Why "NAND Prices Are Rising" Depends on the Product: Three TrendForce Releases, March to September

本文另有中文版：「NAND 漲價」漲的是哪一類？TrendForce 三份稿寫出兩種走向

On Aug. 25, TrendForce estimated that DRAM and NAND flash together would account for 47% of major cloud providers’ capital spending in 2026, and 68% in 2027. Four weeks later, on Sept. 21, it said U.S. cloud providers had raised their enterprise SSD demand forecasts within the previous two weeks.

Both headlines invite the same shorthand, that NAND is scarce and getting more expensive. TrendForce’s own releases from March, July and September say something more specific, because they describe several NAND products and the wording changes from product to product and from quarter to quarter. Every number below is a TrendForce forecast or estimate, and none is a realized transaction price.

Three releases, two products and one overall price, no single NAND story

The table sets the three releases side by side, and cells marked “not covered” mean the release did not address that product.

The next sections take the eMMC/UFS row first and the enterprise SSD row second, because those two moved furthest apart.

In March, eMMC and UFS had the tightest supply gap of any NAND product

TrendForce’s Mar. 31 outlook for the second quarter said eMMC and UFS (TrendForce ties their demand to smartphones, automotive and industrial use) had “the tightest supply gap across all product segments”, and it expected prices to rise sharply. The release ties demand to flagship phones with AI features, which still needed fast transmission even though the smartphone market was weak, and to a modest recovery in automotive and industrial demand.

On the supply side, the same passage says eMMC/UFS shares overlapping process capacity with enterprise SSDs but earns significantly lower margins. These are two different products, and what they share, in TrendForce’s description, is process capacity.

One quarter later, the 3Q outlook described eMMC and UFS supply differently

The Jul. 3 outlook for the third quarter reads differently, although March looked at the second quarter and July at the third, so the two statements are outlooks for different quarters. TrendForce wrote that eMMC and UFS supply, “which had previously been heavily constrained”, had become “relatively more abundant”.

The release gives its reasons: overall demand was weakening, OEMs were less willing to absorb higher costs, and end-market demand was softer. Together, TrendForce said, these had weakened suppliers’ pricing power, and prices were still expected to rise, but by smaller amounts than in the prior quarter.

The September release does not mention eMMC or UFS, so the July wording is the latest in this set, and it is not a reversal recorded at one moment.

March and July describe more capacity for enterprise SSDs; September describes stronger demand and possibly higher orders

The March release said NAND capacity was increasingly allocated to enterprise SSDs. It described orders as showing no signs of slowing and expected a clear shortage in 2026, with meaningful capacity expansion unlikely until late 2027 or 2028, while cloud providers, it said, were willing to accept higher prices and sign long-term agreements (LTAs).

The July release used similar language, saying NAND vendors were allocating more capacity to enterprise SSDs. It also added a limit, because shortages of internally sourced DRAM continued to constrain small-capacity, high-performance enterprise SSDs, and overall prices stayed on an upward trajectory.

The September release changes the subject from capacity to demand. It says recent procurement momentum suggested that fourth-quarter orders “could surpass the already elevated levels seen in 3Q26”, which is a possibility and not a result. TrendForce added that this had given suppliers greater control over capacity allocation and pricing in the enterprise segment.

North America and China are described as using enterprise SSDs differently

The September release names two applications, and it keeps the regions apart.

For North America, TrendForce points to agentic AI. As those systems spread, it says, demand for large volumes of real-time data retrieval and caching is growing quickly, and QLC storage is also seeing broader adoption in vector databases, where high capacity and cost efficiency suit large volumes of unstructured data. TrendForce describes QLC products as offering superior cost efficiency and capacity. It expects wider adoption of agentic AI to further increase QLC enterprise SSD penetration across North American cloud and data center applications.

For China, TrendForce points to AI architectures led by DeepSeek. They use high-capacity QLC enterprise SSDs for KV cache offloading, which means selectively moving cached data out of memory and onto those drives. TrendForce says this can improve inference efficiency at a lower cost, and it adds that the approach could drive broader adoption of QLC enterprise SSDs in China.

KV cache offloading appears in the release only for China, while for North America the named uses are retrieval, caching and vector databases. TrendForce also says demand drivers have widened beyond the early focus on training large language models toward more concrete applications. It notes that the increase is not limited to high-performance TLC products, and that QLC has become a significant growth catalyst.

The forecast increase fell from 70–75% to 10–15%, and it still points up

The Mar. 31 release forecast NAND contract prices up 70–75% quarter over quarter in 2Q26, and the Jul. 3 release forecast 10–15% for 3Q26. TrendForce attributed the slowdown to consumer buyers, because contract prices were at record highs, it wrote, and price tolerance among consumer customers had reached its limit. It described the forecast as “a noticeably slower pace than in previous quarters”.

Slower is not lower.

The forecast is still an increase, and the 10–15% is also an average for NAND overall rather than a number for enterprise SSDs. For that segment, the July release says only that prices stay on an upward trajectory, and the September release says the same for 4Q26 without a percentage.

A longer view comes from the Aug. 25 release. TrendForce says enterprise SSD prices rose around 35% in 2H25 and projects a cumulative rise of 235% in 2026, and those figures are cumulative rather than annual growth rates.

Cloud capex is large in TrendForce’s estimate, and it names two limits

The Aug. 25 release projects total capital spending by major cloud providers rising 98% in 2026 and another 50% in 2027, and the combined memory shares cited at the top cover DRAM and NAND together, not NAND alone.

Two limits appear in the same release.

First, some LTAs signed from 2Q26 onward include price ceilings that could limit further increases. Second, cloud providers could optimize their AI system memory more aggressively by reducing the memory capacity in each system, which would ease the cost of high prices or limited supply allocations.

One note on DRAM, which the series treats separately. TrendForce’s Sept. 23 DRAM Market Bulletin summary says it lifted its 4Q26 DRAM contract-price outlook, but the summary gives no percentage and says nothing about NAND. For NAND, the latest statement in this set is the Sept. 21 “could”.

The next NAND price headline needs one question first

The question is which product the headline is about.

If it concerns enterprise SSDs, TrendForce’s September wording supports an upward trend and a possible order increase in the fourth quarter. If it concerns NAND contract prices overall, the latest forecast in this set is 10–15% for 3Q26. If it concerns eMMC or UFS, the latest wording is from July, which describes supply as relatively more abundant and increases as more modest.

The next article in this series turns to consumer electronics, the customers whose price tolerance TrendForce says had reached its limit by July.

This article summarizes published statistics, press releases and news reports, with forecasts labeled apart from reported figures. It is not investment advice.