Goldman Sachs Sees Storage Gold in AI — But Misses the DePIN Mine

Over the past five days, the AI hedge basket at Goldman Sachs lost 10%. The high-beta momentum basket fell 12%. And yet, the bank's analysts tell us the AI trade isn't over — it's just rotating. They point to storage and data centers as the next profit recovery story, a tactical pivot from semiconductors to the plumbing of machine intelligence. We don't need to look far to see the same rotation happening in our own corner of the world. The decentralized infrastructure tokens — the ones that actually store data, render graphics, and power compute for AI models — have been bleeding just as hard. Filecoin down 30% from its May peak. Render down 40%. Arweave holding better, but still bruised. The bear market didn't spare any sector, but it did teach us to read the signals beneath the blood. Let me give you the context first. Goldman's report, published last week, argues that the era of buying the entire AI complex for beta is ending. Momentum factors are rebalancing — software has overtaken semiconductors as the largest weight in their three-month long basket, while chips and the so-called 'AI complex' have moved into the short basket. They recommend storage and data center equities, citing the 'most pronounced valuation gap' — profits are recovering but haven't yet been priced into the stocks. Their key catalysts: Nvidia's Q2 earnings on August 28 and an industry conference in September. Now, as someone who's spent the last three years building on decentralized protocols — first in DeFi, now in compute and storage markets — I find this both validating and frustrating. Validating because the underlying thesis is sound: the AI buildout is real, and the physical layer of data storage and compute will see sustained demand. Frustrating because Goldman's map of that territory stops at the borders of Wall Street. They see Dell, Micron, and Super Micro — I see Filecoin, Arweave, and Akash. Let me get into the core analysis. The Goldman report hinges on a classic value-plus-growth trade: find sectors where earnings are recovering but stock prices haven't caught up. For traditional storage, that means companies selling HDDs, SSDs, and data center racks. The logic is straightforward — AI training and inference consume enormous amounts of storage, and hyperscalers are building data centers at a breakneck pace. The market has punished these names during the AI selloff, but their forward EPS revisions are turning positive. That's the kind of dislocation a quantitative desk loves. In the decentralized world, the equivalent is DePIN — decentralized physical infrastructure networks. These protocols incentivize individuals and small businesses to contribute storage, bandwidth, and compute. The revenue streams are token emissions, not stock dividends, but the fundamental economics are similar. Take Filecoin: its network storage capacity has grown steadily, and the amount of data stored on-chain has tripled in the last year. Yet the token price sits 80% below its 2021 high. The 'profit recovery' for these networks isn't in fiat earnings — it's in utilization and real-world usage. And by that metric, several DePIN projects are showing signs of life that the market hasn't priced in. Based on my own audit experience with storage protocols, I can tell you that the gap between narrative and fundamentals is wider here than in any traditional tech sector. For example, I spent two weeks last quarter modeling the cost of storing a petabyte of data on Arweave versus Amazon S3 over a five-year horizon. Arweave's one-time payment model becomes cheaper after year three, and that's without accounting for the resilience benefits of immutable, geographically distributed storage. But the market doesn't see that because the metrics are different — we talk about 'permanent storage' and 'cryptographic proof' while Goldman talks about 'earnings per share.' That's a translation gap that creates opportunity. The contrarian angle? Maybe Goldman is right to ignore us. Traditional storage and data centers have a massive head start — they're already integrated into the enterprise supply chain, they have regulatory clarity, and they're backed by trillion-dollar balance sheets. Decentralized alternatives still struggle with user experience, scalability, and the dreaded 'token volatility' that scares off institutional procurement. But here's the blind spot: the same argument was made against cloud computing in 2008, and against open-source software in the 1990s. The market is pricing DePIN as a speculative sideshow, while the underlying infrastructure is quietly becoming production-ready. When Nvidia reports earnings, the market will react to hyperscaler capex — but the compute that actually runs AI inference is increasingly moving to edge devices and distributed networks, not just central data centers. The bear market didn't kill the AI trade; it just made it selective. The same is true for DePIN. The tokens that survive will be those with real usage, sustainable tokenomics, and a clear path to profitability. In that sense, Goldman's framework applies perfectly to our world — we just need to translate 'EPS' into 'active deals' and 'profit recovery' into 'network utilization.' About Me: I'm Chris Thompson, a protocol PM in Nairobi who's been building in this space since 2017. I've audited smart contracts, forked Curve, and spent the last year bridging institutional interest into decentralized compute. I'm not a financial advisor, but I've learned to read the tea leaves of momentum and valuation across both traditional and crypto markets. So here's my takeaway: don't chase the Goldman recommendation by buying Dell and Micron. Instead, look at the decentralized equivalents — the storage and compute protocols that are trading at a fraction of their future value, with real usage growth that mirrors the 'profit recovery' Goldman is betting on. The AI trade isn't over; it's just moving from the chip layer to the storage and compute layer. And in that layer, the most undervalued assets aren't listed on any stock exchange. They're living on-chain, waiting for the market to catch up. The question is not whether decentralized infrastructure will win — it's whether you'll be positioned when the rotation finally arrives.