My two Raspberry Pi boards cost as much as a laptop now - and AI is to blame

LPDDR4 DRAM prices have risen roughly 7x year-over-year, translating to about $25 extra per 4GB and pushing a 16GB Raspberry Pi 5 to ~$305 (vs $120 launch); specific Pi SKUs have seen $25–$150 price bumps. The surge is attributed to AI data-center demand (massive HBM3E/LPDDR5X consumption) and industry concentration—Samsung, SK Hynix and Micron control ~95% of DRAM production—and Micron has shifted away from consumer memory. Given fab build times/costs (~$10bn and 2–4 years) and prioritized supply to high-margin data-center customers, elevated RAM prices are likely to persist through at least 2028, posing a sector-level headwind for consumer electronics and single-board computer demand.

Analysis

Memory tightness driven by AI is not just a component-cost story — it shifts margin pools and distribution channels. Manufacturers and hyperscalers with direct supply contracts and ability to lock HBM/LPDDR capacity will see gross margin expansion, while consumer OEMs face both higher BOMs and longer qualification cycles, pressuring unit economics and incentivizing product consolidation over feature proliferation.

A salient second-order effect is the acceleration of the secondary/refurbishment market and marketplaces that intermediate used compute parts. When durable micro‑assets become uneconomic new, market liquidity and fee capture for platforms and specialty resellers grow; that arbitrage can be sizable and persistent because hobbyist demand is inelastic and substitutes are fragmented.

Timing matters: capacity response for DRAM/HBM is measured in years, not quarters. That makes the current cycle mean-reverting only if one of three low-probability events happens soon — a sudden collapse in AI capex, mass deployment of memory‑efficient model architectures, or a rapid multi‑manufacturer capex blitz. Any of those would compress prices sharply; absent them, memory scarcity is a structural multi-year tailwind for suppliers and aftermarket operators.

Key Decisions for Investors

• Long NVDA (6–12 months): buy NVDA shares or a 6–12 month call spread to express continued AI spend. Rationale: secular GPU/HPC demand and pricing power. Risk/reward: target +25–40% if AI capex stays elevated; downside -30% if model efficiency or regulation materially curbs GPU demand. Size as core conviction, hedge with 1–2% notional in long-dated puts.
• Long MU (Micron) (6–18 months): buy MU stock or LEAP calls to capture DRAM tightness and data-center allocation. Rationale: direct beneficiary of higher memory ASPs and share gains. Risk/reward: target +40%+, but downside -50% if large new capacity ramps; use a 12–18 month horizon and stagger entries on earnings down‑days.
• Long EBAY (3–6 months): buy EBAY shares or short-dated calls for upside from increased used-hardware flows and higher take-rates. Rationale: marketplaces win when new-unit affordability breaks; faster monetization of supply-side. Risk/reward: target +15–25%; downside -15% if retail liquidity normalizes or fee compression occurs. Position size tactical (tradeable alpha).
• Pair trade: Long MU / Short XRT (retail ETF) (3–9 months): equal notional to capture memory ASP tailwind vs stressed consumer retail margins. Rationale: buys a structurally tight input (MU) while hedging general consumer weakness. Risk management: stop pair if spread moves >20% adverse; expected outperformance 400–600bps under base case.