Hedge Funds Slash Mag-7 Positions Before Rebound Hits as Valuations Compress to Decade Lows
Alina Collins
Hedge funds slashed Mag-7 long-short ratios to the 25th percentile since 2018 just before the MAGS index outpaced the Nasdaq 100 and SOX over the past month; JPMorgan notes hyperscaler valuation premiums have compressed to a near-decade low, with room for incremental capital to re-enter.
Hedge funds sold out — and then the rally hit?
JPMorgan data shows Mag-7 long-short ratios fell to the 25th percentile since 2018. Net and gross exposures both reset to clearly underweight levels.
After the position flush, the MAGS index outperformed both the Nasdaq 100 and the Philadelphia Semiconductor Index over the past month. The timing of the unwind and the rally lined up almost exactly.
This means → the rally was not driven by a sudden improvement in fundamentals — it was fueled by short covering after extreme position liquidation.
Valuations at a decade low — how much room is left?
Hyperscaler valuation premiums — the extra multiple investors pay for mega-cap cloud stocks over the S&P 500 — have compressed to a near-decade low. Forward P/E ratios narrowed sharply.
Positioning has partially rebuilt with the rally but remains well below prior highs.
In plain terms = cheaper valuations plus still-light positioning — both conditions stacking up means the pool of potential buyers has not been exhausted.
Is AI actually making money yet?
Cloud revenue growth: AWS up 37% YoY, Azure up 43%, Google Cloud up 82% — demand continues to outstrip supply.
Backlog — signed contracts not yet delivered — keeps swelling: Google Cloud hit $514 billion, adding $52 billion in a single quarter; AWS reached $496 billion, up 36% QoQ and nearly 2.5× YoY.
Amazon also disclosed that its server and networking investments pay back in under three years. This means → capital spent today is recouped within three years, easing the market's fear that AI spending is a black hole with no return.
A $900 billion AI arms race — who is footing the bill?
JPMorgan estimates the market expects total AI capex of roughly $900 billion this year, up about 85% YoY, rising past $1.2 trillion next year.
Hyperscalers are projected to account for roughly 87% of that total. AI spending now makes up nearly 60% of total S&P 500 capital expenditure.
This reflects a structural shift: AI investment is no longer "a tech-sector story" — it is reshaping capital allocation across the entire US equity market.
Where is the biggest risk hiding?
Over the trailing twelve months, hyperscalers earned a combined ~$599 billion in net income — but free cash flow (the actual cash left after capex) came in at only ~$169 billion, a gap of $430 billion.
Except for Microsoft, most hyperscalers are expected to run negative free cash flow through 2026–2027.
Put simply = the profits look strong on paper, but every dollar is being plowed back into data centers, leaving pockets empty. The market is betting that backlog converts on schedule and capex pays back within three years — whether that bet holds is the single largest variable ahead.
Content is for reference only, not financial advice.