High-bandwidth memory will cost buyers more than twice as much in 2027 as it does today, according to a new forecast from market research firm TrendForce. The projection, reported by the Seoul Economic Daily, attributes the sharp escalation to a supply crunch across the HBM segment rather than to demand growth alone.

The forecast lands at a moment when HBM has become the most contested product category in the memory industry. The stacked-die DRAM technology, which packages memory directly alongside logic compute, now anchors the accelerator roadmaps of the major AI hardware vendors. That structural dependence has converted HBM from a niche component into a capacity-allocation battleground, and TrendForce's numbers indicate the imbalance will intensify rather than ease over the next several years.

A price increase of more than 100 percent in a single product cycle is unusual even by the standards of the memory industry, which has always moved in sharp boom-and-bust swings. What distinguishes the current cycle, according to the forecast, is the duration of the squeeze: the crunch is expected to persist through 2027, giving suppliers sustained pricing power across multiple procurement cycles rather than a brief spike.

For buyers, the practical consequence is straightforward. System manufacturers and hyperscale data center operators that have built multi-year accelerator roadmaps around HBM-equipped hardware will face materially higher memory costs at the point when 2027 supply agreements are negotiated. Procurement teams that locked in longer-term contracts will be partly insulated; those buying on spot or short-cycle terms will absorb the full increase.

The supply side explains the arithmetic. HBM carries a substantial manufacturing penalty relative to conventional DRAM. Building the stacked packages requires advanced die fabrication, through-silicon via interconnects, and co-packaged qualification with customer compute silicon — steps that consume wafer capacity, yield, and engineering time. Every HBM wafer therefore displaces conventional DRAM output that a supplier could otherwise ship. When suppliers shift their mix further toward HBM to serve the accelerator market, conventional DRAM supply tightens as well, compounding the crunch across the broader memory market.

TrendForce's projection of a more-than-doubling price level for 2027 also signals where the industry believes the bottleneck sits. Prices at that scale do not emerge from incremental demand growth. They emerge when buyers compete for constrained qualified supply, and when suppliers — even running fabs at full utilization — cannot expand output fast enough to close the gap.

The timing matters for product planning. Accelerator vendors typically qualify memory suppliers years ahead of volume shipments, which means the cost basis for systems shipping in 2027 is being negotiated now. A forecast of doubled HBM pricing implies that the bill-of-materials pressure on next-generation AI hardware will fall disproportionately on the memory subsystem, and that margin allocation between memory suppliers and systems vendors will shift accordingly.

For the memory suppliers themselves, the forecast points to an extended period of pricing power in their highest-value product line. HBM already commands a significant price premium per bit over standard DRAM. A further doubling extends that premium, and it does so in a segment where the customer base is concentrated among a small number of large, well-capitalized buyers.

TrendForce is a Taiwan-based research firm whose memory pricing forecasts are widely followed by procurement and investment professionals across the semiconductor supply chain. The report was relayed by the Seoul Economic Daily, a publication with close coverage of the South Korean memory industry, where two of the world's three major DRAM suppliers are headquartered.

The core takeaway from the forecast is unambiguous: through 2027, HBM will remain a seller's market, and buyers should plan their memory cost curves on the assumption that prices will more than double before the imbalance corrects.