The Hook: When the Smart Money Starts Running Scared
The blockchain's gray matter is humming with a signal most retail investors can't hear. On August 23, Goldman Sachs released a note that sent ripples through institutional trading floors: the AI trade is entering a "deleveraging phase." High-beta momentum portfolios dropped 12% in a single week. AI-focused hedge fund portfolios fell 10% in five days. The leverage that had been piling into AI names—pushing them to stratospheric valuations—is now unwinding with the force of a coiled spring.
But here's what caught my attention, chasing the ghost in the blockchain's gray matter: Goldman didn't say the AI trade is over. They said the way you make money from it has fundamentally changed. The era of buying the whole sector and watching it rise is dead. What's replacing it is something far more interesting—and far more dangerous for those still holding yesterday's narrative.
Context: The Narrative Cycle Repeats, But the Hash Changes
I've been in this industry long enough to recognize the pattern. In 2017, it was ICOs. In 2020, it was DeFi Summer. In 2021, it was NFTs. Each time, the same arc: euphoria, leverage, capitulation, and then—crucially—a rotation into the next narrative while the old one bleeds out.
What Goldman is describing is the same cycle playing out in AI equities. The first phase was pure narrative: "AI will change everything," and every stock with the letters "AI" in its ticker went vertical. That was the beta trade—buy the basket, ride the wave. But now, the wave is breaking.
The key signals from Goldman's note are unmistakable:
Semiconductors and AI complexes have moved into short portfolios. The most beloved trade of the past 18 months—Nvidia and its ilk—is now being actively shorted by professional investors. This isn't a tactical hedge; it's a structural repositioning.
Software has replaced semiconductors as the largest weight in the three-month momentum long portfolio. The momentum factor—which quantifies which stocks have been rising the fastest—has flipped. Money is rotating from the "picks and shovels" of AI (hardware) to the "gold miners" (applications and software).
Storage and data centers are now "tactically the most attractive sectors." Goldman explicitly states that the "profit recovery has not yet been fully reflected in stock prices" for these segments.
This is where the narrative gets interesting. Where code meets the human heartbeat, the story is no longer about who builds the most powerful chips. It's about who stores the data those chips process, and who houses the infrastructure that runs the inference workloads.
Core: The Deleveraging Mechanism and What It Really Means
Let me break down what's actually happening under the hood, because the surface-level reading misses the deeper signal.
The Leverage Unwind
When Goldman says "deleveraging," they're not just talking about margin calls. They're describing a structural unwind of crowded positioning. The AI trade had become the most crowded trade in institutional memory—every hedge fund, every pension fund, every sovereign wealth fund needed AI exposure. That crowding created a feedback loop: rising prices attracted more capital, which pushed prices higher, which attracted more capital.
But here's the thing about feedback loops: they work in reverse too. When the first wave of selling hits, it triggers risk-parity algorithms to reduce exposure. That selling pushes prices down, which triggers more selling. The 12% weekly drop in high-beta momentum portfolios isn't a correction—it's a cascade.
The Rotation Signal
The shift from semiconductors to software in momentum portfolios is the most telling signal in the entire note. Momentum factors are backward-looking—they buy what's been going up. The fact that software has overtaken semiconductors means the price action has already shifted, even if the narrative hasn't caught up.
This is classic narrative hygiene territory. The market is telling us that the "AI infrastructure buildout" story—the one that justified infinite capital expenditure on GPUs—is losing its persuasive power. What's replacing it is a more mundane but ultimately more sustainable story: "AI applications are generating actual revenue."
The Storage and Data Center Anomaly
Goldman's call on storage and data centers deserves special attention. The logic is straightforward: AI inference workloads require massive amounts of memory bandwidth and storage capacity. Every time a user queries an AI model, the system needs to access model weights, retrieve context, and process tokens. This is fundamentally different from training workloads, which are compute-bound. Inference is memory-bound.
The profit recovery in storage and data centers is real—I've seen it in the earnings reports of companies like Micron and the data center REITs. But the market hasn't fully priced it in because the narrative is still fixated on Nvidia's quarterly numbers.
The Nvidia Catalyst
Goldman flags Nvidia's Q2 earnings as a key catalyst. This is where the narrative could go either way. If Nvidia beats and raises, the AI trade gets a temporary reprieve. If they guide lower—or even just in-line—the deleveraging accelerates.
But here's the contrarian angle: the market's obsession with Nvidia's earnings is itself a narrative trap. The stock has already priced in perfection. Any deviation from "blowout" will be punished. The real signal to watch isn't Nvidia's revenue—it's the guidance for data center revenue and the commentary on inference demand.
Contrarian: The Blind Spot in Goldman's Analysis
Reading the invisible signals of digital identity, I can't help but notice what Goldman isn't saying.
First, the "profit recovery" in storage and data centers may not be as AI-driven as the narrative suggests. Traditional enterprise IT spending has been recovering as companies refresh their infrastructure post-pandemic. Cloud service providers are on their own capital expenditure cycles. The AI contribution to storage and data center profits could be smaller than the market assumes—and if that's the case, the "valuation gap" Goldman identifies might be a value trap, not an opportunity.
Second, the rotation into software as the momentum leader is fragile. Software stocks are notoriously volatile in momentum portfolios. If the next round of AI application earnings disappoints—if the "AI revenue" story doesn't materialize as quickly as expected—the momentum factor could flip again, this time into cash or defensive sectors.
Third, and this is the one that keeps me up at night: Goldman's note is a sell-side document. Goldman Sachs is a major investment bank with underwriting relationships across the AI complex. Their "AI trade isn't over" framing could be as much about maintaining market confidence as it is about objective analysis. The note's optimism about storage and data centers conveniently aligns with sectors where their clients hold significant positions.
Takeaway: The Next Narrative Is Already Forming
Unraveling the tapestry of digital mythologies, the AI trade's next phase won't be about hardware or even software. It will be about verification—the human-in-the-loop validation of AI-generated content, the infrastructure that proves what's real and what's synthetic.
The capital rotating into European and Japanese banks, gold miners, and copper stocks isn't just a flight to safety. It's a bet on the physical world reasserting itself over the digital. Copper, in particular, is the unsung hero of AI infrastructure—every data center, every power grid upgrade, every electric vehicle needs copper. The market is starting to understand that AI's bottleneck isn't compute—it's power and materials.
The question I'm asking myself as I follow the trail where others see only noise: when the AI trade fully deleverages, where does the narrative capital go next? The answer might not be in the tech sector at all. It might be in the companies that build the physical foundation for the digital revolution—the miners, the power utilities, the storage providers, the data center operators.
The chain never lies, but people do. And right now, the people at Goldman are telling us something important: the AI trade isn't dead, but the easy money is gone. The next phase belongs to those who can read the signals beneath the surface—who can see that the ghost in the machine isn't the algorithm, but the human decisions that shape where capital flows.
Architecture is just storytelling with constraints. The AI narrative is being rebuilt, and the new story is about infrastructure, verification, and the physical world. The question is whether you're still holding yesterday's narrative—or ready for what comes next.