Goldman Sachs Partner: Market Has Razor-Thin Margin for Error as Crowded Positioning and Leverage Risks Compound
N.R. Finch
Goldman Sachs partner Rich Privorotsky warns that AI spending now underpins hardware valuations, GDP growth, and broad-market performance simultaneously — forming a self-reinforcing loop whose fragility will be exposed the moment the spending cycle stalls.
A Wyoming data center paused — why does one project matter?
Crusoe suspended a 1.8-gigawatt data center in Wyoming at the client's request. The facility was designed to bypass grid bottlenecks with 900 MW of behind-the-meter power — generation built on-site, off the public grid.
The project was still early-stage; the client was unnamed. Privorotsky himself cautioned against drawing macro conclusions from it.
This means → The alarm is not about this one project. It is about a market so dependent on the "AI capex only goes up" narrative that even an isolated deferral forces investors to re-examine their demand assumptions.
AI is splitting into two tiers — what does that mean for the spending cycle?
Privorotsky describes a structural divergence: on one end, centralized cloud-hosted frontier models — expensive and compute-intensive. On the other, a near-free, open-source local AI layer growing powerful enough for most everyday tasks.
In plain terms = routine work runs on a free model on your phone or laptop; only hard problems get sent to cloud-scale compute. Two tiers, each handling its own slice.
Bull case: the total pie expands — edge compute, data centers, memory, power, and networking all grow. Bear case: most economically valuable tasks can soon run on existing hardware, and the demand inflection arrives far sooner than current valuations imply.
This reflects a shifting debate. Privorotsky notes it is "increasingly not about model quality — it is about where inference ultimately runs."
Where exactly is the "circularity"?
AI spending props up the market on three levels at once: ① the core bull thesis for hardware stocks, ② a contributor to GDP growth, ③ a weight driver of broad-index performance.
This means → The three reinforce each other: AI spending lifts hardware stocks → hardware stocks pull up the index → index performance supports economic confidence → confidence justifies more AI spending. Break any link and all three layers come under pressure.
In plain terms = AI spending is both cause and effect. The market uses "AI will keep spending" to prove "AI is worth spending on" — that is the circularity Privorotsky is flagging.
Positioning and leverage — how high is the risk stacked?
Current positioning data: momentum returns sit at the 90th percentile on a five-year lookback (momentum returns — the payoff from buying the strongest stocks and selling the weakest). Gross exposure is at the 99th percentile.
Financing spreads — the cost of borrowing to lever up — have begun to widen. Retail participation via leveraged ETFs (funds that use borrowed money to amplify gains and losses) remains sizable.
This means → Privorotsky argues that the embedded and synthetic exposure created by leveraged products makes owning gamma — options convexity that pays off asymmetrically in sharp moves — far more valuable as portfolio protection than the market broadly appreciates.
When does the test arrive?
This risk is not new, but Privorotsky's read is clear: the market is accelerating toward it.
Whether the AI spending cycle continues to deliver is the core variable that determines if the current valuation framework holds.
In plain terms = the entire market's valuation logic is a bet that AI spending will not slow down. If it does, today's prices are missing their foundation.
Content is for reference only, not financial advice.