Hook
The AI hedge basket dropped 10% in five days. The high-beta momentum basket fell 12%. Semiconductors and AI complexes flipped to the short side of Goldman Sachs' three-month momentum portfolio. If you read the headlines, the conclusion writes itself: AI trade is over, run for the exits.
But Goldman's own analysts are saying the opposite. "The AI trade isn't finished," they wrote. The market just entered a phase where beta isn't free anymore. The alpha is now found in the friction — in the gap between what the price says and what the profit statements reveal.
Charts lie, but the on-chain wallets never sleep. The same logic applies to traditional equities, just with a different ledger.
Context
Let's be precise about what this report actually claims. Goldman's momentum factor analysis reveals a rotation, not a rejection. Software has replaced semiconductors as the largest overweight in the three-month momentum long book. Semiconductors and AI complexes now appear in the short book. Meanwhile, the bank's strategists recommend storage and data center names — Dell, Super Micro, Micron-type infrastructure plays — arguing their profit recovery hasn't been fully priced into the equity.
That's a critical nuance. The AI narrative isn't dead. It's just being repriced. The market is moving from "buy everything with AI in the ticker" to "buy what actually monetizes the compute cycle." The former is a beta trade. The latter is an alpha trade.
I've seen this pattern before, though the sector was different. In the DeFi summer of 2020, the market allocated capital to any protocol with a yield farm. I quantified the real yield versus the token emission schedule and found that 60% of liquidity providers were losing money after impermanent loss and token depreciation. The market was pricing yield. I was pricing the source of the yield.
We didn't miss the crash; we shorted the narrative. Same logic applies here.
Core
Let's break down the actual data points Goldman cites.
First, the AI hedge basket fell 5% in five days. The high-beta momentum basket fell 12%. That's a severe deleveraging event. But it's not a capitulation. Leverage in the AI trade had reached extreme highs — this was the unwind of crowded positioning, not a fundamental shift in AI adoption.
Second, the momentum rotation. The three-month momentum long book now holds software over semiconductors. That's a critical shift in market perception. It means the market believes software has better near-term monetization certainty than chip vendors. That's not a technology bet — that's a deployment bet. The market believes AI infrastructure is built enough, and now it's time for the applications layer to prove it can generate revenue.
Third, the recommendation of storage and data centers. This is the most intellectually honest part of the report. Goldman is saying: the compute cycle has been built, but the memory and infrastructure layer hasn't been fully repriced. The profit recovery in these names hasn't been discounted.
I'd push further. Based on my experience auditing 0x Protocol v1 contracts in 2017, I learned that the biggest opportunities often sit in the boring infrastructure layer — the orders, the matching, the settlement. The same principle applies here. Storage isn't glamorous. Data centers aren't the AI story. But the unit economics are transparent. The demand is growing. And the market hasn't fully priced the 2025-2026 earnings recovery.
Fourth, the money flows. The report notes capital rotating into European and Japanese banks, gold miners, and copper producers. This isn't AI capitulation. This is a rotation in capital allocation. Copper is especially telling — AI data centers are energy and copper intensive. The market is trading the inputs of AI, not just the outputs. That's a mature market signal.
Contrarian
Now the counterintuitive angle. The momentum rotation is a lagging indicator. It reflects the past three months of capital flows. It does not predict the next three months. The software names that now lead the momentum book are the same names that got hammered in 2023 when AI hype peaked. The storage and data center names that Goldman recommends now — they're the same names that underperformed in the first half of 2024.
So the contrarian question: is Goldman's recommendation already a crowded trade? The bank is a sell-side institution. Its clients include companies in the storage and data center sectors. When a bank tells you a trade, ask what the bank's clients have already bought.
The ledger is the only court of final appeal. The momentum factor is a lagging indicator. The EPS revisions are the leading indicator. Goldman says storage and data centers are positioned for profit recovery — but I'd ask: what's the actual quarterly earnings revision trend for these names? If the market's EPS forecasts are already rising, the "valuation gap" may be narrower than the report suggests.
Takeaway
The real signal from this report isn't the short-term trade. It's the structural shift. The AI trade is moving from "concept" to "fundamental." That's a healthy maturation, not a death.
The key catalyst is the NVIDIA Q2 earnings call and the September industry conference. If NVIDIA delivers strong numbers and raises guidance, the momentum rotation will be tested. If they miss, the deleveraging accelerates.
Skepticism is the shield; data is the sword. The AI trade isn't over — but the free lunch is. The next phase belongs to analysts who can read the profit chain, not just the narrative.
I'm watching the storage and data center EPS revision charts. That's the signal. The short-term rotation doesn't matter. The long-term ledger does.
Charts lie. But the ledger — the actual financial statements — never sleeps.