For most of the past year, high-bandwidth memory has been the one corner of the memory market where contract prices stayed relatively contained. That's about to change. In a June 2026 analysis, TrendForce projects that HBM contract prices will rise by multiples in 2027 - and the reason isn't HBM demand alone. It's what's happening in conventional DRAM.

Tight DRAM is the lever

DRAM supply has tightened dramatically. Industry DRAM revenue jumped 81% quarter-over-quarter in Q1 2026, and the price surge has been steep enough to reorder the economics of the whole memory portfolio. By Q1 2026, HBM's per-wafer profitability had actually slipped below that of a high-capacity DDR5 64GB RDIMM - a striking inversion, given HBM is the premium, AI-grade product.

That inversion is exactly what hands suppliers their leverage. With conventional DRAM this profitable, the three major memory makers have every incentive to reallocate wafer capacity toward it - which means HBM only stays strategically prioritized if its pricing catches up. TrendForce's read: suppliers will use the tight-supply backdrop to push HBM contract prices sharply higher in 2027, restoring HBM's premium over commodity DRAM.

Demand is climbing into that tight supply

The pricing pressure lands on top of a steep demand ramp. TrendForce projects HBM's share of total DRAM wafer input rising from roughly 18% at the end of 2025 to 22% in 2026 and 30% in 2027. On a bit basis, HBM's share of supply grows from about 8% in 2025 to 9% in 2026 and 13% in 2027 - a reminder that HBM consumes far more wafer area per bit than conventional DRAM.

Driving it is the next wave of AI accelerators. HBM content per AI chip is climbing from today's 96GB/192GB toward 216GB/288GB in 2026, and NVIDIA's Rubin Ultra platform - expected in 2027 - is slated to carry up to 384GB per GPU. Each generation pulls more HBM into each system, and the supply base is racing to keep up.

What to do in 2026

  • Map your HBM exposure now. Which products use HBM, in what generation and capacity, and against which 2027 build volumes? You can't budget for a multiples price increase you haven't quantified.
  • Lock 2027 contract terms early. If suppliers raise contract prices by multiples next year, the cost of waiting is direct. Booking 2027 demand while this year's terms still apply is the clearest hedge.
  • Don't assume conventional DRAM is the safe harbor. The same tight supply driving HBM pricing is driving DDR5 - budget both, not just the AI line items.
  • Re-baseline product cost models. If a design's bill of materials assumes last year's memory pricing, the standard cost is already stale. Update it before it shows up as a variance.
  • Talk to us about forward coverage. For memory-exposed programs, locking supply and pricing ahead of the 2027 step-up is the lever you still control.

The bottom line for buyers

This isn't the usual supercycle story where new capacity eventually floods in and prices collapse. The constraint is structural: HBM consumes disproportionate wafer area, conventional DRAM is now profitable enough to compete for that same capacity, and AI demand keeps raising the HBM content of every new accelerator. TrendForce's forecast of multiples-higher HBM pricing in 2027 is the market pricing that squeeze in. For anyone with HBM - or high-capacity DDR5 - on a 2027 roadmap, the planning window is now, while this year's contract terms still hold.