Memory's Fastest Unwind of 2026: 27% to 46% in 10 Days, Then an 18% Snapback

Podcasts called it a leverage event, not a demand break. The data agrees.

Memory just had its fastest unwind of 2026. It was not about demand.

In the last 10 trading days of July, Micron fell 27%, Western Digital 34%, Seagate 26%, and Sandisk 46% from their highs. On July 28, semis were down 4% while software was up 3.3%, an outlier rotation day going back to 1989. Then on July 30, Micron bounced 18.4% in one session. The physical market did not change that fast. Positioning did.

What happened

The numbers from the tape:

• Micron (MU): High $991.64 on Jul 9 to low $739.00 on Jul 29 = -25.5% in 14 sessions, then +18.36% on Jul 30 to $874.66
• Podcast-reported drawdowns from highs (Animal Spirits, Jul 29): MU -27%, WDC -34%, SNDK -46%, STX -26%
• Rebound (Moving Markets, Jul 31): SOX +8%, MU +18%, AMD +13%, SK Hynix +17% in one day
• Context: S&P 500 was only ~2% off all-time highs during the unwind. This was a memory-specific leverage flush, not broad market risk-off.

Why podcasts say leverage, not demand

Animal Spirits Podcast — Jul 29: "Semiconductors were down 4% on the day, and software, the anti-AI semi trade, was up 3.3%... absolute outlier... Micron is down 27%, Western Digital down 34%, Sandisk is down 46%, Seagate Technology is down 26%."

Hedge Fund Tips — Jul 29: "average semiconductor stock is down 25 to 45 percent... gamma squeeze, retail is buying... going lower... sell any rips in semiconductors in memory."

Time in the Market — Jul 28: Flagged China risk as trigger: "China steps in and suddenly increases production of memory... puts downward pressure on margins... memory prices have started to roll over a little bit... from exceptionally high levels."

The Data Center Daily — Jul 31: Reframed the bigger picture: Meta's $31.08B quarterly capex including leases plus $25B in new notes shows the AI build is now debt-funded, which makes high-beta memory the first to de-risk when rates or risk appetite wobbles.

Takeaway: No one cited a canceled HBM order or a fab coming online early. The debate was gamma, retail leverage, and China headline risk.

Fundamentals did not break

This is why the unwind looks like positioning:

• Micron FQ3 (ended May 31): Revenue $41.56B, +345.7% YoY, +73.7% QoQ. Gross margin 84.56%, operating margin 80.37%, net margin 68.13%. FCF $17.56B, +951% YoY. Net cash $19.6B.
• Valuation: MU trades at 5.8x forward earnings, 20.1x trailing, PEG 0.12. Market cap ~$1.01T. 21-day performance -9.1% but 63-day +55.3%. RSI 49, neutral.
• Street: 43 analysts, consensus Strong Buy, mean price target $1,508, +68.9% upside to mean. Low $361, high $2,200.
• Sell-side read: BofA reiterated Buy $1,550 on Aug 4, calling pullback an "enhanced buying opportunity," forecasting FY28 EPS ~$150 vs prior cycle peak ~$12. UBS: "Memory undersupply is expected to continue into 2028, fueled by Agentic AI driving growth beyond HBM... almost all incremental DRAM wafer capacity is going toward HBM."

The physical bottleneck is still there. HBM sold out through 2027 and conventional DRAM in outright shortage. SK Hynix printed 76% margins — the pass-through gap is accounting, not demand.

Flow flipped back to accumulation

Algo heatmap for memory basket (21-day window, last 5 sessions Jul 29 - Aug 4):

• MU: avg -4.3 (distribution bias), but last 3 sessions +54.2, -54.6, +55.4, +40.3 = 5-day avg +6.0, flipping to accumulation
• SNDK: avg -3.0, 5-day avg +13.1, with +58.8, +57.9, +51.8 accumulation prints
• WDC: avg -0.4, 5-day avg +1.5
• STX: avg -7.9, 5-day avg -16.1, still lagging

Interpretation: The unwind was real distribution Jul 28-29 (scores -65 for MU, -47 for SNDK), but Aug 3-4 saw aggressive re-accumulation. That matches the 18% bounce.

What to watch

• China supply: CXMT IPO and consumer DRAM flood risk is real for low-end, but 3-player oligopoly for HBM/AI DRAM remains intact. Watch for yield headlines.
• Hyperscaler capex language: Amazon raised 2026 capex to $220B citing higher memory costs as bullish for MU. If capex is memory-price driven, not compute-volume driven, ROI math gets harder.
• LTA structure: Long-term agreements now cover >50% of server DRAM with take-or-pay and floor pricing above prior cycle peaks. That is what turns memory from 1-3x book commodity to secular.
• Gamma: Retail call buying drove the run-up. Same gamma unwind drove the flush. Options positioning still elevated.

The fastest unwind of 2026 looks like a leverage event inside a supercycle, not the end of the supercycle. The next test is whether the re-accumulation holds into earnings.