Samsung Electronics has reportedly decided to review and postpone without a new date its initial production plan for the D1d DRAM, the seventh generation of 10 nm class DRAM, after yields failed to meet internal targets and expected profitability did not justify starting production under current conditions. The information comes from an article by IT Chosun published on April 22, and has not been officially confirmed by Samsung at this time. Therefore, it should be read as a move reported by internal sources, not as a formal corporate announcement.
If this turn of events is confirmed, the impact would go beyond a simple manufacturing adjustment. The D1d appears in the Korean report as a critical piece for the future roadmap of HBM5E, a product not part of Samsung’s immediate 2026 offerings but key to the next major phase of high-bandwidth memory for AI accelerators. In other words, the issue may not be in Samsung’s most immediate short-term plans but rather in the point where the company must support its next technological leap against SK hynix and Micron.
The context helps explain why this matters. Samsung announced in February that it had already begun shipping its HBM4 commercially to clients and expected to start sampling HBM4E in the second half of 2026. Additionally, at GTC 2026, Samsung publicly showcased HBM4E as part of its push for AI infrastructure. This indicates the company is maintaining its short-term roadmap, at least for the upcoming generations, but the reported slowdown of D1d suggests the real bottleneck might be one step higher in the more advanced upcoming iteration of the roadmap.
In fact, Samsung had communicated a much more optimistic outlook on its HBM development in March. News outlets like Reuters and other Korean media quoted company executives stating they were accelerating production and saw no significant issues with AI memory supply, while also seeking multi-year contracts with major clients to reduce demand volatility. This narrative aligns with a company eager to reposition itself strongly in the HBM market. However, the IT Chosun report introduces a significant crack: speeding up development is one thing; turning that technology into profitable and stable production is quite another.
The problem isn’t designing the memory, but manufacturing it with sufficient performance
In semiconductors, a new generation isn’t just won in the lab. It’s achieved when the process delivers enough performance to support volume, quality, and profit margins. IT Chosun reports that Samsung’s management recently reviewed the status of the D1d and concluded that, given current performance levels and expected returns, production is not feasible. The article adds that process complexity has skyrocketed and that the company has preferred not to push the line if it compromises quality or worsens ROI.
This kind of decision fits well within the industrial logic of the memory business. A performance improvement or decline of just a few points can translate into billions of won in results. Therefore, what might seem externally as a technical delay is actually a matter of capital allocation and manufacturing discipline. Starting advanced production too early can support technological narratives but might also erode margins and execution capacity if the process isn’t mature.
The HBM market leaves little room for error. Reuters reported last month that SK hynix controls about 57% of the HBM market, compared to approximately 22% for Samsung, demonstrating the internal pressure on each process node upgrade. At the same time, Samsung has signed significant agreements to supply advanced memory for AI platforms, such as the memorandum with AMD for future Instinct and EPYC solutions. This underscores the importance of keeping its roadmap credible, both in presentations and industrial execution.
The slowdown of D1d doesn’t derail HBM4, but complicates the next phase
The key here is not to mix timelines. The IT Chosun report does not suggest that Samsung will stop pushing HBM4 or that it plans to sample HBM4E this year. In fact, their public statements remain consistent: HBM4 is already in commercial production, and sampling of HBM4E is still scheduled for the second half of 2026. The potential setback would be in the next generation, for which D1d would be necessary according to the Korean article. This would not affect the short term but could strain the mid-term strategy in AI.
Additionally, Samsung explained on its technical blog after GTC 2026 that HBM5 plans to use a core die manufactured in 1c DRAM and a base die in 2 nm process, while still showcasing HBM4E as one of its flagship products for the new AI wave. This official roadmap indicates the company still has room to defend its immediate next generations without relying entirely on D1d at every level. That’s precisely why the reported delay is significant: it indicates the issue isn’t with the already commercialized generation but with the one needed to sustain leadership further down the line.
Another possible interpretation is that Samsung has been trying for months to accelerate its HBM development cycle to keep pace with major AI accelerators. However, development and production are not the same. The company can shorten design times, expand facilities, or showcase prototypes at trade shows, but if the memory node supporting a future generation doesn’t meet performance and profitability thresholds, the actual schedule shifts. This is exactly what the Korean report suggests.
A troubling signal amid the AI memory race
The underlying message is unwelcome for Samsung because it arrives during a period of strong financial results driven by the memory business’s booming AI market and rising DRAM prices. Reuters reported in early April that the company was heading toward record quarterly profits, fueled by AI boom and DRAM price increases. While this improvement offers relief, it also raises expectations: when the market is rewarding advanced memory, any signs of delay in the next generation are seen as a competitive warning.
At present, the accurate picture isn’t a roadmap collapse but rather a serious potential review of a key piece in the next stage. HBM4 remains on track, HBM4E is still on the 2026 radar, and Samsung maintains agreements and deployments that support its AI race. Nevertheless, if D1d requires more time, the real battle will shift to the area that hurts industry most: who can reach profitable volume first, with stable quality and a sufficiently mature offer to capture the next big wave of contracts.
Frequently Asked Questions
Has Samsung officially confirmed the delay of D1d DRAM?
No. The information comes from a report by IT Chosun based on internal sources. Samsung has not yet released an official statement confirming this indefinite delay.
Does this affect the HBM4 that Samsung is already shipping to clients?
Not directly. Samsung announced in February that it had begun shipping HBM4 to clients and maintains that HBM4E sampling is still scheduled for the second half of 2026.
Why is D1d so important to Samsung’s strategy?
Because, according to IT Chosun, this DRAM generation is a key piece for future HBM5E solutions. If the process doesn’t reach adequate performance, it could strain the most advanced part of the AI memory roadmap.
What is Samsung’s current position compared to SK hynix in HBM?
Reuters reported in March that SK hynix controls about 57% of the HBM market, with Samsung holding around 22%. This explains the competitive pressure on each new advanced memory generation.
via: it.chosun