Memory prices may rise again as AI demand gives chip makers more leverage

Memory makers may gain even more pricing power in 2027 as demand for AI hardware keeps pressure on HBM, DRAM, and other high value memory products. New TrendForce estimates suggest that some DDR5 RDIMM modules have become more profitable per wafer than HBM, giving suppliers a stronger position as they negotiate future AI memory contracts.

That matters because HBM has become one of the most important parts of the AI hardware boom. Nvidia, custom AI chip makers, cloud companies, and hyperscalers all need large amounts of high bandwidth memory to power advanced accelerators. But the report suggests that memory suppliers may no longer feel pressured to prioritize HBM unless buyers accept better pricing terms.

TrendForce says the issue comes down to contract timing and market movement. Memory contracts often use annual pricing, which means suppliers can be locked into prices even as market conditions change quickly. With memory prices rising sharply during the AI boom, manufacturers have not always been able to raise contract prices fast enough to match demand.

Now, as negotiations shift toward HBM4 for 2027, suppliers may have more room to push back. TrendForce compared HBM and DRAM using die size, yield rates, and per gigabit pricing. Its analysis found that HBM per wafer revenue fell below DDR5 64GB RDIMM revenue in the first quarter of 2026. That also made certain RDIMM modules more profitable than HBM.

This gives memory manufacturers a simple negotiating advantage. If AI companies do not accept higher HBM prices, suppliers can allocate more limited production capacity to DRAM products that may deliver stronger profitability. That would tighten HBM supply further and could push prices up anyway.

The situation shows how complicated the AI supply chain has become. AI chip makers do not only need advanced processors. They also need enough memory, packaging capacity, substrates, power delivery, and cooling. A shortage or pricing shift in any one part can affect the whole system.

HBM demand is expected to remain strong through 2026 and 2027, but the reasons may change. TrendForce expects custom AI chips to drive much of the demand in 2026, while Nvidia’s Rubin Ultra platform is expected to become a major driver in 2027. That would keep pressure on suppliers even as buyers try to manage costs.

The likely result is more expensive AI infrastructure. If memory costs rise, companies building AI servers may face higher system prices. Those costs could eventually affect cloud AI services, enterprise AI deployments, and the economics of running large models.

For normal PC buyers, the direct impact is less clear, but the wider memory market is already tight. When AI demand absorbs more production capacity, it can affect DRAM and storage pricing elsewhere. That is one reason memory upgrades, SSDs, and high capacity modules have become more expensive in recent months.

The key point is that memory makers are no longer only suppliers reacting to demand. They now have leverage. AI companies need HBM, but suppliers can compare that business against profitable DDR5 RDIMMs and decide where their wafers are best used.

If TrendForce’s analysis proves accurate, 2027 could bring another round of HBM price pressure. AI firms may still pay because they need the memory to keep building larger systems. But the balance of power appears to be shifting, and memory manufacturers may be in a stronger position than they have been in years.