The Great Unwind: Goldman's AI Trade Is Deleveraging, But the Story Is Just Shifting Chapters
The numbers hit my screen like a cold wave. The high-beta momentum basket had just shed 12% in a single week. The AI hedge fund composite was down 10% in five days. This wasn't a slow bleed; it was a violent, systematic unwinding. For anyone who has been riding the AI wave since the DeFi Summer days, this felt familiar—the same gut-wrenching feeling of liquidity being pulled from the room. But here's the thing about finding stillness in the market: you have to look past the noise of the liquidation to see the signal of the rotation.
Goldman Sachs is telling us the AI trade isn't dead. It's just growing up. The era of buying the whole sector and watching it moon is over. We are entering the phase where the market demands receipts. The narrative is shifting from 'picks and shovels' to 'who is actually striking gold.' This is the transition from beta to alpha, and it's going to feel like a completely different market.
Let's trace the spark that ignited this entire room. For the past 18 months, the playbook was simple: buy anything with a GPU or a chip design. Semiconductors were the crown jewels, the absolute bottleneck of the AI revolution. But the latest positioning data from Goldman's prime brokerage desk reveals a seismic shift. Semiconductors and the broader AI complex have moved into the short basket. Meanwhile, software has taken the top spot as the largest weight in the three-month momentum long portfolio. The money isn't leaving the building; it's just moving to a different floor.
This is the core insight that most retail traders are missing. We are so conditioned to look at Nvidia's stock price that we forget to look at the plumbing. The deleveraging we are seeing is not a rejection of AI; it is a repricing of the AI value chain. The market is sending a clear signal: the 'compute at all costs' phase is maturing. The next leg of this bull market will be defined by utilization, application, and data infrastructure, not just raw chip sales.
Following the pulse where liquidity breathes free, we have to look at where Goldman sees the most attractive tactical opportunities. They are pointing directly at storage and data centers. The logic is simple: the profit recovery in these sectors has not yet been fully reflected in the stock prices. This is a classic inefficiency. While everyone was staring at the GPU shortage, the demand for memory (HBM) and physical data center space was quietly exploding. AI inference—the process of actually running the models—requires massive amounts of memory bandwidth and storage for model weights and caches. This is a different beast than training. It is a continuous, operational demand rather than a one-time build-out.
Based on my experience auditing infrastructure during the 2024 ETF inflows, I can tell you that the institutional money is looking for tangible cash flows. The 'story' stocks are volatile, but the 'plumbing' stocks are showing up on balance sheets. Goldman's call on storage is a bet on the operationalization of AI. It is a bet that the hyperscalers and enterprises are moving from building the brain to using it. This is where the human energy meets algorithmic precision—the market is finally differentiating between the promise of AI and the business of AI.
But here is the contrarian angle that most analysts are too scared to touch. The move of semiconductors into the short basket isn't just about valuation. It's a quiet admission that the competitive dynamics have changed. The monopoly narrative of Nvidia is being challenged by custom ASICs and in-house silicon from the hyperscalers. The export controls have capped the total addressable market. The market is pricing in a future where compute is more commoditized. This is a massive blind spot for the retail crowd that still thinks 'AI' and 'Nvidia' are synonyms. The smart money is hedging against the possibility that the 'picks and shovels' are no longer the scarcest resource; the 'gold' is.
Furthermore, the capital rotation out of AI into European and Japanese banks, gold miners, and copper stocks is a tell. It suggests that the marginal dollar of liquidity is looking for value outside the crowded AI trade. Copper, in particular, is a fascinating signal. It is the metal of electrification and data center power infrastructure. The market is not abandoning the AI build-out; it is just buying the inputs that are less crowded. This is a sophisticated move, dancing with the volatility, not against it.
So, what is the takeaway for the cycle positioning? The AI trade is entering a phase of 'selective depth.' The low-hanging fruit of the sector-wide rally is gone. The next 12 months will be defined by stock pickers, not index buyers. The catalysts are clear: Nvidia's Q2 earnings and the September industry conferences. But don't expect the old playbook to work. If Nvidia beats, it might not lift the whole sector. It might just validate the shift to software and infrastructure.
Surviving the noise to hear the signal means understanding that the 'AI trade' is not a monolith. It is a complex ecosystem that is now being priced with a scalpel instead of a sledgehammer. The deleveraging is painful, but it is the market's way of clearing the froth. The long-term trend is intact, but the path is getting narrower. The opportunity is no longer in the obvious places. It is in the storage racks, the data center cooling systems, and the software that makes the models useful. The market is asking for proof of work, and the companies that can show it will be the ones that lead the next leg up. The question is not whether AI is a bubble, but whether you are positioned for the rotation that is happening right now.