Nvidia’s decision to test versions of its upcoming Rubin Ultra chip with less memory could paradoxically increase demand for high-bandwidth memory, providing another boost to leading suppliers such as Micron Technology and SK Hynix.
UBS analyst Timothy Arcuri said Monday that Nvidia appears to have moved to “de-spec” its upcoming Rubin Ultra offering, a term used when certain specifications or capabilities of a chip are reduced, MarketWatch reported.
The Information reported that Nvidia is testing different versions of the chip with lower memory capacity because of concerns over the availability of high-bandwidth memory, or HBM.
At first glance, using less HBM in each chip could appear negative for memory suppliers.
But Arcuri believes the change could allow Nvidia to produce more chips, potentially resulting in greater overall HBM consumption in 2027 than previously expected.
The development comes as HBM supply remains tight and pricing for newer generations of the technology strengthens.
UBS said pricing for HBM4 and HBM4E is now “even stronger than our prior expectations.”
Micron is one of the world’s leading HBM suppliers, alongside South Korea’s SK Hynix and Samsung Electronics.
The implications could be particularly important for SK Hynix, which remains the largest HBM supplier globally and an important supplier to Nvidia.
HBM pricing outlook strengthens
Arcuri said “memory suppliers are rewidening” the premium charged for HBM, prompting him to sharply raise his expectations for pricing.
The analyst now expects HBM average selling prices to rise by about 79% from a year earlier, compared with his previous estimate of a 67% increase.
The pricing strength reflects the growing amount of memory required by AI accelerators and the difficulty manufacturers face in rapidly adding HBM capacity.
HBM is increasingly critical to the performance of advanced AI systems because it allows processors to access large amounts of data at high speeds.
For Micron, stronger HBM pricing could translate into a significant improvement in earnings and cash generation.
The company has invested heavily in expanding its HBM manufacturing capacity as demand from AI data centres accelerates.
Arcuri expects Micron’s earnings per share to remain above $160 in 2029. He also forecasts more than $450 billion in cumulative free cash flow through 2028.
“To us, this increasingly looks like a structural reset in earnings power — where this kind of durability/visibility should command a shift toward a broader semi multiple,” Arcuri wrote.
A broader semiconductor multiple would imply that investors could become willing to value Micron more like other high-performing chip companies, rather than assigning it the lower valuation typically associated with the cyclical memory industry.
Micron stock remains up by more than 172% but is down by 30% from its all-time high.
NAND demand also shows strength
The bullish outlook extends beyond HBM.
Contract pricing for NAND flash memory is also moving higher, according to Arcuri.
He attributed the improvement to stronger-than-expected demand for server and storage solid-state drives, which has helped offset weakness in the personal-computer market.
Although Arcuri expects sequential growth in NAND average selling prices to be less powerful than previously forecast, he said “the demand backdrop remains constructive.”
The analyst raised his expectations for NAND bit demand growth to 23% this year and 26% in 2027.
A bit refers to the smallest unit of digital information and is commonly used as a measure of memory consumption.
The improvement in server and storage demand adds another potential source of support for memory manufacturers as AI infrastructure investment expands.
Micron sees supply remaining tight beyond 2027
Micron executives have also offered a bullish assessment of the supply-demand balance.
Chief business officer Sumit Sadana said at the KeyBanc Technology Leadership Forum on Monday that the company expects tight memory supply to persist beyond 2027 as customer demand continues to rise.
“With all of the efforts that we are doing, we still don’t have line of sight as to when the supply is going to be able to meet demand, because demand continues to escalate at a very rapid pace over time,” Sadana said, according to a FactSet transcript.
Sadana said memory demand during the AI era is “very different” from previous cycles because AI technology is still developing rapidly.
That rapid technological progress is increasing the amount of memory required by advanced computing systems, while simultaneously making it difficult for manufacturers to predict exactly where future demand will settle.
HBM expansion puts pressure on other memory
The HBM boom is also creating challenges elsewhere in the memory industry.
Sadana said rising HBM demand has put pressure on wafer supply for other memory products.
Expanding wafer capacity is not a quick process, he added, because it requires significant manufacturing investment and lengthy capacity expansion timelines.
That dynamic could help sustain pricing across the broader memory market if manufacturers remain unable to add capacity quickly enough.
Micron’s customers have so far shown little sign of retreating despite higher prices.
Sadana said customers in the company’s data-centre business and other segments continue to seek additional memory supplies even as prices rise.
That willingness to pay could provide further evidence that AI infrastructure demand is supporting a fundamentally different memory cycle.
The post Why Micron, SK Hynix could gain from Nvidia’s de-spec of Rubin Ultra? appeared first on Invezz
