SK Hynix begins shifting some HBM capacity back to regular DRAM

Has HBM demand shrunk?
Here's the direct conclusion: No
On the contrary, this is SK Hynix's current profit-maximizing choice
>> Because regular DRAM is now more profitable than HBM
And there’s no rush to expand HBM capacity
According to a report by The Korea JoongAng Daily today
SK Hynix is slowing down the expansion of HBM4
Reallocating some production lines originally planned for HBM3E to HBM4
To supply the tight supply of regular DRAM
The reason given by the report is
SK Hynix’s HBM revenue share has exceeded 40%, with a solid lead
Rather than continuing to pour capacity into HBM expansion wars
It’s better to shift to more profitable regular DRAM and earn more
There are two counterintuitive reasons:
1️⃣ Regular DRAM is more profitable than HBM right now
According to a report by The Korea JoongAng Daily
By Q1 2026, the unit price per Gb of regular DRAM will still be lower than HBM
> But the operating profit margin gap has already reversed
Regular DRAM surpasses HBM by more than 15%
The same report cites a forecast from Daishin Securities
The operating profit margin of regular DRAM could theoretically reach 90% within the year
Compared to SK Hynix’s own profit levels
The overall operating profit margin that quarter was about 72%, a record high
And now, the profit margin for regular DRAM is even higher than that
Over the past two years, the three major manufacturers shifted a large portion of wafer capacity to HBM for higher profits
> Regular DRAM supply has been insufficient for a long time, now in severe shortage
According to SK Hynix’s Q1 2026 financial report
The average selling price of DRAM that quarter increased by about 60% quarter-over-quarter
According to TradingKey citing industry expectations
Contract prices in Q2 are expected to rise another 50–70% quarter-over-quarter
Single-quarter operating profit could reach 60 trillion to 70 trillion Korean won
Low costs, sharp price increases
Almost all of the rising part is pure profit
Shifting capacity to higher-margin products is the most straightforward calculation
There’s also an amplifier here—
According to TrendForce’s December 2025 estimate
1GB of HBM consumes roughly the same wafer area as 4GB of regular DRAM
Micron also provided an approximate 3:1 conversion ratio
Meaning: each wafer piece converted from HBM back to regular DRAM
Releases 3–4 times the sellable bits
Splitting some HBM capacity to DRAM
Both volume and price of DRAM go up together
The Korea JoongAng Daily also quoted sources
>> Management cannot ignore that Samsung is earning huge profits from regular DRAM, not HBM
Samsung demonstrated first, SK Hynix followed
The rivalry of the top two, always like this
2️⃣ The share of HBM4 is already locked, no need to fight for it anymore
According to a report by KeyBanc in April
Due to delays in HBM4 validation at that time
Nvidia lowered this year’s Rubin production target from 2 million units to about 1.5 million units
But by June, the validation hurdle was cleared
Huang Renxun confirmed in Seoul
All three major players have passed HBM4 supply certification on the Vera Rubin platform
Rubin has entered full-scale mass production, with delivery expected in Q3
The key is market share—
A January report by Yonhap News Agency
>> Nvidia’s Rubin HBM4 is nearly 70% allocated to SK Hynix
Higher than the previous expectation of just over 50%
SK Hynix’s large batch of paid samples has passed final validation without issues
In other words,
The share SK Hynix should get for this generation of HBM4 is basically secured
Since certification is done and share is locked in,
> There’s no reason to continue pouring capacity into expansion wars for this generation
The marginal capacity freed up can be used for more profitable regular DRAM
A judgment by Goldman Sachs echoes the same idea
SK Hynix just needs to maintain its dominant share of HBM3 / HBM3E and HBM4
No need to compete in an expansion race again
>> So, is HBM really cooling off? The current generation is booming
According to TrendForce data
Demand for Blackwell in the first half of 2026 will surge
Nvidia raised shipment targets and increased HBM3E orders
While lowering Rubin’s schedule
So, the real demand truth is
HBM3E is running at full capacity to grab market share, HBM4’s share is already locked, waiting for customer platform volume
The variable to watch now shifts to the next generation, HBM4E
According to a January report by Hankyung
Samsung is racing ahead with 1c process technology and has matched SK Hynix’s HBM4 pricing
Previously, its 12-layer HBM3E was about 30% cheaper
This generation, HBM4E,
is the real battleground for SK Hynix and Samsung’s next round of competition
So, returning to the initial question,
SK Hynix shifting to DRAM is not a sign of waning HBM demand
HBM4’s share is already locked in, and regular DRAM is more profitable now
HBM3E is still in full production
SK Hynix is just reallocating marginal capacity to the most profitable area right now