The Great Unwind: Goldman Sachs, AI Deleveraging, and the Ghost of Value in a Decentralized Void
What if the most important signal in the AI trade isn't the price of Nvidia shares, but the quiet rotation happening in Goldman Sachs' momentum portfolios? Consider this: while the financial press obsesses over whether we're in an AI bubble, the real story is a structural shift in how capital allocates across the AI stack. Over the past seven days, the market has witnessed something peculiar—not a crash, not a melt-up, but a surgical repositioning that tells us more about the next twelve months than any single earnings report could.
On August 23rd, Goldman Sachs released a note that cut through the noise. Their message: the AI trade isn't over, but the era of indiscriminate beta is. The high-beta momentum basket fell 12% in a week. The AI hedge basket dropped 10% in five days. Leverage is coming off the table. But here's the counter-intuitive part—Goldman isn't calling for a crash. They're calling for a transition. The first phase of the AI trade, where you bought anything with a GPU and watched it appreciate, is dead. The second phase, where you need to actually identify which parts of the stack generate real earnings, has begun.
This is the narrative shift I've been tracking since the 2020 DeFi yield farming mania. Back then, the story was 'liquidity mining APY is free money.' It wasn't. It was a subsidy for TVL numbers. Today, the story is 'AI is the new internet.' Maybe. But the market is starting to ask a very uncomfortable question: which parts of the AI stack actually make money, and which are just burning capital to look relevant?
Let me break down what Goldman is actually saying, because the surface-level takeaway—'AI is still good, just be selective'—misses the deeper mechanics. The note identifies three specific moves: semiconductors and AI complexes are now in the short basket, software has replaced semiconductors as the largest weight in the three-month momentum long basket, and storage and data centers are 'tactically the most attractive sectors' because their 'profit recovery hasn't been fully reflected in stock prices.'
Each of these moves deserves scrutiny. The semiconductor short is the most aggressive signal. For two years, semis were the unquestioned kings of the AI trade. Nvidia's dominance was treated as an axiom, not a hypothesis. Now, Goldman is telling clients to short them. This isn't a tactical call on a single quarter—it's a structural re-rating. The market is beginning to price in the possibility that Nvidia's moat is narrowing. AMD's MI series is gaining traction. Custom ASICs are eating into general-purpose GPU demand. Cloud providers are designing their own silicon. And export controls are shrinking the addressable market for high-end chips. The era of 'Nvidia or nothing' is ending, and the market is adjusting to a world where the AI chip market is contested.
The software rotation is equally telling. Momentum factors don't lie—they reflect where capital is actually flowing. Software becoming the largest weight in the momentum long basket means that, over the past three months, software stocks have outperformed semiconductors. This is a bet on AI application-layer monetization. The market is saying: 'We've priced in the picks and shovels. Now we want to see the gold.' AI programming assistants, AI agents, enterprise AI SaaS—these are the categories where revenue is starting to show up. But here's the catch: software is a crowded trade. Everyone sees the same momentum data. The question is whether the AI application layer can deliver earnings growth that justifies the rotation, or whether this is just another narrative shift that will reverse when the next earnings season disappoints.
Then there's the storage and data center call. This is the most interesting signal, because it's not obvious. Storage has been a value trap for a decade. The memory industry is a cyclical nightmare, with three players (Samsung, SK Hynix, Micron) engaged in a perpetual game of capacity chicken. But AI changes the calculus. HBM (high-bandwidth memory) is now a critical component for AI training chips, and the supply is highly concentrated. Enterprise SSDs are seeing demand from AI inference workloads that require massive data retrieval. And data centers—the physical infrastructure of AI—are seeing utilization rates and rental prices climb as inference demand scales. Goldman's call is that these profit recoveries haven't been fully priced in. The market is still treating storage and data centers as 'old tech,' while Goldman sees them as 'new AI infrastructure.'
But let me push back on this, because my job isn't to parrot Goldman's talking points. The 'profit recovery' in storage and data centers might not be as AI-driven as the narrative suggests. Traditional enterprise IT spending is recovering. Cloud providers are in a capex cycle. The question is how much of the storage recovery is AI-specific and how much is just the cyclical rebound of a depressed industry. If it's the latter, the 'valuation gap' Goldman identifies might be a value trap, not an opportunity. The market isn't always wrong about these things.
Now, let's zoom out and look at the broader picture. Goldman's note is a microcosm of a larger shift in how the market values AI. The first phase of the AI trade (2023-2024) was narrative-driven. The market was paying for 'AI vision'—the potential of the technology, not its current earnings. This is typical of any transformative technology cycle. In the late 1990s, the market paid for 'internet vision' and got burned when the earnings didn't materialize. In 2017, the market paid for 'blockchain vision' and got burned when the ICO bubble burst. The question is whether AI is different.
Goldman's answer is a qualified yes. They're not calling for a crash. They're saying the trade is transitioning from 'narrative' to 'fundamentals.' This is the 'profit realization' phase. The market is starting to demand that AI companies actually make money, not just tell stories. This is healthy, but it's also painful. The transition from narrative to fundamentals is always accompanied by volatility, because the market has to reprice everything.
Here's where my experience as a crypto analyst comes in. I've seen this movie before. In 2021, the NFT market was the hottest trade in crypto. Everyone was buying Bored Apes and CryptoPunks, convinced they were buying 'digital art.' My research showed they were actually buying 'digital status symbols'—tribal identity markers, not art. The market eventually figured this out, and the NFT bubble burst. The same dynamic is playing out in AI. The market is starting to figure out which parts of the AI stack are 'real' (generating earnings) and which are 'narrative' (burning capital).
The contrarian angle here is that the semiconductor short might be overdone. Goldman is a sophisticated player, but they're also a momentum chaser. The same factors that drove semis into the short basket could reverse quickly if Nvidia's Q2 earnings (due at the end of August) blow past expectations. Nvidia has a habit of surprising to the upside. If they do, the short basket could get squeezed, and the momentum rotation back into semis could be violent. The market is positioned for a slowdown in AI capex, but what if the opposite happens? What if the cloud providers' AI capex cycle is just getting started? The bears are betting on a slowdown, but the data doesn't clearly support that yet.
Another contrarian angle: the capital rotation out of AI into European and Japanese banks, gold miners, and copper stocks. Goldman mentions this as a sign that capital is 'spilling over' into neglected areas. But this could also be a sign of AI fatigue. When capital starts rotating into value sectors, it often means the growth trade is losing momentum. The AI trade has been the dominant growth trade for two years. If capital is now looking for value elsewhere, it might be a signal that the AI trade is entering a consolidation phase, not just a rotation.
Let me also address the elephant in the room: the 'AI bubble' debate. Goldman explicitly says the AI trade isn't over. But they're a sell-side firm with significant investment banking relationships with AI companies. They have a vested interest in maintaining bullish sentiment. My own view, based on 29 years of observing technology cycles, is that we're not in a bubble—yet. But we're in a 'froth' phase. The difference is that bubbles are characterized by a disconnect between prices and fundamentals that eventually corrects violently. Froth is characterized by elevated valuations that are still supported by real earnings growth. The AI trade is frothy, but not bubbly. The key risk is if AI earnings growth slows faster than expected. If Nvidia's guidance disappoints, or if the cloud providers' AI capex cycle peaks, the froth could turn into a bubble burst.
Now, let's talk about the specific opportunity in storage and data centers. Goldman's call here is based on a 'valuation gap'—the idea that profit recovery hasn't been fully reflected in stock prices. This is a classic value-investing signal. But it's also a classic value trap. The storage industry has a history of false recoveries. The memory cycle is brutal, and the three major players (Samsung, SK Hynix, Micron) have a history of over-investing and then cutting production to prop up prices. The AI-driven demand for HBM is real, but it's also a small portion of the overall memory market. The question is whether AI demand can offset the cyclicality of the traditional memory market.
My take: the storage trade is worth watching, but it's not a slam dunk. The data center trade is more interesting, because it's less cyclical. Data center REITs and IDC operators have long-term contracts, stable cash flows, and are benefiting from the secular trend of AI inference demand. The 'profit recovery' in data centers is more sustainable than in storage. But the market is also more efficient at pricing data center stocks, so the 'valuation gap' might be smaller.
Let me also address the 'software rotation' from a different angle. Software becoming the largest weight in the momentum long basket is a signal that the market is betting on AI application-layer monetization. But the software sector is also facing headwinds. Interest rates are still elevated, which pressures high-multiple software stocks. And the AI application layer is still nascent. Most AI software companies are spending heavily on R&D and not yet generating meaningful revenue. The rotation into software might be premature. The market might be getting ahead of itself, pricing in AI application revenue that won't materialize for another 12-18 months.
This is where my 'Narrative Hunter' instinct kicks in. The market is always looking for the next narrative. The 'AI infrastructure' narrative (semis, data centers) is getting tired. The 'AI application' narrative (software) is fresh. But narratives don't always align with fundamentals. The market is rotating into software because it's the new story, not necessarily because the earnings are there. This is a classic narrative-driven trade, and it's vulnerable to disappointment.
Let me now bring this back to my own experience. In 2017, I audited the Parallax Coin whitepaper and found a critical flaw in their ZK-Snarks implementation. The market was hyped on privacy coins, but the technical reality didn't match the narrative. I published a 15-page rebuttal, and it went viral. The lesson: narratives can drive prices, but they can't sustain them. Eventually, the technical reality catches up. The same is true in AI. The narrative is 'AI is the new electricity.' The reality is that AI is a transformative technology, but it's also a capital-intensive one. The market is starting to realize that not every AI company will be a winner. The 'rising tide lifts all boats' phase is over. Now we're in the 'separating the wheat from the chaff' phase.
This is also reminiscent of the 2022 Terra/LUNA collapse. The narrative was 'algorithmic stablecoins are the future.' The reality was that the mechanism was a death spiral. I led a team that audited the peg mechanism and identified the flaw. The market eventually figured it out, and the collapse was spectacular. The lesson: when a narrative is too good to be true, it usually is. The AI trade isn't a death spiral, but it is a narrative that's being stress-tested. The market is asking: 'Where is the real value?' And the answer is: 'It's in the companies that are actually generating earnings, not just telling stories.'
So, what's the takeaway? The AI trade is entering a new phase. The era of indiscriminate beta is over. The era of selective alpha has begun. Goldman's note is a roadmap for this transition. The key signals are: (1) semis are being shorted, (2) software is the new momentum favorite, (3) storage and data centers are the value plays, and (4) capital is rotating into neglected sectors. But these signals are not without risk. The semiconductor short could reverse. The software rotation could be premature. The storage and data center value plays could be value traps. And the capital rotation could be a sign of AI fatigue, not just rotation.
My advice: don't chase the momentum. Don't buy the narrative. Look for the companies that are actually generating earnings. The AI trade is not over, but it's changing. The winners will be the companies that can translate AI hype into real revenue. The losers will be the companies that are just riding the narrative. This is the 'profit realization' phase, and it's where the real money is made—or lost.
Let me also address the broader implications for the crypto market. The AI trade and the crypto market are increasingly intertwined. AI agents are transacting on-chain. AI-generated content is creating new NFT markets. AI-driven trading algorithms are influencing crypto prices. The 'AI x Crypto' intersection is one of the most exciting areas of the market. But it's also one of the most overhyped. The same dynamics that are playing out in the AI stock market are playing out in the AI crypto market. The narrative is ahead of the fundamentals. The market is paying for 'AI vision' in crypto, but the earnings are still nascent.
My view: the 'AI x Crypto' trade is a long-term opportunity, but it's a short-term risk. The market is overhyped, and there will be a correction. But the correction will be healthy. It will separate the real projects from the vaporware. The projects that are actually building useful AI applications on-chain will survive. The projects that are just adding 'AI' to their name to pump the token will die. This is the same pattern we saw in the DeFi summer of 2020 and the NFT mania of 2021. The narrative drives the price, but the fundamentals determine the long-term value.
Now, let me get into the technical details of the Goldman note, because there's more here than meets the eye. The 'high-beta momentum basket' falling 12% in a week is a significant move. This basket is composed of the highest-beta stocks in the market, which are typically the most speculative. A 12% weekly decline is a classic deleveraging signal. It means that leveraged players are being forced to sell, and the selling is concentrated in the most speculative names. This is the 'air coming out of the balloon' phase. The AI hedge basket falling 10% in five days is even more significant. This basket is specifically designed to capture AI-related alpha. A 10% decline in five days suggests that the AI trade is being unwound, and the unwinding is happening faster than the market can absorb.
But here's the key: Goldman is not saying the AI trade is over. They're saying the 'easy money' phase is over. The first phase of the AI trade was characterized by 'beta'—you bought any AI stock and it went up. The second phase is characterized by 'alpha'—you need to be selective. This is a classic market cycle. The 'beta' phase is driven by liquidity and narrative. The 'alpha' phase is driven by fundamentals and differentiation. The transition from beta to alpha is always accompanied by volatility, because the market has to reprice everything.
The 'storage and data centers' call is the most interesting part of the note. Goldman is saying that these sectors have 'profit recovery' that hasn't been fully reflected in stock prices. This is a classic value signal. But it's also a classic value trap. The storage industry is cyclical, and the 'profit recovery' might be a cyclical rebound, not a structural shift. The data center industry is less cyclical, but it's also more efficiently priced. The 'valuation gap' might be smaller than Goldman suggests.
Let me also address the 'capital rotation' signal. Goldman mentions that capital is rotating into European and Japanese banks, gold miners, and copper stocks. This is a classic 'risk-off' rotation. When capital rotates out of growth and into value, it's often a sign that the market is becoming more cautious. The AI trade has been the dominant growth trade for two years. If capital is now rotating into value, it might be a sign that the AI trade is losing momentum. But it could also be a sign of 'broadening'—the market is becoming healthier, with more sectors participating in the rally. The key is to watch whether the rotation is a short-term tactical move or a long-term strategic shift.
Now, let me talk about the 'catalyst' angle. Goldman identifies Nvidia's Q2 earnings (due at the end of August) and September industry conferences as key catalysts. This is a classic 'event-driven' approach. The market is waiting for a signal to determine the direction of the AI trade. If Nvidia beats expectations, the AI trade could resume its upward trajectory. If Nvidia disappoints, the AI trade could face a more significant correction. The September conferences are also important, because they provide a platform for AI companies to announce new products and partnerships. The market will be watching for any signs of slowing demand or intensifying competition.
My view: Nvidia's earnings are the single most important catalyst for the AI trade. Nvidia is the bellwether for the entire AI stack. If Nvidia's guidance is strong, it will validate the AI trade and could trigger a rally. If Nvidia's guidance is weak, it could trigger a sell-off. The market is positioned for a slowdown, but Nvidia has a history of surprising to the upside. The key metric to watch is Nvidia's data center revenue, which is the core of its AI business. If data center revenue growth is accelerating, the AI trade is healthy. If it's decelerating, the AI trade is in trouble.
Let me also address the 'momentum factor' angle. The fact that software has replaced semiconductors as the largest weight in the three-month momentum long basket is a significant signal. Momentum factors are based on price performance, and they reflect where capital is flowing. The shift from semis to software suggests that the market is rotating from 'AI infrastructure' to 'AI applications.' This is a classic 'late-cycle' rotation. In the early stages of a technology cycle, the market buys the infrastructure (semis, data centers). In the later stages, the market buys the applications (software). The shift from semis to software suggests that the market believes the AI infrastructure buildout is largely complete, and the next phase is the application layer.
But this rotation is not without risk. The software sector is facing headwinds from elevated interest rates. High-multiple software stocks are sensitive to interest rates, and if rates stay elevated, the software rotation could stall. Additionally, the AI application layer is still nascent. Most AI software companies are not yet generating meaningful revenue. The market might be getting ahead of itself, pricing in AI application revenue that won't materialize for another 12-18 months.
Now, let me bring this back to my own framework. I've been analyzing technology cycles for 29 years, and I've seen this pattern before. The 'narrative' phase is always followed by the 'fundamentals' phase. The market pays for 'vision' first, then demands 'earnings.' The AI trade is in the transition phase. The narrative is still strong, but the market is starting to demand earnings. This is healthy, but it's also painful. The transition is always accompanied by volatility, because the market has to reprice everything.
The key question is: which companies will survive the transition? The answer is: the companies that are actually generating earnings. In the AI stack, the companies that are generating earnings are the infrastructure providers (Nvidia, TSMC, the data center operators) and the application providers (Microsoft, Google, the AI software companies). The companies that are not generating earnings are the 'narrative' plays—the companies that are telling AI stories but not delivering AI revenue. These companies will be punished in the 'fundamentals' phase.
Let me also address the 'decentralized void' angle. I've been writing about the 'ghost of value in a decentralized void' for years. This is the idea that in a decentralized market, value is often a narrative construct, not a fundamental reality. The AI trade is a perfect example. The market is paying for 'AI vision,' but the 'value' is still a narrative. The 'fundamentals' phase is when the market starts to demand real value, not just narrative. This is the 'ghost of value' becoming 'real value.' The transition is painful, but it's necessary.
In conclusion, the Goldman note is a roadmap for the next phase of the AI trade. The key signals are: (1) semis are being shorted, (2) software is the new momentum favorite, (3) storage and data centers are the value plays, and (4) capital is rotating into neglected sectors. But these signals are not without risk. The semiconductor short could reverse. The software rotation could be premature. The storage and data center value plays could be value traps. And the capital rotation could be a sign of AI fatigue, not just rotation.
My advice: don't chase the momentum. Don't buy the narrative. Look for the companies that are actually generating earnings. The AI trade is not over, but it's changing. The winners will be the companies that can translate AI hype into real revenue. The losers will be the companies that are just riding the narrative. This is the 'profit realization' phase, and it's where the real money is made—or lost.
As for the crypto market, the 'AI x Crypto' intersection is a long-term opportunity, but it's a short-term risk. The market is overhyped, and there will be a correction. But the correction will be healthy. It will separate the real projects from the vaporware. The projects that are actually building useful AI applications on-chain will survive. The projects that are just adding 'AI' to their name to pump the token will die. This is the same pattern we saw in the DeFi summer of 2020 and the NFT mania of 2021. The narrative drives the price, but the fundamentals determine the long-term value.
So, what's the next narrative? The next narrative is 'AI applications.' The market is rotating from 'AI infrastructure' to 'AI applications.' The companies that are building useful AI applications—whether in software, crypto, or any other sector—will be the winners. The companies that are just telling AI stories will be the losers. This is the 'profit realization' phase, and it's where the real money is made—or lost.
Chasing the ghost of value in a decentralized void, I've learned that the only way to survive is to focus on fundamentals. The narrative will always be there, but the fundamentals are what matter in the long run. The AI trade is entering the 'fundamentals' phase, and the winners will be the companies that can deliver real earnings. The losers will be the companies that are just riding the narrative. This is the 'profit realization' phase, and it's where the real money is made—or lost.
Let me leave you with this: the AI trade is not over, but it's changing. The era of indiscriminate beta is over. The era of selective alpha has begun. The market is demanding earnings, not just narratives. The companies that can deliver earnings will be rewarded. The companies that can't will be punished. This is the 'profit realization' phase, and it's where the real money is made—or lost. The ghost of value is becoming real. Are you ready?