Sustained high memory demand and escalating prices continue to pressure the margins of smartphone manufacturers, especially during a period of mild market demand, according to GF Securities.
In particular, higher prices for NAND flash and LPDDR (Low-Power Double Data Rate) have driven up memory to range from 20% to 30% of the bill of materials for smartphones. It was previously 10% to 20%. GF Securities expects global smartphone shipments to decline by 6% year over year in 2026.
“On the other hand, we expect Apple to continue outperforming, thanks to its strong procurement capabilities, and we believe that Apple is in negotiation with Korean suppliers aiming to reach a favorable deal,” said GF Securities analysts Evan Lee and Jeff Pu in an investor note.
In the AI data center market, the financial firm finds that higher memory costs could cause capital expenditures among the hyperscalers to increase by 7% to 8%. During Microsoft’s (MSFT) most recent earnings call, executives noted that rising memory prices would affect capex.
“However, our proprietary analysis suggests the recent memory price hike (HBM3e/LPDDR5/NAND) translates to an incremental cost of $0.3m per rack or $1.8bn per 1GW data center deployment,” Lee said. “This equals to +7% BOM for racks and +5% in total investment of $35bn (22.5bn for GB300 Racks and 12.5bn for infrastructure) for 1GW capacity.”
This is likely behind Meta Platforms’ (META) recent upward revision to capex.
“As such, we think this suggests the recent upward CAPEX revisions from CSPs, like Meta, are primarily to reflect the higher memory costs, alongside the hike in general servers,” Lee noted. “As general servers usually account for 20%-30% of CSPs’ CAPEX, a 30% increase in BOM would also contribute to a 7-8% rise for CAPEX. Of note, we believe the AI supply chain is increasingly focused on this issue, particularly neo-clouds that impact AI startups.”
The ongoing memory issue will also apply pricing pressure to HGX 8-GPU servers and general servers.
“According to our analysis, memory cost hikes will lead to an 11% BOM increase for the HGX 8-GPU B300 (NVDA) AI server, where we agree with the recent news of the price upward adjustment of 15-20%,” Lee said. “For general servers, memory and storage costs are expected to reach 72% of the total BOM for now (up from 42%), which creates a pricing pressure for server brands typically with a 20% gross margin.”