Loading today’s stories
Invest1 distinct publisher2 min readPublished
TrendForce puts 3Q26 DRAM contracts up 13 to 18% and NAND up 10 to 15%, slower than recent quarters, though nothing on the supply side loosened; what gave way was the PC and smartphone buyer's budget, strained by record prices.
The Investor · Invest desk
invest
Memory absorbs 110% of the 2027 increase in major cloud capex1 distinct publisher
build
China's accelerator swap makes Cambricon supply, not export policy, your ship-date risk1 distinct publisher
build
H200s reach China at 2.5% of the order book, and Hong Kong holds the rest2 distinct publishers
build
SMIC's first $3 billion quarter comes with a wafer price increase attached1 distinct publisher
Compiled by The InvestorSomething wrong?How this is made
Hold the midpoints for four quarters and the arithmetic of moderation looks strange: 15.5% a quarter on DRAM compounds to 78% over a year, 12.5% on NAND to 60% [16], and even the bottom of the DRAM range annualizes to 63% while the top reaches 94% [18]. DRAM's midpoint runs three points above NAND's [19], so the flash side is where the buyer is pushing back hardest.
The supply side, meanwhile, is moving away from the customers doing the complaining. TrendForce has leading manufacturers keeping production cuts disciplined while Taiwanese DDR4 additions fail to cover the reduction [14], suppliers still steering capacity toward AI applications in a way that keeps LPDRAM tight and prices climbing [11], and major suppliers accelerating their withdrawal from consumer DRAM, with the orders migrating out of televisions and set-top boxes keeping underlying demand from falling even as that end market stays weak [13]. Capacity leaving the PC socket shows no sign of coming back. A price increase that decelerates while available supply keeps narrowing points to the buyer running out of room to pay, with the fab's output unconstrained.
Graphics DRAM is the cleanest test in the release. NVIDIA's RTX PRO 6000 Blackwell did not generate the GDDR7 wave suppliers expected, and softer notebook shipments pulled GDDR6 and GDDR7 demand down with it, yet both rose in line with the broader DRAM trend because capacity was flexibly reallocated to other mainstream products [12]. End demand fell and the contract price went up anyway, which is what a market looks like when the marginal seller sets the number.
Two readings would blunt this. Server DRAM stays undersupplied through the quarter, but its gains moderate partly because a share of procurement runs through long-term supply agreements [8], so some of the reported deceleration is paper signed months ago rather than resistance now. And TrendForce names a higher comparison base as one of the two moderating forces alongside weaker consumer demand [20], which means a smaller percentage on a record base can still be a larger absolute increase than a bigger percentage was a year earlier, and the release does not publish the levels required to work out which.
I would still take the demand-destruction read, or rather the version of it that says consumer sockets are being abandoned by suppliers rather than lost to competitors. This is probably wrong in one specific way, and here is the test: if the 4Q26 survey shows PC DRAM allocation restored at the same time as prices decelerate further, then this was an ordinary supply story all along, and the affordability wall was a coincidence of timing.
Ranked by verification strength, evidence, and original report placement.
TrendForce forecasts conventional DRAM contract prices to rise 13-18% QoQ in 3Q26.
TrendForce expects NAND Flash contract prices to increase 10-15% QoQ in 3Q26.
Record-high contract prices mean customers from consumer markets such as PCs and smartphones are reaching their affordability limit, leading to more moderate price increases in 3Q26.
TrendForce says the DRAM market will remain extremely tight in the third quarter of 2026.
TrendForce attributes the moderation in DRAM contract price increases to weaker demand from consumer applications and the impact of a higher comparison base.
NAND Flash demand continues to be driven primarily by AI inference and large-scale data center deployments, and the projected 10-15% increase is a noticeably slower pace than in previous quarters.
Distinct publishers with included, body-backed reporting in this cluster.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One release, no data behind it
Both ranges, the tightness call, the inventory picture and the capacity shifts trace to a single TrendForce press release, and TrendForce publishes the conclusions without the survey: no absolute contract prices, no sample of deals, no named supplier or OEM, no methodology. The specificity of the bands and the firm's standing as the industry's default price tracker keep this from scoring at the floor. What it cannot do is corroborate itself, and nothing else in our coverage tests it.
Behaviour described, never counted
There is real market behaviour in here, and some of it has already happened rather than being forecast: OEMs stockpiled client SSDs through the first half and are now stalling negotiations, majors are walking away from consumer DRAM, part of server procurement is locked under long-term agreements. But every one of these arrives as a characterization. Not a single volume, inventory week, contract count or buyer name is attached, so the picture is broad, plausible and impossible to size.
Restrained framing, unaudited magnitude
TrendForce is telling a deceleration story, which is the opposite of hype, and our own headline follows it rather than the annualized arithmetic. The gap is narrow and comes from magnitude, not spin: a 13-18% quarterly move is an extraordinary number to publish with no absolute prices behind it, and the further the reader carries it, the thinner the ground gets. Compounded across four quarters the DRAM midpoint implies roughly 78% a year, which is arithmetic on one quarter's guidance and not a forecast anyone made.
The forecaster sells the forecast
TrendForce's business is the memory pricing survey this release summarizes, and headline percentage swings are what makes that subscription worth buying. That does not make the numbers wrong, but it does explain the shape: bands wide enough to be defensible, segment detail deep enough to advertise the full report, and no methodology given away. Worth noting too that the release attributes continued price strength partly to "disciplined production cuts by leading memory manufacturers" without pausing on how that phrasing reads in a market with three serious DRAM makers.
Trust the direction, not the decimal
TrendForce has the standing to be believed on where memory pricing is heading, and the internal logic here holds together: supply stays tight, capacity keeps moving to AI, and it is the consumer buyer's budget that cracks. What we cannot do from one interested release with no underlying figures is stand behind the specific bands or the segment-level assertions about NVIDIA card demand and unnamed suppliers exiting consumer DRAM. Directionally usable for planning; not something to model against without a supplier or OEM disclosure alongside it.