Who makes DRAM for AI data centers
DRAM (DDR5 RDIMM/MRDIMM, LPDDR5X for Grace)
constrained
Off-package system memory — DDR5 modules for host CPUs and LPDDR5X soldered near Nvidia's Grace CPU — that holds the working data a server's processors operate on.
DRAM sits a step further from compute than HBM: it serves the host CPU rather than the accelerator, reached over the memory channels described by the buffer chips in the next part, and it is what memory modules (2.03) package into a pluggable form. AI servers lean on host DRAM for data staging, batching, and checkpoint handling ahead of GPU reads, and newer module standards like MRDIMM exist specifically to push more bandwidth through the same socket as core counts rise. A structural pressure worth watching is that HBM and standard DRAM are produced on the same manufacturing lines by the same small set of makers, so wafer capacity a supplier commits to HBM is capacity it is not committing to commodity DRAM — a trade-off that can tighten DRAM supply and pricing even though DRAM itself has not become any more technically exotic. For an investor, this layer is less about a single chokepoint than about how the same three suppliers allocate capacity across HBM, DRAM, and LPDDR product lines as AI demand pulls at all three simultaneously.
AI delta: AI servers' data-staging and checkpointing needs raise host DRAM content per node, and HBM production competing for the same fab capacity tightens standard DRAM supply alongside it.
"This is the most undersupplied the DRAM market has been in decades," per TechInsights DRAM/memory analyst Mike Howard, with AI infrastructure spending expected to keep the shortage running into 2027 or beyond (EE Times, 2026-09-18).
Companies on this part (6)
| Company | Role | Ownership | Ticker | Price | 1d | Confidence | Note |
|---|---|---|---|---|---|---|---|
| Micron | manufacturer | public | MU | 878 USD | +3.8% | disclosed | · source |
| Nanya | manufacturer | public | 2408.TW | 414 TWD | -5.2% | disclosed | · source |
| Samsung Electronics | manufacturer | public | 005930.KS | 215500 KRW | +0.2% | disclosed | · source |
| SK Hynix | manufacturer | public | 000660.KS | 1277000 KRW | +4.4% | disclosed | · source |
| Winbond | manufacturer | public | 2344.TW | 155 TWD | -3.1% | disclosed | · source |
| CXMT | manufacturer | private | disclosed | China · source |
Who competes here
The same table read the other way: every company above competes for DRAM sockets. Confidence tags matter — an inferred placement is a lead, not a confirmed rivalry.
Seen in real machines
DRAM is named in 4 of the rack teardowns and supercomputer breakdowns on this site. Each line below is a piece of that machine. The companies shown are the ones named on that line, not necessarily the maker of this part.
- Nvidia GB300 NVL72 · rack teardown
- Grace CPU and LPDDR5X — Micron, Samsung Electronics, SK Hynix
- JUPITER · supercomputer
- Leonardo · supercomputer
- System memory (Booster 512 GB DDR4-3200 per node; Data Centric 512 GB DDR5-4800 per node) — Micron
- Isambard-AI · supercomputer
- System memory (4x 120 GB LPDDR5X per node, 460 GB usable, on the Grace Hopper modules) — Nvidia
Nearby parts
- Previous in layer: HBM
- Next in layer: Memory modules
- Layer: Layer 2 — Memory & storage
From the blog
Watch
Independent videos — no affiliation; nothing loads from YouTube until you click.
Latest feed
- How Goodram Validates DRAM and SSD Reliability in Europe — EE Times, 2026-10-22Goodram details its European testing methodology for validating the reliability of DRAM and SSDs for industrial and embedded applications.
- Micron pulls the plug on future 4F² DRAM after two decades of research in dramatic shift — Tom's Hardware, 2026-10-22Micron announced it is discontinuing the development of 4F² DRAM because future process nodes offer better bit density improvements for memory.
- [AI Infrastructure Insight] How AI infrastructure will be designed for the future — newsroom:sk-hynix, 2026-10-22Effective real-world AI application requires improvements across faster compute, wider memory bandwidth, higher-performance networking, better storage, and more stable power infrastructure.
- Samsung reportedly overhauls DRAM, NAND R&D incentives to retain talent and accelerate next-gen memory chips — DigiTimes, 2026-10-22Samsung is reportedly restructuring its R&D incentives to accelerate the development of future DRAM and NAND memory chips and improve talent retention.
- China's IC designers expand into AI server memory interconnects beyond DDR5 — DigiTimes, 2026-10-22Chinese IC designers are increasingly entering the high-speed interconnect market to support the growing demand for server CPUs, memory, and connectivity driven by AI applications.
- Edge AI MCU Features Multimodal Compute, Tiered Memory — EE Times, 2026-10-21
- [AI Ecosystem] The semiconductor paradigm shifts toward memory — newsroom:sk-hynix, 2026-10-21The industry trend is shifting towards a greater reliance on memory components, suggesting changes in accelerator and memory demand.
- [AI Ecosystem] The semiconductor paradigm shifts toward memory — newsroom:sk-hynix, 2026-10-21The semiconductor industry is shifting its focus towards memory solutions, indicating a strong trend in memory technology.
- Samsung's HBM4 supply lock-in exposes limits of China's CXMT as Apple and Tesla seek DRAM alternatives — DigiTimes, 2026-10-20Samsung's anticipated high level of pre-committed memory contracts is noted alongside China's CXMT expanding capacity, which draws attention from Western tech firms looking for alternative DRAM suppliers.
- Elon Musk points to China's 2028 DRAM supply surge to challenge Samsung, SK Hynix pricing power — DigiTimes, 2026-10-19Elon Musk suggested that a projected surge in China's DRAM supply by 2028 could challenge the pricing dominance held by Samsung and SK Hynix.
Questions about DRAM
- What is DRAM?
- Off-package system memory — DDR5 modules for host CPUs and LPDDR5X soldered near Nvidia's Grace CPU — that holds the working data a server's processors operate on.
- Which companies make DRAM for AI data centers?
- This catalog lists 6 companies on DRAM, 5 of them with listed equity: Micron, Nanya, Samsung Electronics, SK Hynix, Winbond, CXMT.
- How does AI change demand for DRAM?
- AI servers' data-staging and checkpointing needs raise host DRAM content per node, and HBM production competing for the same fab capacity tightens standard DRAM supply alongside it.
- DRAM — is it a supply bottleneck?
- This catalog flags DRAM as "constrained". "This is the most undersupplied the DRAM market has been in decades," per TechInsights DRAM/memory analyst Mike Howard, with AI infrastructure spending expected to keep the shortage running into 2027 or beyond (EE Times, 2026-09-18). The bottlenecks page lists every part carrying a sourced supply constraint.
- How reliable are these DRAM supplier placements?
- The 6 placements on this part break down as 6 disclosed. "disclosed" means the company or a filing says so; "reported" comes from trade press or a reference-design list; "inferred" is a taxonomy lead with no citation yet.
Sources: placements without a source URL come from the taxonomy seed and are tagged accordingly. How this data is built.· Markdown · JSON