Hook
Soros Fund Management dumped Salesforce and GlobalFoundries in Q2 2025. The market barely blinked. But the real signal is in what they bought: Nebius, DigitalBridge, American Electric Power, Taylor Morrison Home, and Apogee Therapeutics. Five new positions. Five old ones cleared. This isn’t a random rebalance. It’s a structural rotation from legacy tech into the physical backbone of the AI economy. And for anyone watching the intersection of compute, energy, and digital infrastructure, the implications for blockchain and crypto are direct.
Context
The 13F filing, disclosed on August 15, 2025, captures Soros Fund Management’s equity holdings as of June 30, 2025. The firm, now led by Alex Soros, manages roughly $6.5 billion in U.S. equities. This is a mid-sized institutional player, not a whale. But the Soros brand carries outsized signal value. The fund’s history of macro bets—from the 1992 pound short to the 2022 FTX warning—means every quarterly reveal is dissected for directional clues. This quarter’s moves are particularly stark: a clean exit from enterprise software (Salesforce) and mature semiconductor manufacturing (GlobalFoundries), and a simultaneous entry into AI infrastructure, digital real estate, regulated utilities, homebuilding, and biotech. The pattern is coherent. The question is whether the market has priced it in.
Core: The Infrastructure Quadrant
Let’s break down the new positions. Nebius (NBIS) is a GPU cloud provider, re-listed on Nasdaq in October 2024 after a corporate restructuring. Soros bought it in Q2, likely before the AI compute narrative fully repriced the stock. Nebius operates thousands of Nvidia H100 and H200 GPUs, offering cloud services for AI training and inference. This is a direct play on the compute scarcity that underpins not just AI but also proof-of-work mining and zero-knowledge proof generation. DigitalBridge (DBRG) is a digital infrastructure REIT—owns data centers, cell towers, fiber networks. Again, physical assets that host the cloud. American Electric Power (AEP) is a utility with massive exposure to the grid. AI data centers are projected to double U.S. electricity demand by 2030. AEP benefits from that growth, and its regulated earnings provide a defensive buffer. Taylor Morrison Home (TMHC) is a homebuilder—puzzling at first, but logical if you believe the Fed will cut rates and housing supply remains constrained. Apogee Therapeutics (APGE) is a biotech focused on inflammatory diseases—a high-risk, high-reward play that doesn’t fit the infrastructure theme neatly, but adds optionality.
On the sell side: Salesforce (CRM) was a legacy enterprise software bet. The market still loves its AI agents, but Soros clearly saw the growth ceiling. GlobalFoundries (GFS) is a pure-play semiconductor foundry reliant on U.S. CHIPS Act subsidies. Soros sold it, implying the subsidy-driven edge is already priced in, and the company lacks the competitive moat of TSMC or Samsung. The other sold positions—ATI, Align Technology, and a few others—are less significant, but the pattern is clear: out of software and manufacturing, into physical infrastructure and energy.
The core insight: Soros is betting on the “pick-and-shovel” of the AI revolution. Not the AI models themselves, but the power, computing, and real estate that make them possible. This is the same logic that drove early Bitcoin miners to hoard ASICs and secure cheap electricity. The structural demand for compute is not a narrative—it’s a measurable, growing consumption of kilowatt-hours and rack space. In my 23 years of market surveillance, I’ve seen capital flow follow physical bottlenecks. The GPU is the new oil derrick. The data center is the new refinery.
Contrarian: The Blind Spot in Consensus
Most analysts are still obsessed with AI software names—Microsoft, Salesforce, Adobe. They chase the narrative of AI agents replacing workers. Soros’s move suggests the real value accrues to the infrastructure layer, not the application layer. This is a classic “sell the hype, buy the picks-and-shovels” trade. But there’s a deeper contrarian angle: Soros’s bet on American Electric Power and Taylor Morrison Home implies a macroeconomic view that the Fed will cut rates in the second half of 2025, and that inflation will remain sticky enough to keep utility pricing power intact. The market is currently pricing in a soft landing. Soros is hedging with a defensive utility and a rate-sensitive homebuilder. That’s not a pure risk-on bet. It’s a cautious allocation that says “growth is real, but not without headwinds.”
For the crypto world, the blind spot is the assumption that AI and blockchain are separate. They are not. The same GPU clusters that train large language models can also serve as validators for proof-of-stake networks or generate proofs for zk-rollups. The same data centers that host AI workloads also host Bitcoin miners and Ethereum validators. Soros’s investment in DigitalBridge and Nebius is a vote for the convergence of physical digital infrastructure. The market hasn’t priced that convergence yet. When the next crypto bull cycle arrives, the demand for compute will be even more acute, and the infrastructure providers will be the first to benefit.
Takeaway
Soros’s Q2 portfolio is a roadmap for the next 12–18 months. It points to AI infrastructure, utilities, and digital real estate as the structural winners. If you’re holding crypto assets, the question is not whether Soros bought Bitcoin (he didn’t, in this filing). The question is whether his infrastructure bets will squeeze the supply of GPU compute and electricity, raising costs for miners and validators. The answer is likely yes. Watch the next 13F filing in November. If Soros adds more Nebius and DigitalBridge, the signal is confirmed. If he trims, the trade is already crowded. Either way, the arbitrage between AI compute and blockchain compute is widening. Liquidity doesn’t follow narratives; it follows structural demand. And right now, structural demand is screaming for more power and more silicon.
