Goldman's Pasquariello: Surging Leveraged ETF Assets Are Amplifying Intraday Market Volatility
Claire Weston
U.S. 3× leveraged-ETF assets surged from $20 billion to $95 billion in under two months; Goldman warns the buildup is amplifying intraday volatility rather than reflecting fundamental change, while raising its AI capex forecast toward the $1 trillion mark.
Leveraged ETFs nearly quintupled — what is driving the swings?
U.S.-listed 3× leveraged-ETF assets jumped from roughly $20 billion in late April to $95 billion, nearly quintupling in under two months.
Goldman's hedge-fund-business head Tony Pasquariello cited the Philadelphia Semiconductor Index (SOX): last Tuesday SOX swung 11% peak-to-trough, and the 3× leveraged ETF tracking it traded nearly 1.4 billion shares.
This means → leveraged ETFs are not "predicting" direction — they are amplifying existing moves. Up days get sharper; down days get deeper.
In plain terms = leverage acts like options gamma — a measure of how sensitive an option's price is to moves in the underlying — adding fuel to intraday swings, but it does not start the fire.
AI capex — how much is the market underestimating?
Consensus puts 2027 hyperscaler — Amazon, Microsoft, Google and other operators of giant data centers — capex at $920 billion, implying growth slowing sharply from 84% to 22%.
Goldman's U.S. portfolio-strategy team says that estimate is too conservative again: over the past three years, analysts undershot actual capex by an average of 45 percentage points.
This means → if history repeats, 2027 capex could top $1 trillion — roughly $100 billion above the current consensus.
This reflects a market still "catching up to reality" on AI infrastructure demand, rather than running ahead of it.
Where does AI inference run — why is this the core debate?
Goldman senior trader Rich Privorotsky outlined two parallel scenarios: bull case — frontier large models and small efficient models advance together, lifting demand for edge computing, data centers, storage and networking.
Bear case — most economically valuable tasks can ultimately run on existing hardware; the real inflection in inference demand may arrive far later than current valuations imply.
This means → the debate has shifted from "are models good enough?" to "where does inference ultimately run?" — and the answer changes which hardware segments benefit.
AI inference deployment: demand explosion or hardware glut?
BULL
Dual-track expansion
Frontier and small models advance together, sustaining overall compute demand.
History of underestimation
Analysts undershot capex by an average of 45 ppts over three years — underestimation is the norm.
BEAR
Existing hardware may suffice
Most tasks may already run on current equipment, casting doubt on incremental returns.
Inflection could be distant
The real inference inflection may arrive far later than valuations imply.
In plain terms = both sides have a point — the key question is not whether models are good enough, but how much new hardware inference ultimately requires, and where it runs.
Record equity issuance — will it kill the bull market?
U.S. equity issuance hit a record in 2026, yet Goldman argues it will not end the bull market, for three reasons.
IPO count sits at roughly 100, near the 25-year average and well below the 250-plus of 2021 or the nearly 400 of 1999.
Corporate equity issuance of about $700 billion equals just 1% of Russell 3000 market cap, in line with the 2015–2019 average; meanwhile $1 trillion in buybacks will more than offset new supply.
This means → supply looks alarming on its own, but buybacks plus M&A plus overseas inflows provide greater "absorption" — the net effect is demand still exceeds supply.
Rates and Bitcoin — which panic narratives fall apart?
Pasquariello reiterated his rule of thumb: the U.S. 10-year Treasury yield needs a 2-standard-deviation rise within one month to materially weigh on equities — currently around 5%.
In plain terms = rates have to rise "unusually fast and unusually high" to truly hurt stocks; a slow grind higher does not qualify.
On the "Bitcoin leads risk assets" narrative, Pasquariello pushed back directly: since late 2024, Bitcoin has fallen 32% while the S&P 500 gained 29% and the Nasdaq 100 rose 42% — the two have clearly diverged.
What to watch next?
Whether leveraged-ETF assets shrink quickly in the next pullback — or pile higher and amplify the next swing — is the key signal for whether technical pressure can self-correct.
This means → if leveraged money exits fast on a dip, volatility self-heals; if it keeps building, the next bout of turbulence could swing even wider.
Content is for reference only, not financial advice.