The AI Trade Is Not Over—It's Rotating: Goldman's Quiet Signal on Where the Next Phase of Capital Flows

The numbers hit the terminal at 14:32 New York time. The Goldman Sachs prime brokerage desk had just pushed its weekly risk report to institutional clients, and the data inside contradicted every headline screaming about an AI bubble bursting. The AI hedge fund basket had shed 10% in five days. The high-beta momentum basket was down 12%. On the surface, this looked like capitulation—the end of the most crowded trade of the decade. But buried in the same report was a sentence that most retail investors would skim past: "The AI trade is not over." That sentence, sandwiched between liquidity metrics and sector rotation data, is the most important signal in this market right now. Not because it predicts a rebound. But because it reveals a structural shift in how institutional capital is positioning itself for the next phase of the AI cycle. The market is not abandoning artificial intelligence. It is repricing it. And the rotation happening beneath the surface tells you more about the next 12 months than any single earnings print. The context here matters more than the headline numbers. We are in a liquidity environment where the Federal Reserve has maintained a restrictive stance longer than most market participants anticipated. The yield curve has been inverted for over two years—the longest stretch in modern history—and yet equity markets have continued to climb a wall of worry. This is not a normal bull market. It is a liquidity-driven, narrative-fueled advance where capital flows have been concentrated in a narrow set of AI-related assets. The concentration risk has been building for months. When the AI hedge fund basket drops 10% in five days, it is not a random event. It is a deleveraging event. And deleveraging events in a concentrated market are rarely one-day affairs. They are processes. The Goldman report acknowledges this process is underway but frames it as a healthy correction within a longer-term structural trend. The key data point is the momentum factor rebalancing: software has replaced semiconductors as the largest weight in the three-month momentum long portfolio, while semiconductors and the AI complex have moved into the short portfolio. This is not a trivial shift. It represents a fundamental change in how systematic strategies are viewing the AI value chain. Let me be precise about what this means from a macro-liquidity perspective. I have spent the last decade analyzing cross-border payment flows and institutional capital allocation patterns, and the one lesson that has never failed me is this: liquidity dictates asset prices, not narratives. The narrative around AI has been overwhelmingly positive for two years. The liquidity, however, is now being reallocated. When momentum factors—which are essentially quantifiable measures of capital flow direction—start shifting from semiconductors to software, it tells you that the marginal buyer of AI chips is exhausted. The marginal buyer of AI software is just beginning to engage. This is a classic late-cycle rotation within a secular trend. The infrastructure has been built. The chips have been purchased. The data centers are under construction. Now the market is asking a harder question: who is going to use all this compute, and what will they pay for it? That question cannot be answered by another GPU shipment. It can only be answered by earnings reports from software companies that are actually monetizing AI capabilities. Goldman's specific recommendation is telling. They are pointing clients toward storage and data center names, arguing that the valuation gap is the most significant and that profit recovery has not yet been fully reflected in share prices. This is a direct challenge to the prevailing narrative that AI value creation is concentrated in the semiconductor layer. My own analysis of cross-border capital flows supports this view. I have been tracking institutional flows into European and Japanese banks, gold miners, and copper producers—sectors that have been ignored for years. The fact that these flows are accelerating while AI semiconductor flows are decelerating suggests a broader rotation is underway. This is not capital fleeing AI. It is capital seeking the next marginal opportunity. The AI trade is not over. It is rotating. And the rotation is happening along the lines of fundamental value creation rather than narrative momentum. The contrarian angle here is uncomfortable for most market participants. The conventional wisdom is that AI is a winner-take-all market dominated by a few semiconductor names. The data suggests otherwise. The momentum factor shift indicates that software companies are now viewed as having more certain earnings visibility than chipmakers. This is counterintuitive because software has been the laggard in the AI trade. But it makes perfect sense from a liquidity perspective. The capital expenditure cycle for AI infrastructure is peaking. The operational expenditure cycle for AI applications is just beginning. When you are in the middle of a transition from capex to opex, the companies that benefit are not the ones selling the picks and shovels. They are the ones using the picks and shovels to build something that generates recurring revenue. This is the same pattern we saw in the early 2000s with the internet. The infrastructure providers—Cisco, Lucent, Nortel—were the darlings of the market in 1999. By 2002, they were trading at a fraction of their peak valuations. Meanwhile, the companies that actually used the internet to transform their business models—Amazon, Google, eBay—were just beginning their secular ascent. The AI trade is following the same arc. The infrastructure phase is maturing. The application phase is beginning. My own experience in the 2022 bear market taught me a brutal lesson about liquidity risk. When Terra/Luna collapsed, I was tracking stablecoin de-pegging risks and centralized exchange insolvency in real time. The pattern was clear: capital flight from risky assets was not a judgment on the underlying technology. It was a liquidity event. The same dynamic is playing out now. The AI hedge fund basket dropping 10% in five days is not a judgment on the long-term viability of artificial intelligence. It is a deleveraging event driven by position unwinding and risk parity rebalancing. The question is whether this deleveraging has run its course. Goldman's data suggests it has not fully completed. The high-beta momentum basket is still down 12%, which indicates that the most speculative positions are still being unwound. But the recommendation to buy storage and data center names suggests that the desk believes the next phase of the AI trade will be led by infrastructure beneficiaries that have not yet been fully repriced. Let me drill into the storage and data center thesis because it is the most actionable part of this analysis. The argument is straightforward: these companies have been overlooked because the market has been fixated on semiconductor names. But the profit recovery in storage and data centers is not yet reflected in their share prices. This is a classic value-plus-growth setup. The valuation gap is significant, and the earnings catalyst is approaching. The key question is whether this profit recovery is real or just another narrative. My analysis of the underlying fundamentals suggests it is real. The buildout of AI data centers is not slowing down. The demand for high-bandwidth memory, solid-state drives, and enterprise storage solutions is accelerating. The companies that supply these components—Micron, Dell, Super Micro—are seeing order books that extend well into 2025. The market has been so fixated on NVIDIA's earnings that it has ignored the fact that the entire AI infrastructure stack is experiencing demand pull. This is the information gain that most retail investors are missing. The AI trade is not just about GPUs. It is about the entire ecosystem of storage, networking, cooling, and power infrastructure that makes AI compute possible. The catalyst calendar is clear. NVIDIA's second-quarter earnings and the September industry conferences will be the inflection points. If NVIDIA delivers a strong print and maintains its capital expenditure guidance, the entire AI complex will rally. If the company disappoints, the deleveraging will resume. But here is the nuance that most analysts miss: even if NVIDIA disappoints, the storage and data center names may not sell off as much. The reason is that their earnings visibility is less dependent on a single product cycle. The demand for storage is driven by the cumulative buildout of AI infrastructure, not by the marginal performance of one chip. This is a diversification benefit that the market has not yet priced in. The risk is asymmetric. The upside is significant if the profit recovery materializes. The downside is limited if the AI trade continues to deleverage. This is the kind of risk-reward profile that institutional investors find attractive. I want to address the elephant in the room: the flow of capital into non-AI sectors. Goldman notes that money is rotating into European and Japanese banks, gold miners, and copper producers. This is being interpreted by some as a sign that the AI trade is over. I disagree. This is a sign that the market is broadening its opportunity set. The AI trade has been so dominant that it has crowded out almost everything else. Now that the AI trade is consolidating, capital is seeking the next marginal opportunity. The copper thesis is particularly interesting because it connects directly to AI infrastructure. Data centers consume enormous amounts of electricity, and the electrical infrastructure requires copper. The gold thesis is a hedge against the risk that the AI trade fails to deliver on its promises. The bank thesis is a bet on the normalization of the yield curve and the return of traditional financial intermediation. These are not anti-AI trades. They are complementary trades that balance the risk of an AI-specific drawdown. From a systemic risk perspective, I am watching the leverage levels in the AI complex. The deleveraging we have seen is healthy, but it may not be complete. The AI hedge fund basket is still down 10% from its peak, and the high-beta momentum basket is down 12%. These are significant drawdowns, but they are not capitulation-level moves. A true capitulation would be a 20-30% drawdown in a matter of days. We have not seen that yet. This suggests that there is still leverage in the system that needs to be unwound. The risk is that NVIDIA's earnings trigger a second wave of deleveraging if the results disappoint. The mitigation is that the storage and data center names have already been repriced to reflect a more conservative outlook. The market is not pricing in a complete collapse of AI infrastructure spending. It is pricing in a slowdown in the rate of growth. This is a subtle but important distinction. The AI trade is not over. It is maturing. My framework for navigating this environment is simple: focus on liquidity metrics, not narratives. The momentum factor data from Goldman is a liquidity metric. It tells you where capital is flowing, not where it should flow. The recommendation to buy storage and data center names is a liquidity signal. It tells you that the desk believes these names are under-owned relative to their earnings potential. The flow of capital into non-AI sectors is a liquidity signal. It tells you that the market is broadening its opportunity set. When you combine these signals, the picture is clear: the AI trade is entering a new phase. The phase of indiscriminate buying is over. The phase of selective accumulation is beginning. This is not a bearish signal. It is a maturation signal. The market is becoming more sophisticated in how it prices AI-related assets. This is a sign of a healthy market, not a bubble about to burst. Let me be direct about the risks. The first risk is that NVIDIA's earnings disappoint and trigger a second wave of deleveraging. This is a real risk, and it is the reason why I would not be aggressively adding to AI positions before the earnings print. The second risk is that the storage and data center profit recovery is slower than expected. This is a moderate risk, and it is the reason why I would focus on the highest-quality names in the sector. The third risk is a broader market liquidity event, such as a sudden tightening of financial conditions or a geopolitical shock. This is a low-probability but high-impact risk, and it is the reason why I would maintain some cash reserves. These risks are manageable. They are not reasons to avoid the AI trade. They are reasons to be selective and patient. The opportunity set is clear. The first opportunity is to build positions in storage and data center names that have not yet repriced their earnings recovery. The second opportunity is to use NVIDIA's earnings as a volatility event to enter or add to positions at more favorable prices. The third opportunity is to position for the rotation into non-AI sectors that are benefiting from the broadening of the market. These opportunities are not mutually exclusive. They can be pursued simultaneously with appropriate position sizing and risk management. The time horizon is 3-6 months for the storage and data center trade, 1-2 weeks for the NVIDIA earnings trade, and 6-12 months for the non-AI rotation trade. This is a diversified approach that balances short-term catalysts with medium-term structural trends. I have been analyzing cross-border payment flows and institutional capital allocation for over a decade. I have seen multiple market cycles, from the ICO boom of 2017 to the DeFi summer of 2020 to the NFT mania of 2021 to the liquidity crisis of 2022. The one constant across all of these cycles is that liquidity dictates asset prices. The narrative is always compelling. The liquidity is always the truth. Right now, the liquidity is telling us that the AI trade is not over. It is rotating. The rotation is from semiconductors to software, from infrastructure to applications, from AI-specific assets to AI-adjacent assets. This is a healthy development. It is a sign that the market is becoming more sophisticated in how it prices the AI opportunity. The investors who understand this rotation will be positioned for the next phase of the AI trade. The investors who are fixated on the narrative will be left behind. The takeaway is not to abandon the AI trade. The takeaway is to understand that the AI trade is evolving. The phase of indiscriminate buying is over. The phase of selective accumulation is beginning. The storage and data center names are the most attractive opportunity because their profit recovery is not yet fully priced. The NVIDIA earnings print is the key catalyst because it will determine whether the deleveraging continues or reverses. The rotation into non-AI sectors is a sign of market broadening, not a sign of AI's demise. The investors who understand these dynamics will be positioned for the next phase of the AI trade. The investors who are fixated on the narrative will be left behind. The question is not whether AI will transform the global economy. The question is whether you are positioned to capture the value creation that is happening right now. The data says the AI trade is not over. The data says it is rotating. The question is whether you are paying attention.