Hook: The Signal No One Wants to See
A week ago, the high-beta momentum basket bled out 12%. The Goldman Sachs AI hedge basket? Down 10% in five days. Leverage on the AI trade got wrecked from extreme highs. And yet, the bank's latest note doesn't scream 'bubble pop.' It whispers something far more calculated: The free money phase of the AI trade is over. The stock-picker's market has begun.
I've watched this playbook before. It's not the exit. It's the reset. The market is moving from buying the whole sector to buying the one name that prints. Goldman's not leaving the party. They're just throwing out the cheap beer.
Context: The Game Has Changed
For two years, the AI narrative was simple: buy anything with a chip, a GPU, or a 'cloud' in the name. The 2023-2024H1 phase was pure beta. Liquidity waves lifted every boat, from the semiconductor kings to the laggard storage players. But the August 23rd positioning report from Goldman signals a structural shift in the capital flows. The 'AI trade' isn't dead. It's just no longer a tide that lifts all boats. It's a scalpel.
This is the pivot from 'narrative-driven' to 'fundamentals-driven' pricing. The market is now asking one question: Show me the revenue. The AI narrative has been priced in; the AI earnings are just starting to show up. The disconnect between price and profit is the new trading floor.
Core: The Data Behind the New Map
The key facts are in the factor flows. Goldman's own playbook is now openly contradictory to the last two years of consensus.
- Semiconductors are in the short basket. That's not a typo. The same sector that minted millionaires in 2023 is now a target for professional shorts. The AI complex is too. Why? Because the price of the 'picks and shovels' is now assuming a gold rush that might not yield gold for everyone. The consensus has moved from if AI is real to who actually profits. And the market is betting the GPU monopoly gets a lot more crowded.
- Software is the new momentum king. For the first time in three months, Software has replaced Semiconductors as the largest weight in the 3-month momentum long basket. That is a massive signal. It means the market is pricing that the application layer will capture the value, not just the infrastructure. The ones selling the shovels are getting out-earned by the ones selling the maps to the gold.
- Storage and Data Centers are the 'tactically most attractive.' The report points to a specific divergence: 'profit recovery not yet fully reflected in stock prices.' This is where the forensic analysis gets juicy. AI inference is the new phase. It needs more than compute. It needs memory. It needs bandwidth. It needs the physical space to hold the weights. The profit for these entities is coming in, but the stock price hasn't caught up.
This is the core insight. The AI trade is in Phase 2. Phase 1 was about who had the best GPU. Phase 2 is about who has the data center with power, the storage with capacity, and the software that actually works for the enterprise.
The Catalyst: Nvidia's Q2 Earnings
The next check is Nvidia's Q2 earnings, due at the end of August. That is the high-tension wire. If Nvidia beats on data center revenue and gives a strong guidance, that could be the catalyst for a final wave in the infrastructure. If it's soft? The shorts on the semiconductor basket will be validated. The deleveraging could go from a 'correction' to a 'rout.' The market is waiting for that number. It's the last macro-level catalyst before the September industry conferences.
Contrarian: The Blind Spot Everyone's Missing
Here's the piece that no one is talking about. The Goldman report is bullish on 'profit recovery' in storage and data centers. But what if that profit recovery isn't from AI at all? It's a trap. The 'recovery' could be driven by the traditional IT spend cycle. The enterprise refresh cycle. The cloud service providers (CSPs) are still spending on traditional servers to run their own legacy apps.
Hype is a trap; data is the only map I trust. If you strip out the AI hype, the storage cycle is notoriously cyclical. The current 'recovery' might be the tail end of a normal 3-year refresh. If that's the case, the AI narrative is being conflated with a cyclical bump. You are buying a cycle, not a structural change. The real AI-driven storage demand (HBM, cache, model weights) is real, but it's still a fraction of the total market. The 'profit recovery' could be a red herring.
Also, the rotation out of AI into European banks, gold miners, and copper miners is a signal of crowding. It's the 'smart money' taking profits and moving to value. That's not a 'melt-up' signal. It's a sign of the last marginal dollar being in. The AI sector is getting crowded, and the excess capital is now chasing the neglected.
Takeaway: The Watchlist
I've seen this movie before. The move from 'beta' to 'alpha' is the most dangerous time for the passive investor. The market is now going to be a show-me-the-money game.
- Watch the Nvidia earnings guidance. That will set the tone for the entire sector for the next quarter. If the forward guidance is strong, the 'deleveraging' is over. If it's weak, the shorts have a new fuel.
- Watch the storage names. The divergence between the EPS and the price is the biggest arbitrage opportunity I see. If the profit recovery is real and AI-driven, the repricing will be violent. If it's cyclical, the stock will catch down to the price.
- Watch the momentum factor. The Software > Semiconductor rotation is a canary in the coal mine. If software names start to stall, the whole 'application layer' thesis cracks.
The next 30 days will define the second half of 2024. The AI trade isn't dead, but it's no longer for the lazy. Execution is everything. Data over drama. Always.
Are you ready for the split?