The front-runners are already inside the block. George Soros’s Q2 2025 13F filing, published with the standard 45-day delay on August 15, shows a clean rotation: exit Salesforce and GlobalFoundries, enter Nebius, DigitalBridge, Taylor Morrison, American Electric Power, and Apogee Therapeutics. The market reads this as a bet on AI infrastructure, housing, and inflation-proof utilities. But the filing is a snapshot, not a strategy. The real trade—the one that matters—was likely unwound before the ink dried on the SEC submission.
Context: The 13F Trap
Every quarter, institutional managers with over $100 million in US equities must file a 13F within 45 days of the quarter’s end. Soros Fund Management, controlled by Alex Soros since 2023, reported a $65 billion portfolio as of Q1 2025. The Q2 filing, released August 15, 2025 (based on Nebius’s October 2024 relisting and standard reporting timelines), shows five new positions and five closures. The closures include Salesforce (CRM) and GlobalFoundries (GFS)—both former consensus picks. The newcomers: Nebius (NBIS), DigitalBridge (DBRG), Taylor Morrison Home (TMHC), American Electric Power (AEP), and Apogee Therapeutics (APGE).
The fund has transitioned from a macro hedge fund to a family office. Its signal value has diminished. Yet the market still treats it as a grail. Based on my experience auditing DeFi protocols, I’ve seen the same pattern: capital rotates into infrastructure before the narrative catches up, and by the time the 13F is filed, the early movers have already repositioned. Code does not lie, but it does hide—and so does the 13F.
Core: Deconstructing the Rotations
Let’s dissect each new position through the lens of a security auditor—not as a financial analyst, but as someone who looks for structural vulnerabilities and hidden dependencies.
Nebius (NBIS) – AI Cloud Infrastructure. Nebius operates a GPU cloud service, competing with CoreWeave and Lambda. The thesis: AI inference demand is exploding, and capacity is scarce. But Nebius is a 2024 IPO with a complex history (spun off from Yandex). Its balance sheet is weak, and its dependency on NVIDIA’s H100/GB200 supply chain is a single point of failure. Soros’s bet is a binary one: either NVIDIA delivers on time, or Nebius’s revenue collapses. The filing does not reveal the size of the position. A small, symbolic buy means nothing. A large allocation would be a vote of confidence in the AI supply chain. The market, however, treats it as a directional signal.
DigitalBridge (DBRG) – Digital Infrastructure REIT. This is a safer bet: it owns data centers and cell towers. AI-driven demand for colocation is structural. DBRG’s AUM growth is a leading indicator. The 13F does not tell us the entry price. If Soros bought at Q2’s peak, the position is underwater. If he bought during a dip, it’s a different story. The filing hides the cost basis.
Taylor Morrison Home (TMHC) – Homebuilder. This is a counter-cyclical bet. TMHC benefits from a housing shortage and a potential rate cut. But Soros also bought AEP, a utility that profits from inflation and rate stability. The two positions are in tension: rate cuts help TMHC, but hurt AEP’s yield. This is not a monolithic macro trade; it’s a collection of alpha bets. The market often mistakes a portfolio for a thesis.
American Electric Power (AEP) – Regulated Utility. The logic: AI data centers will drive a 10-20% increase in US electricity demand by 2030. AEP operates in high-growth regions (Texas, Midwest). But utilities are rate-regulated, and their profits are capped. The real upside is in independent power producers (VST, CEG) that can sell into merchant markets. Soros choosing AEP over VST suggests a preference for stability over volatility—or a lack of understanding of the power market. The 13F does not reveal whether he hedged with a short on VST.
Apogee Therapeutics (APGE) – Biotech. This is a binary bet on a clinical-stage drug. Apogee is developing an antibody for asthma and chronic obstructive pulmonary disease. The 13F filing tells us nothing about the probability of success. It’s a lottery ticket, not a position.
The closures are equally telling. Soros sold Salesforce (CRM) and GlobalFoundries (GFS). CRM is a legacy software company facing AI-native disruption. GFS is a mature-node foundry that received CHIPS Act subsidies but lacks the technology to compete with TSMC. The market interpreted this as a bet against 'old tech.' But the filing does not show the timing: Soros could have sold CRM in April and bought it back in July. The 13F is a snapshot, not a diary.
Contrarian: The Blind Spots
The 13F is a dangerous tool. It shows only long equity positions. It does not include short positions, options, swaps, or futures. Soros could be shorting the S&P 500 while buying these stocks—or shorting Nebius itself via puts. The filing does not reveal the hedge. The best audit is the one you never see. The true signal is not in the 13F, but in the derivatives market. If Soros was buying put protection on AI stocks, the long positions are a distraction.
Furthermore, the 45-day delay means the market is reacting to stale data. Between June 30 and August 15, the AI narrative shifted. The ‘DeepSeek’ moment in early 2025 triggered a sell-off in GPU-related stocks. Nebius dropped 30% in July. Soros may have already exited the position. The retail investors who buy NBIS based on this filing are buying into a narrative that expired a month ago.
Takeaway: The Real Signal
Ignore the stock picks. The real insight is the rotation pattern: from old tech to new infrastructure. This mirrors the shift in crypto from layer-1 speculation to layer-2 scalability and infrastructure tokens. The market is repricing assets not based on current earnings, but on future utility. Soros’s 13F is a lagging indicator of that trend. The front-runners are already inside the block. The next 13F—due in mid-November 2025—will reveal whether Nebius was a conviction bet or a fleeting trade. Until then, treat the filing as noise, not signal.