Amazon puts the memory crunch on the shelf: Echo Dot up 60%, Kindle up $40

Work the Echo Dot backwards. If J.P. Morgan's projection holds and DRAM runs up 400% or more between the start of 2024 and the end of 2026 ^[5], the memory inside the old unit would have to have been worth roughly $7.50 to $10 at 2024 prices for $79.99 to be plain cost recovery ^[3]. That is 15% to 20% of what the entire device used to retail for. Plausible on a memory-heavy product sold at thin margin, and equally consistent with a hardware line whose owner has decided it will not be sold at $50 any more. The Kindle shows what restraint looks like next to it: $40 onto a $109.99 base is 36% ^[1].

Amazon sits on both sides of the same chip. TrendForce has the nine largest clouds spending more than $886.7 billion in 2026 capex, up as much as 90% year over year ^[7], which puts the prior-year base somewhere near $467 billion ^[2]. J.P. Morgan attributes part of the shortage to hyperscalers signing long-term supply agreements ^[5]. Amazon has already repriced the compute those chips end up in, lifting EC2 Capacity Blocks for Machine Learning about 20% in July after 15% in January ^[14], close to 38% compounded inside a year ^[4]. The devices group is now buying into a market the cloud group helped clear.

The pass-through so far is partial, which is the part worth modeling. J.P. Morgan's composite index for software, hardware and storage prices is up 23% since the end of 2024 ^[6], against a DRAM curve several times steeper. IDC expects PC shipments to fall 11.3% in 2026 and smartphones 12.9%, with revenue roughly flat because average selling prices are rising ^[9]. That is an industry shipping meaningfully fewer units for the same money, and it happens before most of the DRAM curve has reached a shelf price.

Duration is the other half. Micron says 16 strategic customers have committed $22 billion to secure memory chips, and chief executive Sanjay Mehrotra told Reuters he expects constrained conditions to persist beyond 2027 ^[8]. IDC's own outlook does not show pricing returning to 2025 levels inside its forecast period ^[12]. So the input will not normalise within a product cycle, and Amazon has volunteered to find out whether a 60% consumer increase survives contact with a buyer ^[2]. Anyone costing hardware for next year should treat memory as a supplier with pricing power rather than a commodity input, and should read the announced increases as the opening instalment rather than the settlement.