Consumer Data Cracked, Broadcom Gapped, and SOXL Lost 8%

A Consumer Print, Not a Chip Print

Two macro releases landed before the August 14 open, and both were bad. July retail sales fell 0.6% — the biggest monthly drop since May 2025 and the first decline in nine months — with nonstore retailers off 2.2% and motor vehicle and parts dealers down 1.8%, per Census Bureau data. An hour later the University of Michigan preliminary sentiment index printed 51.0 against a 54.5 consensus, down from 55.2 in July. Survey director Joanne Hsu flagged the internals as the real damage: short-run expectations down 11%, long-run down 17%.

The broad tape shrugged. The S&P 500 slipped 0.21% to 7,782.65 and the Nasdaq lost 0.50% to 26,664.30, one session after the index had set a record close at 7,798.99, clearing 7,800 for the first time. A quarter-percent give-back after a record is noise.

Semis did not get the memo. Broadcom dropped more than 6% intraday to roughly $391.50 before closing near $395, and there was no fresh filing or press release behind it — it reads as profit-taking into the company's fiscal Q3 report in early September plus lingering questions about gross-margin mix as revenue shifts toward custom AI silicon. Broadcom is about 5.46% of the index SOXL references, so that alone is a meaningful drag on a day when the rest of the sector was already soft. Micron, notably, went the other way.

The Daily Reset Is Doing Most of the Work

None of the above is an 8% story on its own. The leverage is.

SOXL targets 300% of the *one-day* return of the ICE Semiconductor Index — the 30 largest U.S.-listed chip names — and the multiplier resets at every U.S. close. Back out the math and an 8% move in the perp implies something on the order of a 2.7% decline in the underlying basket. That is a bad-but-ordinary day for semiconductors. The product converts it into a number that looks like a crisis.

Over longer horizons the divergence gets ugly rather than merely amplified. From the June 22 high through late July, SOXL fell 63% while the unleveraged iShares Semiconductor ETF lost 25% — nowhere near 3x. Volatility decay scales with the square of leverage, so a 3x fund bleeds at roughly six times the rate of the underlying's variance where a 2x fund bleeds at two times. In a sector that has been chopping violently since June, that drag compounds every single session.

At $140.20 the perp sits roughly 54% below the June 2026 high of $302.00. If the index round-trips back to its June level, SOXL does not. That is not a bug in the fund; it is the disclosed design.

Flows Say Nobody Believes the Decay

The positioning data is the part worth sitting with. SOXL took in $6.9 billion of net inflows during July 2026, its largest monthly intake since launch — while total assets under management *fell* by $906 million over the same period. Investors put in nearly $7 billion and the fund got smaller. That spread is the decay, paid in cash, in one month. Roughly $4.53 billion has come in over the trailing month, and leveraged chip ETFs collectively now hold about $21 billion. Fidelity's Jurrien Timmer has been warning the semiconductor cycle is near a peak; the flows have been completely indifferent to it.

The Hyperliquid perp stacks one more layer on top. Up to 10x on an instrument that is already 3x levered is roughly 30x effective exposure to a 30-stock index — a 1% move in the SOX basket is a 30% move in the account. With $11.46 million of 24h volume on this HIP-3 market, the book is thin relative to the underlying equity, which widens slippage in exactly the sessions where it matters most.

What to Watch

The next three prints define the tape: NVIDIA on August 26, Marvell on August 27, and Broadcom on September 2. Broadcom's is the most directly relevant to Friday's move, since that is what positioning was de-risking into. The PHLX Semiconductor Index entered bear-market territory in late July, more than 20% off its June record, so these reports land into an already-damaged sector rather than a confident one.

The consumer data deserves more weight than it is getting. The entire chip drawdown since June has been an argument about AI capex — whether hyperscalers are overspending. A retail sales miss of this size opens a second front: if demand is genuinely cooling, semis have to defend both the capex thesis and the cyclical one at the same time. One month is not a trend, but the August sentiment collapse says households are not treating it as noise.

One structural note specific to this market: the perp trades 24/7 while the underlying equity does not. Weekend headlines get repriced in the perp with no cash market to arbitrage against, and the reference price gaps at Monday's open. On a 3x product that is a real risk, not a footnote.