The ledger never sleeps, only updates. And right now, Goldman Sachs is pushing a hard update to the AI trade narrative that most retail portfolios haven't synced with yet.
Chaos is not noise; it is unindexed data. For months, the market treated AI as a monolith—a rising tide lifting every ticker with a GPU or a chatbot demo. That era just ended. Goldman's latest desk note, dated August 23, isn't a call to abandon ship. It's a surgical re-indexing of where the value actually sits. The headline: AI trade isn't over, but the phase of getting rich off the entire sector's beta is dead. What's replacing it is a far more interesting, and brutal, game of picking winners in the plumbing.
Let's decode the signals, because the details are where the truth hides.
The Context: A Deleveraging Event, Not a Narrative Break
First, the macro backdrop. This isn't a crash. This is a correction in positioning. Goldman's data points to a classic deleveraging event: the high-beta momentum portfolio dropped 12% in a single week. Their proprietary AI hedge fund basket fell 10% in five days. Leverage in the AI complex has retreated from extreme highs. This is the market exhaling after a parabolic run, not a fundamental rejection of the technology.
The key phrase from the note is that the phase of gaining excess returns from the overall sector rally is changing. That's banker-speak for the beta trade is closed. The easy money, the kind you made by just owning anything AI-related, has been made. The next phase is about alpha—finding the specific names where earnings are recovering but the stock price hasn't caught up.
The Core: Where the Cracks Are Forming
This is where the analysis gets granular. Goldman's positioning shifts are not random. They are a map of where smart money thinks the next bottleneck and the next windfall will be.
Signal 1: Semiconductors Are Now a Short. This is the most jarring data point. Semiconductors and the broader AI complex have been added to Goldman's short basket. The very sector that defined the AI trade—Nvidia and its ilk—is now a candidate for underperformance. Why? The market is pricing in a few scenarios: export controls are capping the addressable market, cloud giants might be slowing their capex growth, and custom ASICs are starting to eat into the general-purpose GPU's dominance. The 'pick and shovel' sellers are no longer a guaranteed bet.
Signal 2: Software Took the Crown in Momentum. In a stark reversal, software has overtaken semiconductors as the largest weight in the three-month momentum long portfolio. The market's momentum engine is now betting on the 'gold miners'—the application layer—over the 'shovel sellers.' This reflects a growing belief that AI is moving from capability demos to actual revenue contributions. AI agents, coding assistants, and enterprise SaaS are starting to show up in income statements. The market is starting to pay for adoption, not just infrastructure.
Signal 3: Storage and Data Centers Are the 'Most Attractive Tactically.' This is the hidden gem in the report. Goldman flags storage and data centers as having the most obvious valuation gap—profit recovery that isn't yet reflected in the stock prices. This aligns with my own read on the tech stack. We are transitioning from a training-dominated AI cycle to an inference-dominated one. Inference is far more demanding on memory bandwidth and data throughput. Model weights, KV caches, and retrieval-augmented generation (RAG) databases all need massive, high-speed storage. The data centers that host this infrastructure are seeing utilization and pricing power return. The market is still pricing them like legacy IT, not like the critical backbone of the AI economy.
Signal 4: The Capital Rotation. Goldman notes capital is rotating into previously ignored areas: European and Japanese banks, gold miners, and copper stocks. This is a tell. It suggests the AI trade is getting crowded, and the marginal dollar is looking for value elsewhere. The copper mention is particularly telling—it's a play on the electrification of everything, including AI data centers' insatiable power needs. The market is broadening its definition of what counts as an 'AI trade.'
The Contrarian Angle: The Narrative-Reality Divergence
Here's where the mainstream takes on this report miss the mark. Most will frame this as 'Goldman says AI is over.' That's lazy. The more accurate read is that Goldman is flagging a systemic shift in how the AI value chain is being priced. The market is moving from paying for potential to paying for realized cash flow.
But there's a deeper, more uncomfortable layer. The move into storage and data centers is a bet on the physical world. It's a bet that the 'digital gold rush' requires an enormous amount of physical infrastructure—silicon for memory, steel and concrete for data halls, and copper for power distribution. The market is quietly admitting that the AI revolution is, at its core, a massive industrial build-out. This is a far less sexy but far more concrete reality than the 'digital intelligence' narrative.
My own experience auditing infrastructure projects tells me this is the correct read. For years, I've watched teams focus on the flashy smart contract layer while ignoring the oracle and data availability problem. The same is happening here. Everyone was watching the GPU; the real bottleneck is the memory bus and the power grid. If it isn't on-chain, it didn't happen. If it isn't in a data center, it isn't running.
The Takeaway: The Next Block to Watch
Speed is the only moat in a borderless war. The next major signal is Nvidia's Q2 earnings at the end of August. This is the catalyst that will determine if the deleveraging continues or if the infrastructure thesis gets a new bid. But the real alpha is in the shadows. Watch the storage names. Watch the data center REITs. Watch the momentum factor weekly—if software's relative strength vs. semiconductors persists, the rotation is real.
The AI trade isn't over. It's just getting harder. The question now is whether you can adapt to the new indexing, or if you'll get front-run by your own assumptions. The truth is hidden in the block height—and this block just got a lot more complex.
