AI memory crowd-out → non-LTA buyers are the residual claimants → device units fall but revenue holds → the consumer dollar transfers from OEM to memory supplier

The funding source for the memory supercycle is the consumer-device OEM. AI-server allocation and multi-year LTAs insulate the hyperscalers, which routes the entire price increase onto non-LTA buyers — PC and smartphone OEMs. IDC: PC units −11.3% but PC revenue +1.6%; smartphone units −12.9% but revenue −0.5%. Units fall double-digits while revenue is flat-to-up: the value is not destroyed, it migrates to the memory supplier. Thin-margin Android/PC OEMs (HPQ, DELL, Acer, ASUS, Lenovo, Xiaomi, Transsion) eat it; AAPL and Samsung are structurally hedged.

• Step 1: Memory suppliers publicly restore consumer allocation priority, or consumer DRAM/NAND availability normalises.
• Step 2: SK Hynix / Samsung confirm LTA price caps are scrapped — LTA-covered CSPs would rejoin the price-taking pool and the seam this step rests on collapses.
• Step 3: A named OEM rescinds a memory-driven price increase, or DDR5 module prices retrace toward the $40 pre-shock level.
• Step 4: Realized 2026 device revenue falls in line with units (i.e. revenue does NOT hold up), which would mean demand destruction, not value transfer — a different chain with different victims.
• Step 5: AAPL or Samsung guide down on memory cost, showing the 12–24 month hedge did not insulate them — which would collapse the up-market share-shift leg.

• **The unit forecasts are analyst projections, not realized data.** Omdia reported the smartphone market down only ~4% YoY in Q2 2026 against IDC's −12.9% and CCS Insight's −14.8% full-year forecasts. The back half must be much worse for those forecasts to land. Step 4's *direction* is corroborated independently (IDC + CCS Insight); its *magnitude* is a forecast.
• **The shock may be peaking, which caps the chain.** Q3 2026 DRAM contract increases decelerate to 13–18% QoQ from ~58–63% in Q2 and ~93–98% in Q1 — explicitly because consumer makers are hitting their affordability limit. Demand destruction capping the price move is *the mechanism working*, but it also means the OEM margin hit may be nearer its end than its beginning.
• **Step 2's LTA protection may already be dissolving.** One unverified secondary report claims SK Hynix has scrapped long-term contract price caps while Samsung targets another 20% Q3 hike. If true, LTA-covered CSPs rejoin the price-taking pool and the clean non-LTA/LTA seam blurs. Headline-only, unfetched — a live risk to step 2.
• **OEM pull-forward partially defers the hit.** Lenovo reportedly stockpiled RAM to 50% above usual, enough to last through 2026; PC OEMs "built inventories aggressively during the first half of the year." A hoarded OEM eats the shock later than an unhedged one — which blunts the near-term earnings expression even if the chain is right.
• **Pass-through may protect OEM revenue better than margin, or vice versa, and nothing here sizes it.** IDC gives units and industry revenue; no source in this pass gives OEM *gross margin* impact per name. The chain establishes the transfer's direction and the victims' identity — not the EPS delta.

• **The un-owned leg is the HURT side.** The benefit side (MU, Samsung, SK Hynix) is already owned via hbm-cowos-as-binding-bottleneck. What is new here is the identity of the funding source: **HPQ, DELL** (10–30% hikes on commercial PCs), Acer (2353.TW), ASUS (2357.TW), Lenovo (0992.HK), Xiaomi (1810.HK), Transsion (688036.SS).
• **Sort by hedge, not by brand.** AAPL and Samsung are relative beneficiaries *within* the OEM set — 12–24 month advance procurement means they absorb the shock later and shallower, and share shifts up-market during a cost shock. The trade is the spread, not the sector.
• **Contract structure is the screen.** Step 2 generalises: in a supply shock, ask who holds an LTA. LTA holders are insulated; non-LTA buyers are the residual claimants and therefore the P&L victims. See lta-contract-structure-as-price-insulation.
• **This rhymes with iran-fuel-shock-consumer-bifurcation** — same K-shaped incidence, different forcing function. Convergent evidence for the bifurcation pattern, not an independent chain.