AI memory demand pushes notebook BOM share to 68%; consumer costs now carry the trade
The memory shortage has reached the bill of materials. TrendForce’s latest benchmark shows CPU, DRAM and SSD accounting for 68% of a $900 notebook’s component cost in the third quarter of 2026, up from 45% in the first quarter of 2025. The 23-point increase gives memory suppliers more pricing power and leaves device makers with a narrower margin buffer. TrendForce expects 2026 notebook shipments to fall 9.4%, with only 47% of the year’s shipments left for the second half after buyers pulled demand forward, adding volume risk to OEM gross margins. (TrendForce)
The investment signal is a two-sided allocation trade. AI infrastructure is paying more for scarce memory, while consumer electronics must absorb the cost, pass it to customers or sell fewer units. The new 68% measure shows the pressure reaching system economics; it does not make every device maker a direct beneficiary or victim by the same amount.
Server memory is winning the allocation fight
TrendForce estimates HBM and RDIMM together will consume 51% of 2026 DRAM bit supply. HBM is especially demanding on wafer capacity: by year-end it is projected to use 22% of the top three suppliers’ DRAM wafer input while producing 9% of DRAM bits, raising the capacity cost of each additional bit. That mix gives suppliers a strong incentive to prioritize AI and conventional server workloads over lower-margin consumer demand.
The competition is not simply HBM versus PCs. TrendForce says 64GB DDR5 RDIMM overtook HBM in per-wafer revenue and profitability in the first quarter. It expects server-DRAM contract prices to rise about 270% during 2026, including another 13–18% sequential increase in its third-quarter forecast. Server DRAM and HBM therefore form a broader pricing complex, with hyperscalers and enterprise buyers competing for the same manufacturing capacity.
Long-term contracts are shifting who receives supply visibility. SK hynix said in July that it had agreements with about 10 customers. Micron disclosed 16 strategic customer agreements; 14 carry about $100 billion of cumulative revenue at contractual minimum prices, and signed agreements are expected to generate about $22 billion of customer deposits and related financial commitments. Those contracts protect selected buyers’ access while leaving shorter-term customers more exposed to price and allocation changes. (Micron filing)
Consumer devices face the pass-through test
The mobile market is already showing the cost pressure. TrendForce estimates that combined memory costs in a 256GB iPhone 18 Pro could be nearly 400% higher year on year and forecasts retail prices for the new range 10–20% higher, squeezing device gross margins. Its third-quarter mobile-DRAM work projects another 8–13% sequential contract-price increase even as smartphone inventories improve and end demand softens. The 400% figure combines LPDDR and NAND and is a research estimate, not an Apple pricing commitment. (TrendForce mobile analysis)
Notebook makers have already seen a similar channel response. Microsoft said PC inventories were elevated partly because buyers were purchasing ahead of memory-price increases. Microsoft’s Xbox pricing update then raised US console prices by $100 for 512GB models and $150 for 1TB models, citing memory and storage costs more than 2.5 times higher. The pass-through can protect gross margin, but it also raises the risk that price-sensitive demand breaks.
NAND is splitting by application. Enterprise SSD revenue for the top five suppliers reached $37.59 billion in the second quarter, up 103.6% sequentially, and TrendForce projects enterprise-SSD prices to rise about 235% during 2026. Client SSD inventories are higher, eMMC and UFS supply is more abundant, and its August spot table showed 512Gb TLC NAND down 1.47%. The AI pull is therefore strongest in enterprise storage, not retail NAND as a whole. (TrendForce NAND outlook)
Relief is a late-cycle event
TrendForce expects meaningful new DRAM production in the second half of 2027, with larger contributions deferred until 2028. Micron targets first wafer output from its Idaho fab in mid-2027; its Singapore NAND cleanroom is scheduled for the second half of 2028. Until that supply arrives, the cleanest bullish exposure remains server DRAM, HBM and enterprise SSD pricing power. The bear case is a faster demand break: if notebook and smartphone prices destroy enough volume, or hyperscaler spending cuts release capacity toward consumer products, memory prices can roll over before new fabs matter.
The settling data are quarterly contract prices and device economics. Two consecutive quarters of falling server-DRAM prices alongside normalized allocation would weaken the shortage thesis; firm server pricing with client-SSD easing would confirm that AI is tightening the memory complex unevenly.