS Act subsidies. But the company’s revenue is flat, and its advanced node roadmap is behind TSMC. Soros sold GFS to reallocate into AI infrastructure. This is a clear capital rotation from “subsidized legacy” to “market-driven growth.”
Contrarian: The Blind Spots
Correlation does not equal causation. The 13F doesn’t show derivative positions. Soros could have bought put options on the S&P 500 or sold call spreads against these positions. The filing only shows long equity. The new positions could be part of a paired trade where the short side (not visible) hedges the risk.
Also, the market has already priced in the AI infrastructure narrative. NBIS is up 120% from its IPO price. Soros’s buy could be a momentum follow, not a value discovery. The 13F doesn’t reveal the entry price. If he bought at the June peak, he’s underwater now. The follow-through in Q3 will tell the real story.
Another blind spot: The exits from CRM and GFS could be tax-loss harvesting or portfolio rebalancing, not a directional bet. The 13F doesn’t show the reasoning. The data detective must interpret the pattern, but never assume intent.
Takeaway: The Next Signal
The Q3 2025 13F, due mid-November, will be the confirmation or denial. If Soros adds to NBIS, DBRG, and AEP, the AI infrastructure rotation is a long-term thesis. If he sells, it was a trade. The smart money is watching the same data.
Follow the metadata, not the mood. The 13F is a lagging indicator, but the sector rotation is real. The question is whether the market has already priced it in. The answer lies in the next filing.
Data doesn’t care about your timeline. Neither does Soros.