Hook
On August 15, 2025, Soros Fund Management filed its Q2 13F. The headline: five new positions, five full exits. The market yawned. But if you peel back the tickers, you see something that the crypto-native analyst community has entirely missed—a clean rotation from old-tech manufacturing into AI infrastructure that mirrors exactly the value migration happening in decentralized compute networks right now.
Context
Let me be clear: Soros Fund Management is no longer the macro hedge fund of the 1990s. Since Alex Soros took control in mid-2023, the fund has shifted toward a more concentrated, thematic equity portfolio. As of Q1 2025, its U.S. equity holdings sat at roughly $6.5 billion—medium-sized by institutional standards, but the signal-to-noise ratio remains high because of the Soros brand.
The Q2 2025 filing (disclosed August 15, covering positions as of June 30) shows the following:
New Buys: - Nebius Group (NBIS) – AI infrastructure / GPU cloud - DigitalBridge Group (DBRG) – digital infrastructure REIT - Apogee Therapeutics (APGE) – biotech (obesity/inflammation) - Taylor Morrison Home (TMHC) – homebuilder - American Electric Power (AEP) – regulated utility
Exits: - Salesforce (CRM) – enterprise software - GlobalFoundries (GFS) – semiconductor manufacturing - Hologic (HOLX) – medical devices - Two unnamed positions (per the filing, but the analysis confirms at least five exits)
At first glance, this looks like a typical diversified portfolio. But the pattern is unmistakable: Soros is selling legacy tech and buying infrastructure that supports the physical layer of the AI economy. And that pattern has direct implications for crypto—specifically for the DePIN and AI compute token sectors.
Core: The Value Migration Playbook
The most telling move is the swap from GlobalFoundries (GFS) to Nebius (NBIS). GFS is a U.S.-based semiconductor foundry that benefited from the CHIPS Act subsidies. NBIS is a Dutch company that operates massive GPU clusters for AI training and inference. Soros didn't just rotate sectors; he rotated along the value chain—from chip manufacturing to compute services.
Why does this matter for crypto? Because the same value migration is happening in decentralized networks. The 2021–2023 narrative was all about GPU supply—miners, ASICs, and chip scarcity. The 2025 narrative is about compute utilization, latency, and service-layer margins. In DeFi terms, think of it as moving from being a liquidity provider to being the protocol that aggregates and routes liquidity.
I've seen this pattern before. During DeFi Summer 2020, the early money flowed into lending protocols (Compound, Aave). By 2021, the value had migrated to yield aggregators (Yearn) and insurance protocols (Nexus Mutual). The infrastructure layer that enabled the ecosystem eventually captured more value than the base protocols themselves.
Soros' bet on Nebius is a bet that the compute layer—not the chip layer—will capture the lion's share of AI value. In crypto, that translates directly to projects like Render Network (RNDR), Akash Network (AKT), and io.net (IO). These are decentralized GPU marketplaces that provide the same service as Nebius but with token incentives and permissionless access.
But here's the catch: Nebius is a centralized company with audited financials, a management team, and a Nasdaq listing. Decentralized compute networks are still experimental, with unproven demand and tokenomics that often leak value to speculators rather than actual users. Soros is not buying crypto. He's buying the centralized analogue. The question is: will the decentralized version ever achieve the same institutional credibility?
The Utility Signal
Soros also bought American Electric Power (AEP), a regulated utility. The conventional read is that this is a defensive play against rate cuts. But look deeper: AEP's earnings calls in 2024 and 2025 repeatedly cited AI data center load growth as a driver of capital expenditure and rate base expansion. AEP is essentially a proxy for AI power demand.
In crypto, this maps to energy-focused projects like Powerledger (POWR) and Energy Web Token (EWT). But again, the institutional flow is into regulated utilities, not tokenized energy credits. The gap between the two is still wide.
Contrarian: What the Market Misses
The common narrative around this 13F is "Soros is bullish on AI." That's shallow. The contrarian angle is that Soros exited Salesforce (CRM) and GlobalFoundries (GFS)—two stocks that the market still considers AI winners. CRM has its Agentforce AI platform; GFS has government subsidies. Soros sold both. Why?
Because he sees the same thing I saw during the LUNA collapse: narrative can sustain a price only until the underlying economics fail. Salesforce is a legacy software company trying to bolt on AI features. GlobalFoundries is a mature-node foundry competing with TSMC and Intel for a shrinking piece of the pie. The AI trade has already moved past these names.
In crypto, the equivalent mistake is buying tokens of projects that have strong narratives but weak unit economics. For example, many AI agent tokens have no revenue, no users, and no moat. Soros' playbook says: buy the infrastructure that everyone needs, not the application that might be disrupted next quarter.
The 13F Lag Trap
Here's the real risk: the filing is as of June 30, 2025. It's now late August. Soros could have sold everything in July. The market's tendency to copy 13F filings blindly is a behavioral bias that I've exploited in my own trading. In 2021, I watched retail traders pile into stocks that hedge funds had already dumped, based on 13F data that was 45 days stale.
For crypto traders, the lesson is: don't buy NBIS stock just because Soros bought it. Instead, analyze the thesis behind the trade and apply it to the crypto analogue. If the thesis is "compute services will capture more value than chip manufacturing," then look at decentralized compute tokens that have actual revenue and a path to profitability.
Takeaway
Soros' Q2 2025 13F is not a crypto play, but it's a roadmap for where institutional capital is flowing. The rotation from old tech to AI infrastructure is clear. The next question for crypto is: which decentralized projects can credibly capture that same value?
Patience is a tactical advantage, not a virtue. Wait for the Q3 13F (due mid-November) to confirm whether Soros held or added to Nebius. If he did, the AI infrastructure theme is sticky. If he didn't, it was a quarter-long trade.
Either way, the data is clear: the value is migrating from supply to service. Code does not negotiate. It executes or it fails.
The chart shows fear; the order book shows intent. Soros' order book says AI infrastructure. The crypto market should take notes—but not blindly copy the trade.