Korean giant SK Hynix is reportedly delaying production of its next-gen memory chips, which could impact Nvidia’s next-gen Rubin line of GPUs.

According to ZDNet Korea, Hynix will increase capacity of its HBM4 chip offering in the third quarter of 2026, instead of the second quarter as originally planned.

The latest generation high-bandwidth memory chip is a key part of the Rubin GPU architecture from Nvidia, which is scheduled for release in 2026.

Reports had swirled in recent weeks that Rubin could face a production delay, claims which Nvidia has denied.

Sources familiar with the matter told ZDNet that while Hynix had planned to gradually increase HBM4 production in the first half of 2026, the increased chance of a delay in the Nvidia Rubin launch schedule has made it re-focus on HBM3E production for existing Blackwell chips.

The firm has also been facing high demand for HBM3E capacity amid the AI-driven memory boom which has seen dynamic random access memory (DRAM) and NAND flash memory prices spike between 50% and 300% in the past six months.

SK Hynix rival Samsung has reportedly raised its memory prices by 60%, putting it on good footing to re-claim its position as the world’s biggest DRAM provider from Hynix, which led in market share in the last quarter at 33.2%.

Memory skirmish in Seoul

While it has yet to put the chips into mass production, Samsung is also working on an HBM4 offering, along with U.S. competitor Micron.

Samsung and SK Hynix together command around 70% of the DRAM market; conversely, The Korea Economic Daily reported Samsung Electronics' memory division is capable of supplying only 70% of current DRAM orders, with the squeeze so tight that its semiconductors division is rejecting orders from Samsung’s own phone making segment so as to prioritize enterprise clients, in a similar fashion to Micron closing its consumer division last week.

At an investor briefing hosted by Morgan Stanley last month, a Samsung official told The Korea Economic Daily: “Customers want long-term contracts of several years, but Samsung Electronics does not want to tie up volumes with a specific customer in a situation where prices are rising rapidly.”

The exception though may be the formidable Nvidia, for whom Samsung will supply half of the Small Outline Compression Attached Memory Module 2 (SOCAMM2) that Nvidia requires for Rubin production.

In addition, Samsung reportedly sent HBM4 samples to Nvidia in November and is currently awaiting final approval.

Impact on Nvidia?

Recent reports citing Fubon Research claimed Nvidia was going back to the drawing board for a slight redesign of Rubin to better compete with AMD's upcoming rival GPU, the MI450.

But Nvidia has moved to quash claims Rubin will be delayed, with the GPU due to drop sometime in 2026 as part of the chip giant's 'annual rhythm.'

The company said the claims were false and that Rubin remains on schedule.

Back in August, Nvidia revealed both its Rubin GPU and upcoming Vera CPU line have been taped out – the final stage where chip designs are made ready for fabrication.

While it remains to be confirmed if Rubin delay rumors are accurate, Nvidia has some form for postponing products: its current-gen Blackwell GPUs were pushed back by three months after manufacturer TSMC uncovered a production error related to the processor die that connects two Blackwell GPUs on a GB200.

There were also suggestions its flagship 72-GPU AI rack-scale server needed rejigging after the processors inside were overheating when linked together. In addition, its BlueField-4 data processing unit (DPU), for example, was finally released this October after first being announced in 2021.