Projections now put the average selling price of high bandwidth memory (HBM) at 121% higher next year, a figure that stands out even by the standards of a memory market that has spent the past two years repricing itself around AI accelerator demand.

The forecast, reported by finance.biggo.com, links the price surge directly to a structural change in the product mix: suppliers are moving their output toward 8-high stacks, and that transition is redefining what an "average" HBM unit costs.

Why the mix shift drives the number

HBM pricing is not a single-line item. Manufacturers sell the memory in stacked configurations, and the count of DRAM dies in the stack is the primary differentiator between product tiers. An 8-high module carries roughly twice the die count of a 4-high module, along with a more demanding assembly process, higher yield risk, and a larger package footprint.

When suppliers shift capacity from lower stacks to 8-high production, the weighted average selling price across their HBM shipments rises mechanically. The projected 121% increase reflects this arithmetic as much as it reflects any tightening of supply or strengthening of demand: the average unit being sold next year is simply a bigger, more complex unit than the one sold this year.

That distinction matters for buyers. A headline ASP jump of this size does not translate one-to-one into a per-bit cost increase. Customers purchasing 8-high products receive correspondingly more capacity and bandwidth per module. The effective price movement on a per-gigabyte or per-terabyte-per-second basis will be flatter than the headline figure suggests — though the projection still indicates firm pricing power on the supplier side.

What it means for the supply chain

For memory manufacturers, the shift to 8-high output is a capacity decision with consequences across the fab. Every 8-high stack consumes more wafer starts than a 4-high equivalent, and advanced stacking steps — through-silicon via formation, microbump bonding, testing of taller packages — take more time and carry lower cumulative yields. Allocating wafer capacity to HBM in general, and to 8-high tiers specifically, means less capacity available for conventional DRAM and other product lines, a trade-off suppliers have been making in growing increments as AI accelerator roadmaps call for denser memory configurations.

For system integrators and accelerator vendors, the projection signals continued cost pressure at the memory interface. HBM already represents one of the largest line items in the bill of materials for high-end AI hardware, and a doubling of average selling prices will flow into module, board, and system-level pricing wherever 8-high configurations become the default specification.

The context behind the number

A 121% projected ASP increase is large even for a component category that has seen repeated double-digit price moves since AI demand began absorbing HBM supply. Memory markets move cyclically, and analysts' forward price estimates have historically tracked the mix of products shipped as closely as they track supply-demand balance. The 8-high transition gives the projection a concrete mechanism: suppliers are not merely charging more for the same product, they are changing what the product is.

Whether the full 121% materializes will depend on how quickly 8-high output ramps, how yields on taller stacks evolve, and how negotiation leverage distributes across the concentrated set of HBM suppliers and their large accelerator customers. The direction, however, is unambiguous in the projection: next year's HBM market will be built around 8-high stacks, and the average selling price will reflect it.