Soros Fund Management filed its Q2 2025 13F on August 15. The headline: new positions in Nebius Group (NBIS), DigitalBridge (DBRG), American Electric Power (AEP), Taylor Morrison Home (TMHC), and Apogee Therapeutics (APGE). The exits: Salesforce (CRM) and GlobalFoundries (GFS). At first glance, this looks like a scattergun portfolio shuffle. It is not. It is a structured rotation from the old guard of enterprise software and chip manufacturing to the raw infrastructure of the next cycle: AI compute, digital real estate, and power generation.
Let me state the obvious first: 13F data is a lagging snapshot. It captures June 30 holdings, disclosed 45 days later. Soros could have sold everything by August. But the pattern of what he bought and sold tells a story about where he thinks the structural value lies — and where it does not.
Context: The 13F Trap and the Soros Signal
Soros Fund Management, now controlled by Alex Soros, manages roughly $65 billion in U.S. equities. The firm has evolved from a macro hedge fund into a long-biased family office. Its signal-to-noise ratio is lower than the George Soros era. But the sector-level choices still carry weight because they reflect a multi-billion-dollar team’s research.
I’ve seen this play before. In 2017, I audited 50 ICOs and rejected 90% based on code quality and revenue models. The market called me a bear. Then the crash proved that hype without fundamentals is a liquidity trap. Soros’ Q2 moves echo that same empirical skepticism: he is rotating out of stories that have been fully priced and into sectors where the fundamentals are just beginning to compound.
Core: The Rotation Deconstructed
Let’s break down each leg of the trade.
Sell: Salesforce (CRM) – Enterprise software is being disrupted by AI-native tools. Salesforce’s Agentforce is a reaction, not a revolution. The market has been pricing CRM as an AI winner, but the company’s core revenue growth has slowed. Soros is saying: the AI narrative is already in the price. Sell the story.
Sell: GlobalFoundries (GFS) – This is the most telling exit. GFS is a mature-node chip foundry that received billions in CHIPS Act subsidies. Yet the stock has underperformed. Why? Because subsidies do not equal competitive advantage. AI demand is for advanced nodes (TSMC, Samsung) and for GPU compute, not for legacy chips. Soros is betting that government handouts cannot replace market-driven efficiency. I have seen this pattern in DeFi: protocols with grant money but no product-market fit eventually bleed out.
Buy: Nebius Group (NBIS) – NBIS is a GPU cloud provider operating out of Europe, with a focus on AI inference workloads. It is small, unprofitable, and capital-intensive. But it sits at the intersection of two structural trends: AI compute scarcity and the need for alternative cloud providers to AWS/Azure. Soros is buying the picks and shovels of the AI gold rush, not the gold miners. Volatility is the tax on undiscerned capital — and NBIS is volatile, but its underlying demand is real.
Buy: DigitalBridge (DBRG) – This is a digital infrastructure REIT, owning data centers and cell towers. Data centers are the new oil wells. AI training and inference require massive physical footprint. DBRG benefits from both AI demand and the broader trend of cloud migration. Soros is essentially buying a leveraged play on the real estate of the internet.
Buy: American Electric Power (AEP) – A utility that generates and transmits electricity. The conventional view is that utilities are boring bond proxies. But AI data centers are projected to add 10-20% to U.S. electricity demand by 2030. That is a structural shift for a sector that has seen near-zero load growth for decades. AEP operates in regions with high data center concentration (Ohio, Virginia). This is a bet on electricity demand, not on interest rates. Yield without protocol is just delayed loss — but AEP’s yield is backed by a regulated monopoly with pricing power.
Buy: Taylor Morrison Home (TMHC) – A homebuilder. This seems contradictory to the AI theme. But the logic is a soft landing: if the Fed cuts rates, mortgage rates drop, and housing demand picks up. U.S. housing supply is structurally short. TMHC is a play on the demographic tailwind of millennials buying homes. It is a hedge: if AI demand fails, the economy still has housing.
Buy: Apogee Therapeutics (APGE) – A biotech focused on inflammatory diseases. High-risk, high-reward. This is a small allocation to a binary outcome. Not a core theme, but a signal that Soros is willing to take asymmetric bets in areas where the downside is limited and the upside is 10x.
Contrarian: Why Retail Is Missing the Real Signal
The market is obsessed with AI software: ChatGPT, Copilot, Salesforce’s Agentforce. Retail traders chase the headlines. Soros is selling that. The smart money is moving downstream to the infrastructure that makes AI possible: compute, power, and real estate. The same happened in DeFi in 2020: everyone bought tokens, but the real value accrued to protocol infrastructure and liquidity providers.
Here is the counter-intuitive angle: the 13F shows Soros buying homebuilders and utilities, which are defensive sectors. But the interpretation is not defensive — it is a bet on growth. AEP and TMHC are not defensive; they are cyclical assets that benefit from a specific macro scenario: gentle rate cuts and sustained demand. This is a nuanced view, not a macro hedge.
Also, note that Soros dropped GFS, a direct beneficiary of government policy. The common narrative is that the CHIPS Act is a tailwind for U.S. semiconductor manufacturing. Soros disagrees. He is betting that the market will reward companies with real competitive advantages, not those that depend on subsidies. This is a lesson for crypto: protocols that rely on grants or token incentives without organic demand will eventually fail. I trade the ledger, not the hype cycle.
Takeaway: Actionable Price Levels and the AI Infrastructure Thesis
The Soros Q2 13F is a map of where institutional capital is flowing. The core thesis: AI infrastructure (compute, power, data centers) is under-owned relative to AI software. The market pays for clarity, not complexity. The clarity here is that NBIS, DBRG, and AEP are the structural beneficiaries of the AI buildout.
But the lag is real. By the time you read this, Soros may have already taken profits or added more. The real signal is not the specific stocks — it is the sector rotation. For crypto traders, the equivalent is rotating from AI agent tokens to DePIN projects that provide GPU compute, decentralized storage, or energy credits. Look at projects like Akash Network, Render, or Power Ledger. The same pattern applies: infrastructure wins over application hype.
