AI Trading's 'Deleveraging' Phase: A Crypto-Native's Guide to the Value Shift

The market is bleeding, but not the way you think. Over the past week, Goldman Sachs’ high-beta momentum portfolio—a basket of stocks tied to the AI narrative—has dropped 12%. Their AI hedge fund strategy, a bundle of long and short positions, lost 10% in five days. To a crypto-native, this sounds like a familiar tune: leverage unwinding, crowded trades collapsing, and the herd scrambling for the exits. But here’s the twist: Goldman says the AI trade isn’t dead. It’s just changing shape. And for those of us who’ve watched DeFi cycles, that’s a signal worth reading. Let me rewind. I’m Olivia Walker, a decentralized protocol PM based in Buenos Aires, and I’ve been watching this convergence of traditional finance and blockchain since 2016. When I first started explaining Hyperledger to skeptical bankers, they’d ask, “Why should I trust a trustless system?” Now, I’m asking myself the same question about Goldman’s analysis. The report, dated August 23, 2024, is a single-source document from the investment bank—no independent verification, no counterpoints. But it’s rich with data points that resonate with anyone who’s survived a DeFi summer or a Terra collapse. The key takeaway? AI trading is entering a “deleveraging and rebalancing” phase, and the value is shifting from the picks-and-shovels (hardware) to the prospectors (software and infrastructure). Sound familiar? It’s the same pattern we saw in blockchain: from mining rigs to dApps. Goldman identifies three sectors as “tactically most attractive”: storage, data centers, and software. The logic is that these areas have “profit recovery that isn’t yet fully reflected in stock prices.” Specifically, storage and data centers are showing the biggest valuation gap—meaning their earnings have improved, but share prices haven’t caught up. This is a clear signal that the market is underpricing the downstream effects of AI deployment. Storage isn’t just about holding data; it’s about the memory and caching needs of inference models, which are exploding as AI moves from training to real-world application. Data centers, meanwhile, are the physical backbone of that deployment, and their utilization rates and rents are rising. But here’s where it gets interesting for blockchain enthusiasts: Goldman also notes that semiconductors have entered the short portfolio in their momentum model, while software has become the largest weight in the long portfolio. This is a rotation from hardware to software, from the “seller of shovels” to the “miner of gold.” It’s analogous to the shift we saw from Ethereum’s POW to POS, where the value accrual moved from miners to stakers and validators. Based on my experience auditing smart contracts for DeFi protocols, I’ve seen how these patterns play out. In 2020, during DeFi Summer, the early movers were the protocols that built the infrastructure—Uniswap for liquidity, Aave for lending. But within a year, the value had shifted to the applications built on top—yield aggregators, insurance protocols, and gaming platforms. The same thing is happening in AI: the initial hype was around NVIDIA’s GPUs and the training race, but now the market is pricing in the operational phase. Storage and data centers are the “liquidity pools” of AI, while software is the “yield farming.” The contrarian perspective? Goldman’s analysis might be too optimistic. The report is a single source, and the bank has a vested interest in maintaining market confidence—they’re underwriters for many of these companies. But for the blockchain community, the real lesson is in the methodology: look for the “value gap” between earnings and price, just as we look for the “TVL gap” between deposits and token value. Let me challenge this with a softer touch. I’ve been in too many DAO governance calls where the loudest voice was the one with the most tokens, not the most insight. Goldman’s report is a classic example of top-down analysis: it’s all about capital flows, leverage, and momentum. It ignores the human element—the developers, the users, the communities building these technologies. In my work with Art Blocks, I saw how generative artists used blockchain to reclaim ownership of their work. The real value wasn’t in the GPU cycles; it was in the narrative. Similarly, AI’s long-term value will come from the applications that solve real human problems, not from the hardware that runs the models. The shift to software and storage is a positive sign, but it’s still a financial abstraction. The true test will be whether these companies are building for users, not just for investors. What are the risks? First, the storage and data center “profit recovery” might be driven by factors other than AI—like traditional IT spending cycles or cloud service provider upgrades. If the recovery is cyclical, not structural, the valuation gap could close in the wrong direction. Second, the semiconductor short might be a tactical bet on short-term overvaluation, not a strategic view on the industry’s long-term potential. If NVIDIA’s Q2 earnings report (expected late August) surprises to the upside, the momentum could reverse, and the software rotation might be premature. Third, the capital flowing out of AI into European banks, gold miners, and copper stocks suggests that the sector is becoming crowded—a classic sign of a top in a bull run. For crypto-native readers, this is like watching the DeFi summer of 2020, where the reflexive nature of the market led to a crash when liquidity dried up. The opportunities, however, are clear. For those with a long-term horizon, the storage and data center sectors offer a high-conviction entry point. Look at companies like Micron, SK Hynix, or Samsung for storage, and Equinix or Digital Realty for data centers. The key is to verify the earnings recovery signals—check their quarterly reports for guidance on HBM (high-bandwidth memory) demand and data center utilization rates. For software, the focus should be on companies with a clear AI revenue model, not just AI hype. Think of them as the “blue-chip DeFi protocols” of this cycle: they have the distribution, the data, and the network effects to sustain growth. But here’s my real concern: the report doesn’t address the ethical and systemic risks that I’ve seen in both AI and blockchain. The USDT problem—Tether’s lack of a transparent audit—is a parallel to the AI industry’s lack of transparency around energy consumption and data privacy. Goldmans’ analysis is purely financial, ignoring the social costs. In 2022, after the Terra collapse, I helped a DAO rebuild its governance framework by focusing on values-first decision-making. The same principle applies here: we need to ask not just which sectors are profitable, but which are sustainable. The AI industry is facing a “blob saturation” problem, similar to what I predict for Layer 2s after the Dencun upgrade. Storage and data center demand will double, but so will the energy and environmental costs. The protocols that succeed will be those that embed human-in-the-loop verification and ethical guardrails, just as I negotiated in the AI protocol committee in 2025. Connect first, transact second. Always. If you’re reading this as a crypto investor, don’t just chase the momentum. Understand the technology, the community, and the values behind it. The AI trade isn’t over—it’s evolving. The real alpha will come from those who see the human cost behind the capital flows. Education is the ultimate form of protection. In my workshops in Latin America, I taught users to read smart contract audits before they deposited funds. The same logic applies here: read the earnings reports, not just the headlines. The storage and data center sectors are attractive, but only if the profit recovery is from AI demand, not cyclical IT spending. Verify the data. The most decentralized system is the one that remembers why it was built. AI and blockchain are both tools for human empowerment, not just instruments for financial speculation. As we move from the “training phase” to the “inference phase,” let’s ensure that the value flows to the creators, not just the capital allocators. The contrarian perspective is that Goldman’s analysis is a mirror of the market’s collective bias—it’s still trapped in the old paradigm of “returns at all costs.” The future belongs to those who build with integrity. So, what’s the takeaway? The AI trade is entering a new phase, but it’s not a bubble bursting—it’s the market waking up to the reality that value lies in application, not just infrastructure. For the blockchain community, this is a moment to reflect: are we repeating the same mistakes of the 2020 DeFi summer, or are we building something more resilient? The answer depends on how we choose to act. Connect first, transact second. Always.