The Q2 13F from Soros Fund Management landed on August 15. Most traders saw a list of new names. I saw a clean sector rotation from old tech to physical infrastructure. The model didn't flinch; it's tracing the gas leaks before the code compiles.
Context: The 13F Trap
Every quarter, the market waits for Soros' filings like a scripture. But the 13F is a snapshot with a 45-day delay. The positions are as of June 30, 2025. By August 15, the fund could have already reversed half of them. I've been through this cycle since 2017—Golem audit, Uniswap V2 liquidity mining, the LUNA autopsy. The pattern is always the same: noise traders chase the headlines, while the real edge is in the exits.
Soros' Q2 filing shows five new longs: Nebius (NBIS), DigitalBridge (DBRG), Taylor Morrison Home (TMHC), American Electric Power (AEP), and Apogee Therapeutics (APGE). He also exited five: Salesforce (CRM), GlobalFoundries (GFS), and a few others. The market chatter focused on the AI infrastructure narrative—Nebius, the GPU cloud startup, got the most attention. But the contrarian signal is in what he sold.
Core: The Infrastructure Super-Cycle
Let me break down the signal-to-noise ratio. The new positions form a coherent thesis: Soros is betting on the physical layer of the AI economy, not the software layer.
- Nebius (NBIS): A GPU cloud operator that provides compute for AI training and inference. The classic “picks and shovels” play. The bull case is simple: AI demand for compute is exponential, and Nebius is one of the few independent providers not tied to a hyperscaler. But the real insight is the timing. Nebius re-listed in October 2024 after a restructuring. By Q2 2025, the stock had already run up. Soros likely bought during a pullback or built a position over time. The risk is supply chain dependency on Nvidia—if H200 deliveries slip, Nebius gets crushed.
- DigitalBridge (DBRG): A digital infrastructure REIT focusing on data centers, cell towers, and fiber. This is a bet on the physical footprint of the internet. Data center rents are sticky, and the AI boom is forcing operators to expand faster than expected. Soros is buying the land and power, not the chips. That's a structural hedge.
- Taylor Morrison (TMHC): A homebuilder. At first glance, this seems unrelated to AI. But the macro link is rate sensitivity. If the Fed cuts rates in H2 2025, homebuilders explode higher. Soros is betting on softening inflation and a pause in rate hikes. But the hidden logic is housing supply shortage—the US is underbuilding by 1.5 million units per year. TMHC is a play on structural demand, not cyclical speculation.
- American Electric Power (AEP): A regulated utility. Most analysts call this a defensive bet. I call it a power thesis. AI data centers are projected to consume 10% of US electricity by 2030. AEP serves regions with high data center activity—Ohio, Virginia, West Virginia. Soros is buying the electrical grid. The tariff-regulated model ensures stable cash flows, but the real alpha is the multi-year growth in power demand. This is not a defensive play; it's a growth play disguised as a utility.
- Apogee Therapeutics (APGE): A biotech focused on obesity and inflammation. This is the outlier. But biotech is a classic Soros move—bearish on the macro, bullish on innovation. He's betting on drug approval cycles that are uncorrelated with interest rates.
The Exit Signal: What Soros Sold
The real story is in the exits. He sold Salesforce (CRM) and GlobalFoundries (GFS).
Salesforce is the poster child for AI software hype. The stock rallied on AI agent narratives, but the revenue growth is decelerating. Soros is saying: the market is pricing in too much AI software value too early. He's selling the dream and buying the reality.
GlobalFoundries is a semiconductor foundry that benefited from the CHIPS Act subsidies. But the stock has been flat. The subsidies are priced in, but the competitive moat is weak. Taiwan Semiconductor and Intel are eating the high-end. GFS is stuck in mature nodes. Soros is saying: policy tailwinds don't create sustainable competitive advantage. The rug wasn't pulled, it was always a rug.
Contrarian: The Market Misreads the Signal
The popular narrative is that Soros is bullish on AI. I disagree. He is bullish on the infrastructure that enables AI, but bearish on the AI software layer. He's selling the narrative and buying the implementation. That's a contrarian position because retail is piling into AI software stocks like Salesforce, while the smart money is rotating into utilities and data centers.
But there's a deeper contrarian angle: the 13F doesn't show hedges. Soros could be short the NASDAQ or long puts on the AI software ETF. The 13F only shows long equity positions. The real directional bet might be the opposite of what appears. We're flying blind on the derivative book.
Also, the new positions are small relative to the fund's $6.5 billion US equity portfolio. Nebius is a micro-cap. DigitalBridge is mid-cap. The signal is more about the direction of the sector rotation than the magnitude of the bet. If Soros wanted to make a massive AI infrastructure bet, he would have bought Nvidia, not a tiny GPU cloud operator. The scale suggests a tactical allocation, not a mega-thesis.
Takeaway: The Q3 13F Is the Real Catalyst
The model is clear: Soros is rotating out of software and manufacturing into energy, housing, and compute infrastructure. But the 45-day delay means you can't trade on this data. You have to wait for the Q3 13F, due in mid-November 2025. If he adds to Nebius and DigitalBridge, the thesis is confirmed. If he sells them, it was a quarter-end window-dressing trade.
Two weeks in the lab, one second in the field. The Q2 filing gives you the lab work. The Q3 filing will tell you if the field execution matches.
For now, the actionable insight is not to chase Soros's picks. Instead, look at the sectors he's exiting: software and semiconductor manufacturing. Those are the areas where the market is still optimistic, but the smart money is leaving. The opposite trade—shorting Salesforce and going long utilities—is a higher-probability play than buying Nebius at current levels.
Debugging the market means reading the exits, not the entries. Soros's Q2 13F is a map of where the value is moving. Follow the infrastructure, not the hype.