If you are sourcing standard DRAM or NAND right now, you are competing against the most well-funded supply chain operation in semiconductor history — and losing.
SK Hynix, the world's leading HBM supplier, is in the process of redirecting its production capacity toward custom, co-developed memory products built exclusively for Nvidia. The commodity memory market — DDR5, server DIMMs, LPDDR4, eMMC, NAND — is what remains after that allocation is satisfied. And what remains is shrinking.
Founder & Lead Analyst of Semiconductor Business Intelligence, Claus Aasholm, has been tracking this shift through purchasing commitments, earnings data, and capacity signals across the memory supply chain. His conclusion is unambiguous.
"Beyond HBM4, SK Hynix will move into custom HBM," Aasholm said. "That means they're not going to be a memory company anymore in its traditional form."
For procurement teams, that sentence should trigger an immediate review of your memory sourcing strategy.
The Supply Warning Hidden in SK Hynix's Q1 2026 Guidance
SK Hynix's Q1 2026 guidance did not get the attention it deserved. Buried beneath record profitability headlines were two numbers that tell the real story: flat DRAM shipments and negative NAND shipments.
This is not a company struggling with demand. This is a company that has made a deliberate choice about where its capacity goes — and standard memory products are not the priority.
HBM production is extraordinarily capacity-intensive. A wafer allocated to HBM is a wafer that does not become a DDR5 module, a server DIMM, or an automotive eMMC chip. As SK Hynix commits more of its fabs to HBM and eventually custom HBM, the supply available to the rest of the market contracts accordingly.
The pricing data confirms this is already happening. Aasholm flagged a detail that none of the major memory makers have publicly addressed: HBM's share of SK Hynix's revenue declined last quarter for the first time, not because HBM demand weakened, but because commodity DRAM prices are now rising so fast they are closing the gap with HBM.
"The price increases in non-HBM are now violent," Aasholm said. "The HBMs are guarded a little bit by long-term agreements, so you can't raise prices as fast as you can in the more commoditized memory products. None of them mentioned this, and I think that's because they don't want to tell their clients that it's now becoming more expensive than potentially HBM."
Standard commodity DRAM is approaching HBM price parity. If your procurement model assumes memory is a stable, predictable input cost, that model is now broken.
How Nvidia Locked Up the Supply Chain While Everyone Else Watched
The current shortage did not happen overnight. It is the direct result of Nvidia making procurement decisions years ago that most of the industry either ignored or could not match.
While memory markets were still depressed, Nvidia moved in with upfront payments, multi-year capacity commitments, and co-development agreements. SK Hynix, coming out of one of the worst memory downturns on record, accepted. The result was a guaranteed supply relationship that has compounded in Nvidia's favor every quarter since.
"Even two years ago, you heard from Hynix that they had sold all of the HBM," Aasholm said. "That's because Nvidia just came in and bought it. No questions asked."
The custom HBM transition that Aasholm describes is that relationship taken to its logical conclusion. SK Hynix will not just be selling Nvidia a standard HBM product — it will be co-designing memory architectures built specifically around Nvidia's next-generation AI accelerators, locked in under long-term contracts that functionally remove that capacity from the open market permanently.
"They're going to try and move almost as much capacity as they can into custom HBMs made for Nvidia on long-term contracts and co-developed," Aasholm said. "And that's going to create a completely new memory market."
A new memory market. One in which the largest, most advanced supplier is no longer a merchant selling to the highest bidder, but a dedicated partner serving one customer. If you are not that customer, you are working with what is left.
What Is Left: DRAM, NAND, and Automotive Supply Outlook
DRAM is tight and getting tighter. Clean room shortages, not equipment, not investment willingness, are the hard ceiling on capacity expansion. No meaningful new DRAM supply is expected before mid-2027.
"You can't just buy equipment," Aasholm said. "There's a shortage of clean rooms. So we are still some distance away from reasonable capacity increase."
Pricing is expected to continue to rise through 2026 and remain elevated through 2027 until new production capacity starts to come online
NAND has now entered the same cycle DRAM went through months earlier. Hyperscaler demand is consuming available supply faster than it can be replenished. SSD pricing is rising. Older formats are seeing unexpected demand spikes. There is early evidence of renewed hard disk drive interest from buyers priced out of NAND-based storage, a sign of how distorted the market has become.
Automotive and industrial memory is among the most exposed. LPDDR4, eMMC, and DDR3 are legacy products that Korean memory manufacturers have never prioritized. Automotive-grade certification is expensive, volumes are relatively small, and margins are thin. When supply gets tight, these segments get deprioritized first.
The consequences are already visible. Some automotive OEMs are reporting 50% fulfillment rates on eMMC orders. Spot market pricing for automotive-grade memory is rising sharply. At least one major server manufacturer has begun shipping systems with minimum memory configurations — or none at all — telling customers to source DIMMs independently and expect delivery in two quarters.
This is not a supply chain stress test. It is a supply chain failure for segments that sat too far down the priority queue for too long.
What Procurement Teams Should Do Now
The window for proactive action is narrow. Here is where it stands by supplier:
SK Hynix is increasingly committed to HBM and custom programs. Standard memory allocations will continue to tighten as more capacity is directed toward co-development with Nvidia. If you do not have a long-term agreement in place, assume spot market conditions going forward.
Samsung is in a similar trajectory, though its HBM execution challenges have kept more standard capacity available in the near term. That may not last as Samsung works to close the gap with Hynix on HBM quality and yield.
Micron is the most accessible major supplier for customers outside the hyperscaler tier. It has shown more willingness than its Korean competitors to support automotive and industrial segments and is worth prioritizing for relationship development if you are in those markets.
CXMT and Chinese suppliers are frequently cited as relief valves but should not be factored into near-term planning. Scaling advanced memory production takes years, and equipment shortages are slowing Chinese ramp timelines regardless of state investment levels. "They still need to scale," Aasholm said. "It's not like you can go and build something in ten seconds."
The broader strategic reality is that the memory market has been permanently reorganized around a tier of customers, Nvidia, the major hyperscalers, Apple, who secured their supply through early, large, and unconditional commitments. Everyone else is in a different market now, one defined by constrained allocation, rising prices, and limited ability to accelerate supply regardless of what you are willing to pay.
As Aasholm put it: "It doesn't matter how much pressure you apply to it, water doesn't budge, and memory capacity doesn't budge right now. So there's only one thing that can give and that's price." Plan accordingly.
Browse available memory now to secure DRAM and NAND supply as hyperscaler demand continues to tighten allocation.
For ongoing insights into memory market dynamics, follow Claus Aasholm and Semiconductor Business Review on LinkedIn and Substack.
Key Takeaways
- SK Hynix is transitioning toward custom HBM co-developed with Nvidia, permanently removing that capacity from the standard memory market
- Commodity DRAM prices are rising faster than HBM prices — a sign of how severe standard memory tightness has become
- Flat DRAM and negative NAND shipments in Q1 2026 signal no near-term supply relief
- Automotive and industrial memory segments face the most severe allocation constraints
- Micron is the most accessible major supplier for non-hyperscaler customers
- Memory pricing relief is not expected before mid-2027; new fab capacity not online until early 2028
- Companies without long-term supply agreements should plan for continued spot market pricing pressure through 2026 and into 2027
Frequently Asked Questions
Why is SK Hynix reducing standard DRAM and NAND supply?
SK Hynix is not reducing supply because of weak demand — it is redirecting capacity toward high bandwidth memory (HBM) and, increasingly, custom HBM programs co-developed with Nvidia. HBM production is far more capacity-intensive than standard memory, meaning every wafer committed to HBM directly reduces the output available for DDR5, LPDDR4, eMMC, NAND, and other commodity products. As that transition accelerates beyond HBM4, the share of SK Hynix capacity available to the open market will shrink further.
When will DRAM and NAND prices come down?
Based on current capacity data and clean room availability, meaningful price relief is not expected before mid-2027. New fab capacity is unlikely to come online at scale until early 2028. Prices are expected to continue rising through mid-2026, with a potential plateau around August 2026 — but remaining elevated well beyond that point. Companies planning procurement budgets around a near-term price correction should revise those assumptions.
Which memory segments are most at risk right now?
Automotive and industrial memory are the most exposed. Products like LPDDR4, eMMC, and DDR3 have historically been deprioritized by Korean memory manufacturers, and the current shortage environment has made that worse. Some automotive OEMs are already reporting fulfillment rates as low as 50% on eMMC orders. Server DRAM and DDR5 are also tight, with at least one major server manufacturer now shipping systems with no memory installed and asking customers to source DIMMs independently.
Is Chinese memory production a viable alternative supply source?
Not in the near term. Chinese manufacturers like CXMT are scaling, but advanced memory production takes years to ramp and requires equipment that is itself in short supply. Chinese memory output is not yet at a scale or quality level that meaningfully offsets tightness in the broader market. Procurement teams should not factor Chinese supply into near-term planning horizons.
Which major memory supplier is most accessible for non-hyperscaler buyers?
Micron is currently the most accessible of the three major DRAM suppliers for customers outside the top hyperscaler tier. Unlike its Korean competitors, Micron has demonstrated more consistent willingness to support automotive and industrial customers, even as those segments become less attractive to SK Hynix and Samsung. For procurement teams in those markets, building or strengthening a Micron relationship now is worth prioritizing.
What should procurement teams do immediately given these conditions?
Three things: first, audit your current memory exposure — identify which products, lead times, and supplier relationships are most vulnerable to further tightening. Second, do not assume spot market availability will improve; if you need memory in 2026 and into 2027, secure it now at current pricing rather than waiting for a correction that is unlikely to come. Third, diversify supplier relationships where possible, with Micron as the most viable near-term alternative for customers being squeezed out of SK Hynix and Samsung allocations.
