Hook: The Leverage Has Left the Building
Over the past week, the high-beta momentum portfolio tracked by Goldman Sachs fell 12%. The bank's own AI hedge basket dropped 10% in five days. These are not corrections. These are deleveraging events. When a cohort of traders who crowded into the same narrative on margin starts to unwind, the tape does not bleed—it gapes.
Logic does not bleed, but code leaves traces. And the trace here is unmistakable: the AI trade, as a monolithic bet, is over. Not the technology. Not the infrastructure buildout. The trade.
Goldman Sachs, in an August 23 note, delivered a message that most retail participants are not prepared to hear: AI trading is not finished, but the phase of extracting excess returns through broad sector beta is ending. The era of buying anything with "AI" in the ticker and watching it appreciate is gone. What remains is a more brutal, more technical game: stock selection, valuation discipline, and the uncomfortable work of reading actual earnings reports.
Context: The Anatomy of a Crowded Trade
To understand why Goldman's positioning signals matter, you have to understand how the AI trade was built. From late 2023 through mid-2024, the equity market experienced what can only be described as a liquidity-driven narrative feedback loop. The release of ChatGPT had ignited a race among hyperscalers—Microsoft, Google, Amazon, Meta—to secure GPU supply. Nvidia's data center revenue exploded, and the market extrapolated that growth curve into perpetuity.
The trade was simple: buy semiconductors, buy AI-enabled software, buy anything with GPU exposure, and short the old economy. Momentum factors rotated almost entirely into AI-adjacent names. The high-beta momentum portfolio—a quantitative construct that holds the most volatile, most trending stocks—became essentially an AI-only basket.
This worked beautifully until it didn't. The problem with momentum crowding is that it creates a structural fragility. When everyone holds the same position for the same reason, there is no marginal buyer left. The bid disappears exactly when it's needed most.
Goldman's data shows the unwind is underway. The high-beta momentum portfolio's 12% weekly decline is not a normal pullback—it's a forced deleveraging. Margin calls, risk parity adjustments, and quant factor rebalancing all compound the downside. The AI hedge basket falling 10% in five days tells us that even sophisticated investors who structured hedges are not immune to the cascade.
But here's what the market is missing: Goldman explicitly says the AI trade is not ending. It's evolving. The question is not whether AI infrastructure spending continues—it does. The question is whether the market can distinguish between companies that will capture value from that spending and companies that merely rode the narrative wave.
Core: The Sector Rotation Signal Most Investors Will Misread
Let me break down the specific positioning shifts Goldman identified, because they tell a coherent story about where the AI value chain is heading.
Semiconductors Enter the Short Book
This is the most significant signal. Goldman reports that semiconductors and AI complexes have entered the short portfolio. For the past 18 months, semiconductors were the momentum trade. They were the high-beta expression of AI optimism. Now, the smart money is positioned against them.
The interpretation requires nuance. This does not mean the AI chip cycle is over. It means the market has priced in the obvious growth and is now looking for the cracks. Consider the competitive dynamics: Nvidia's dominance is no longer uncontested. AMD's MI300 series has gained traction. Custom ASIC designs from Google (TPU), Amazon (Trainium), and Microsoft (Maia) are absorbing incremental workloads. And export controls to China have structurally limited the total addressable market for high-end GPUs.
The short positioning also reflects a cyclical concern. Semiconductor inventory cycles are notoriously violent. If hyperscaler capital expenditure growth decelerates—and there are early signs of digestion after two years of unprecedented buildout—the earnings revisions could be sharp. The market is not saying AI is dead. It's saying the semiconductor trade has become consensus, and consensus trades have poor risk-reward.
Software Takes the Top Weight in the Momentum Long Book
This is the more interesting signal. Goldman notes that software has replaced semiconductors as the largest weight in the three-month momentum long portfolio. This is a quant factor shift—it reflects actual price performance. Software stocks have been outperforming semiconductors over the past quarter.
What does this tell us? The market is shifting its AI value-capture thesis from "picks and shovels" (hardware) to "gold miners" (applications). The logic is straightforward: the infrastructure is being built. Now the question is who deploys that infrastructure to generate revenue. Companies with proprietary data, distribution channels, and integration capabilities are positioned to monetize AI.
This is also a maturity signal. The AI stack is moving from training to inference. Training requires massive GPU clusters—that's the semiconductor story. Inference requires software optimization, model deployment, and integration into existing workflows—that's the application story. As AI transitions from capability demonstration to revenue contribution, the value chain shifts.
Storage and Data Centers: The Unloved Middle Layer
Goldman identifies storage and data centers as the most tactically attractive sectors, with the valuation gap being most pronounced. This is the key insight that most market participants will overlook.
The narrative around AI has been dominated by GPUs. But every GPU cluster requires storage—for model weights, training data, inference caches, and checkpointing. The scale of data generated and processed by AI workloads is unprecedented. High-bandwidth memory (HBM) has become a critical bottleneck, and enterprise SSD demand is accelerating.
The data center story is similarly underappreciated. AI inference workloads have different requirements than training workloads. They require lower latency, higher density, and more distributed architectures. This creates opportunities for operators who can provide the right infrastructure mix.
Goldman's logic is that the profit recovery in these sectors has not yet been reflected in stock prices. In other words, the earnings are improving, but the market is still pricing these companies based on their pre-AI growth trajectories. This is a classic mispricing opportunity.
Based on my audit experience across crypto and traditional markets, this pattern is familiar. The market has a tendency to anchor on narratives and ignore the unglamorous components of the value chain. In crypto, we saw this with infrastructure tokens—the market overpaid for layer-1 protocols while ignoring the oracle and indexing layers that actually captured usage. The same dynamic is playing out in AI equities.
The Capital Rotation Beyond AI
Goldman also notes capital is rotating into previously overlooked areas: European and Japanese banks, gold miners, and copper stocks. This is a signal worth examining.
The rotation to banks suggests the market is positioning for a rate environment that benefits financials. The rotation to gold miners reflects persistent concerns about fiat debasement and geopolitical risk. The copper trade is the most interesting—copper is a critical material for data center electrical infrastructure. The market is essentially building a secondary AI trade: not the chips, but the physical inputs required to power and cool the data centers that run the AI.
This capital exodus from AI into value sectors also tells us something important: the AI trade has become crowded to the point where marginal capital is seeking less efficient markets. When the easy money has been made in a sector, the next wave of returns requires either deeper analysis (alpha) or rotation to less discovered opportunities.
Contrarian: What the Bulls Got Right
I've spent considerable time deconstructing the bearish signals, so it's only fair to examine the counter-argument. The AI skeptics have been wrong before, and they will be wrong again. Here's what the bulls understand that the shorts may be missing.
First, the AI buildout is still in its early innings. We are arguably in the equivalent of 1996 in the internet cycle—the infrastructure is being laid, but the killer applications that will justify the spending have not yet fully emerged. The hyperscaler capital expenditure cycle has historically persisted longer than market participants expect. When the market predicted a pullback in cloud capex in 2022, it continued to grow. The same dynamic may repeat.
Second, the revenue visibility for AI infrastructure is stronger than for any previous technology cycle. Nvidia's data center revenue is not speculative—it's backed by purchase orders from companies with real balance sheets. The AI hedge basket's decline does not invalidate the fundamental demand signal.
Third, the storage and data center thesis is fundamentally an earnings story. If the profit recovery is real—if these companies are actually generating higher margins and cash flows—then the valuation gap will close. The market can ignore earnings for a quarter or two, but not indefinitely.
Fourth, the software rotation may be premature. The momentum factor is a lagging indicator—it reflects what has already happened, not what will happen. The software outperformance may be a catch-up trade rather than a fundamental shift. If AI application revenue disappoints—and there are legitimate questions about monetization timelines—the software trade could reverse just as quickly as it emerged.
The contrarian view is not that AI is a bubble. The contrarian view is that the market's rotation signals may be over-reading short-term price action. The fundamental demand for AI compute is real, and the companies providing the infrastructure have pricing power that has not yet been fully reflected in their valuations.
The Macro Backdrop: Why This Matters for Crypto
As an on-chain detective, I see parallels between the current AI equity market and the crypto market's own cycles of leverage and deleveraging. The mechanics are identical: narrative-driven inflows create crowding, crowding creates fragility, and fragility eventually triggers a repricing event.
The crypto market has experienced this pattern repeatedly. The 2017 ICO mania was a narrative-driven bubble that collapsed when the marginal buyer disappeared. The 2021 NFT frenzy was a liquidity-driven market where wash trading created artificial volume. The 2022 Terra collapse was an algorithmic failure that was obvious to anyone who understood the mechanics of the peg.
The AI equity trade is following the same pattern, just with different instruments. The leverage is in the equity derivatives market rather than DeFi protocols. The crowding is in momentum factors rather than wallet clusters. But the underlying dynamics are identical.
This has implications for crypto investors. If AI equities continue to deleverage, the risk sentiment could spill over into crypto markets. The correlation between tech equities and crypto has been inconsistent, but in periods of stress, the correlation tends to increase. A continued selloff in AI names could create headwinds for crypto assets.
Conversely, the rotation into gold miners and copper stocks is a signal that some investors are positioning for a regime of higher inflation and real asset appreciation. This is traditionally a favorable environment for Bitcoin, which has increasingly been viewed as a hedge against monetary debasement.
The Nvidia Catalyst and the Path Forward
Goldman identifies Nvidia's Q2 earnings as the next catalyst. This is the event that will determine the near-term direction of the AI trade. The expectations are high—the market wants to see continued data center revenue growth, improved margins, and forward guidance that justifies the current valuation.
The risk is asymmetric. If Nvidia beats and raises, the AI trade could stabilize. If the guidance disappoints—even slightly—the selloff could accelerate. The market has priced in perfection, and perfection is a high bar.
But here's what the market is missing: Nvidia's earnings are a backward-looking indicator of the AI trade's health. The more important signal is the deployment rate of AI infrastructure. Are the data centers being built actually being utilized? Is the inference demand materializing? Are the enterprise AI pilots converting to production workloads?
The answer to these questions will determine whether the current storage and data center opportunity is real or a value trap. If AI workloads are growing as expected, the profit recovery in storage and data centers will be sustained. If the deployment is slower than expected, the valuation gap will close through price declines rather than earnings growth.
The Signal in the Noise
Let me step back and give you the framework I use when analyzing any market structure, whether it's a smart contract or a portfolio allocation.
First, identify the primary incentive. In the AI trade, the primary incentive was straightforward: capture the value of the most significant technological shift since the internet. The market priced this incentive with extreme enthusiasm, driving valuations to levels that assumed flawless execution.
Second, identify the fragility points. The AI trade's fragility was its crowding. When everyone holds the same position, the market becomes susceptible to cascading liquidations. The recent deleveraging is the manifestation of this fragility.
Third, identify the structural survivors. Not every company in the AI value chain will thrive. The winners will be those with real earnings, real cash flows, and real pricing power. The storage and data center companies Goldman highlights may be the structural survivors—they have tangible assets, recurring revenue, and competitive moats that are difficult to disrupt.
Fourth, identify the asymmetry. The market is currently pricing AI infrastructure companies with a risk premium that assumes the profit recovery will not materialize. If the recovery does materialize—and the evidence suggests it will—the upside is significant. The asymmetry is favorable.
The Takeaway: The AI Trade Is Not Dead, It's Maturing
The AI trade is entering its second phase. The first phase was characterized by broad beta—everything with AI exposure went up. The second phase is characterized by alpha—only companies that deliver actual earnings will outperform.
The market is transitioning from narrative-driven pricing to fundamentals-driven pricing. This is a healthy maturation, not a death spiral. The companies with real revenue, real margins, and real competitive positions will be rewarded. The companies that merely rode the narrative will be punished.
For investors, the signal is clear: the era of passive AI exposure is over. The era of active, research-driven selection has begun. The storage and data center trade is the current opportunity—the profit recovery is real, and the market has not yet priced it in.
Imagination is infinite, but liquidity is finite. The market is learning this lesson the hard way, but it will emerge stronger for the experience. The AI trade is not over. It's just getting more serious.
The question is not whether AI will transform the economy—it will. The question is whether you can identify the companies that will capture the value before the market does. The next quarter will separate the analysts from the momentum chasers.
Watch the Nvidia earnings. Watch the storage and data center earnings. Watch the momentum factor rotations. The market is telling you where the value is moving. The only question is whether you're listening.
Volume is noise; the wallet cluster is signal. In the AI trade, the wallet cluster is the earnings report. Follow the earnings, and you'll find the signal.