The numbers landed like a block confirmation: the AI hedge basket fell 10% in five days. The high-beta momentum basket dropped 12%. For anyone tracking capital flows the way a cartographer tracks river systems, this was not a crash. It was a redirection. Goldman Sachs did not declare the AI trade dead. They declared it rotating. And in that rotation lies a structural signal that most retail portfolios have not yet priced.
The Context: What Goldman Actually Said
Let me strip the noise from the signal. Goldman's core observation is that the era of indiscriminate AI buying is over. You can no longer buy the entire AI complex and expect alpha. The momentum factor—that quantitative measure of which assets have been moving—has shifted in a way that is both subtle and profound.
Software has replaced semiconductors as the largest weight in the three-month momentum long basket. Meanwhile, semiconductors and the AI complex have moved into the short basket. This is not a footnote. This is a pivot point printed in the data.
Goldman is also pointing capital toward storage and data centers, arguing that the "profit recovery" in these sectors has not yet been fully reflected in stock prices. In other words, the market has been so fixated on GPU suppliers that it has ignored the physical infrastructure layer that makes AI actually function. Storage. Memory. Data center capacity. These are the picks-and-shovels of the AI gold rush, and they are trading at a discount to their earnings trajectory.
The catalysts are clear: Nvidia's Q2 earnings and the September industry conferences. These events will determine whether the AI trade re-accelerates or enters a deeper consolidation phase.
The Core: Reading the Rotation as a Macro Signal
Let me take off the equity analyst hat for a moment and put on the blockchain infrastructure one. Because what Goldman is describing is not unique to traditional markets. It is the same pattern we have seen in crypto cycles: the narrative phase gives way to the infrastructure phase, and the infrastructure phase rewards different assets than the narrative phase did.
The Momentum Rebalance
The momentum factor is a lagging indicator. It tells you where capital has been, not where it is going. But when a factor shift is this pronounced—software over semiconductors, storage over compute—it suggests a fundamental re-pricing of where value accrues in the AI stack.
Consider the analogy to crypto: In 2020, DeFi protocols with no revenue were trading at astronomical multiples. The narrative was "DeFi will replace banks." Then the market matured, and the infrastructure layer—oracles, bridges, aggregators—started capturing value. The same rotation is happening in AI. The market has decided that the GPU buildout is priced in, and it is now looking for the layers that have been ignored.
The architecture of value hidden beneath the hype is shifting from compute to memory and physical infrastructure.
This is where my technical background forces me to pause. Storage is not a passive beneficiary of AI. It is a bottleneck. The training runs that produced GPT-4 required massive amounts of high-bandwidth memory (HBM). Inference workloads are even more memory-intensive. When Goldman recommends storage and data centers, they are implicitly betting that the AI compute buildout has created a downstream demand shock that has not yet been fully transmitted to the companies that make the memory, the storage arrays, and the physical facilities.
The Capital Rotation
The more interesting signal is where capital is flowing out of AI and into: European and Japanese banks, gold miners, and copper miners. This is not a flight to safety in the traditional sense. It is a search for value in sectors that have been neglected during the AI mania.
But there is a deeper read here. Copper miners are not a defensive play. Copper is an AI play. Data centers consume enormous amounts of copper for electrical infrastructure, cooling systems, and networking. Gold miners are a hedge against monetary debasement, which is always relevant when you are dealing with a global liquidity cycle. And European and Japanese banks? Those are bets on a global economic recovery that has not yet been fully priced.
The ledger does not lie: capital is not leaving AI because AI is dead. It is leaving because the marginal dollar is better deployed elsewhere until the next catalyst.
This is the same pattern we see in crypto when Bitcoin dominance rises during altcoin corrections. Capital is not leaving the asset class; it is consolidating into the most liquid, most institutionally recognized expression of the thesis.
The Leverage Question
The 10% and 12% drawdowns in AI baskets are not just price movements. They are deleveraging events. When a momentum basket drops that quickly, it means leverage is being unwound. Positions are being liquidated. The question is whether the deleveraging is complete.
My read from the data: the extreme positioning has been reduced, but the system has not fully normalized. If Nvidia's earnings disappoint—even slightly—the second round of deleveraging could hit the storage and data center names that Goldman is recommending. Correlation is a cruel mistress. When the market sells AI, it tends to sell all of AI, regardless of fundamental differentiation.
This is where defensive positioning matters. I wrote about this during the Terra collapse in 2022: the prerequisite for long-term alpha is survival. You cannot capture the recovery if you are liquidated in the drawdown.
The Contrarian Angle: What Goldman Isn't Telling You
Goldman is a sell-side institution. Their clients include the very companies they are recommending. That does not invalidate the analysis, but it does require a discount rate on their enthusiasm. Let me offer a few contrarian observations that the report does not address.
The Decoupling Illusion
The market is treating storage and data centers as a separate trade from semiconductors. But they are not separate. A data center without GPUs is a warehouse. Storage without compute is a vault. The AI stack is an integrated system, and the components are highly correlated in terms of demand.
If Nvidia's guidance next quarter suggests a slowdown in AI capital expenditure—perhaps due to export controls or supply chain constraints—the storage and data center thesis breaks down. The "profit recovery" that Goldman sees has not happened yet. It is an expectation. And expectations are fragile things.
The Energy Blind Spot
Copper miners are mentioned in the report as a destination for rotated capital. But the deeper signal is energy. AI data centers consume enormous amounts of electricity. The buildout of AI infrastructure is, at its core, an energy play. The market has not fully priced the energy constraints on AI growth.
This connects to the crypto thesis directly: proof-of-work mining taught us that computational infrastructure is energy-constrained. The same physics applies to AI data centers. If energy supply cannot keep pace with AI demand, the growth rate of the entire AI complex will be capped. And the companies that benefit most are not the chipmakers or the software firms—they are the utilities, the energy infrastructure companies, and yes, the copper miners who supply the grid.
The Software Skepticism
Goldman's momentum factor now favors software over semiconductors. I am skeptical. Software companies have been the most aggressive in wrapping themselves in the AI narrative, often with minimal actual AI revenue. The momentum shift may reflect narrative capture rather than fundamental improvement.
In crypto, we saw the same phenomenon during the ICO boom: projects with no product and no revenue were trading at billion-dollar valuations because they had "blockchain" in their whitepaper. The software trade in AI has a similar feel. Some of these companies are real. Many are not. The market will eventually separate them, but the momentum factor is not the tool for that separation.
The Takeaway: Positioning for the Pivot
Predicting the pivot before the pivot is printed is the skill that separates surviving investors from liquidated ones. The pivot here is clear: AI is no longer a monolithic trade. It is a differentiated market where infrastructure, software, and compute will diverge based on their ability to convert narrative into earnings.
Here is my positioning framework:
First, respect the deleveraging risk. The AI complex is still carrying elevated leverage. If Nvidia's earnings disappoint, the correlation between storage, data centers, and semiconductors will spike to near-1.0, and everything will sell off together. Do not be caught long on the wrong side of that correlation spike.
Second, treat storage and data centers as a real but conditional opportunity. The fundamental thesis is sound: AI needs memory, and memory is scarce. But the timing is dependent on the broader AI capital expenditure cycle. Wait for the Nvidia earnings confirmation before deploying full size.
Third, watch the energy complex. The rotation into copper is not a rotation out of AI. It is a rotation into the physical constraints of AI. This is a trade that will persist for years, not quarters.
Fourth, maintain defensive optionality. In crypto, I learned that the best hedges are the ones you put on before the crisis, not during it. The same applies here. If you are long storage and data centers, consider a hedge against semiconductor weakness. If you are long AI broadly, consider a hedge against a second round of deleveraging.
The AI trade is not over. It is rotating. And in that rotation, the market is revealing its true priorities: Silence the noise, listen to the block height. The block height here is the earnings date. The confirmation is the guidance. Everything else is just noise.
Structure over sentiment. The structure says: infrastructure is underpriced, software is overhyped, and energy is the hidden constraint. That is the trade. That is the architecture beneath the hype.