Memory chip prices surge heading into 2026 as AI server buildouts tighten supplies of DRAM—especially server DDR5—and high-bandwidth memory (HBM), giving suppliers more leverage in contract negotiations and forcing buyers to plan for higher component costs.
What’s driving the surge
The price pressure is being fueled by rapid AI infrastructure expansion, as cloud service providers upgrade data centers for larger models and higher-performance computing platforms that require more memory per server.
Market researchers tracking contract deals say server DRAM pricing strengthened in late 2025, and suppliers became more willing to raise quotes as large buyers increased order volumes.
One key factor is that HBM and conventional DRAM products compete for the same manufacturing capacity at leading suppliers, so booming HBM demand can effectively “crowd out” other DRAM output and keep overall supply tight.
TrendForce also expects global server shipments to rise about 4% in 2026, while AI-driven memory content per server continues to climb—two forces that push bit demand higher even if unit shipments grow modestly.
Where shortages are most acute
Two categories are central to today’s squeeze: high-end HBM used alongside AI GPUs, and server DDR5 used broadly across AI and cloud servers.
As of 2Q25, TrendForce observed that HBM3e still carried a price premium of more than four times DDR5, reflecting both performance value and tight supply availability.
At the same time, TrendForce raised its late-2025 DRAM outlook, projecting conventional DRAM contract prices to rise 18–23% in 4Q25 (up from an earlier 8–13% view), a move that signals a sharper tightening than previously expected.
This matters because contract pricing—rather than spot pricing—dominates how big OEMs and hyperscalers budget memory for server rollouts and hardware refresh cycles.
Key segments tied to AI
| Memory segment | What it powers in AI systems | Why supply is tight now |
| HBM3e / HBM (stacked DRAM) | AI accelerators and GPU servers | HBM competes for DRAM wafer capacity and remains structurally constrained as AI ramps. |
| Server DDR5 DRAM | CPU memory in AI/cloud servers | Cloud expansion and platform upgrades lift demand; TrendForce expects DDR5 contract prices to keep rising through 2026 (especially 1H). |
| Future HBM4 | Next-gen AI accelerators | TrendForce forecasts 2026 HBM shipments above 30 billion Gb and notes HBM4 advancement brings higher costs and price premiums above 30% in estimates. |
How suppliers are responding
Memory makers are leaning into pricing power while also reshaping product mix—because not all bits are equally profitable, even when demand is strong.
TrendForce expects DDR5 contract prices to maintain an upward trajectory through 2026 (especially in the first half), and it projects a profitability shift where DDR5 profitability is expected to surpass HBM3e starting in 1Q26.
That anticipated shift creates a strategic tradeoff: suppliers may allocate more capacity to server DDR5 to maximize earnings, even while trying to keep HBM supply commitments intact for major AI customers.
TrendForce notes suppliers could also attempt to raise HBM3e ASPs to maintain balanced profitability across portfolios if DDR5 becomes the clearer near-term profit leader.
Company earnings also show how sharply AI-linked memory is lifting financial performance.
Micron reported fiscal Q1 2026 revenue of $13.64 billion (up from $11.32 billion the prior quarter and $8.71 billion a year earlier), with GAAP net income of $5.24 billion, highlighting the strength of the memory upcycle tied to data-center demand.
In South Korea, SK hynix reported Q3 2025 revenue of 24.4489 trillion won and operating profit of 11.3834 trillion won, and said results were driven by strong sales of HBM and high-performance server products.
SK hynix also said HBM supply discussions for next year were completed, and that HBM4 shipments would begin in Q4 2025 with broader sales expansion planned for the following year—an indicator that customers are locking in longer-term supply amid tightness.
Price outlook snapshot (from industry tracking)
| Item | Latest stated view | Timing |
| Conventional DRAM contract pricing forecast | 18–23% growth (revised up from 8–13%) | 4Q25 |
| Server DDR5 contract pricing trend | Expected to stay on an upward path, particularly in 1H 2026 | 2026 |
| HBM vs DDR5 pricing relationship | HBM3e premium observed at more than 4x DDR5 | 2Q25 |
| HBM 2026 shipment scale (forecast) | Total HBM shipments forecast to exceed 30 billion Gb | 2026 |
What it means for buyers
For hyperscalers and AI developers, higher memory costs can raise the all-in price of AI servers—especially for configurations that combine cutting-edge GPUs with large HBM stacks and high-capacity DDR5.
Even when compute chips draw the headlines, memory often becomes the gating item because it is essential for feeding processors with data fast enough to keep expensive accelerators fully utilized.
For PC and device makers, the risk is less about HBM directly and more about second-order effects: as suppliers chase margins in server and AI products, availability and pricing of mainstream DRAM can tighten.
TrendForce’s view that the market faces a “structural supply shortage” into 2026 signals that buyers may need longer lead times, stronger inventory planning, and multi-quarter supply agreements.
Procurement teams are also watching product transitions—HBM3e to HBM4, and older DDR generations to DDR5—because transitions can constrain yields early on and further limit effective supply even when fabs are running at high utilization.
In this environment, many buyers will likely prioritize predictable delivery over the lowest possible price, since missing memory shipments can delay entire server racks and data-center deployment schedules.
Final thoughts
The near-term trajectory points to continued tightness: TrendForce expects DDR5 contract prices to keep rising through 2026 while capacity competition between DDR5 and HBM shapes what comes off production lines.
What happens next depends on how quickly suppliers can expand output and improve yields—particularly for next-generation HBM—without triggering the boom-bust oversupply cycles the memory industry is known for.
For now, the clearest signal from both pricing forecasts and company results is that AI is shifting memory from a cyclical commodity story into a capacity-constrained critical input for the global compute buildout.
