DDR5 AI Server Memory Market Size, Share & Forecast to 2035

The DDR5 AI Server Memory Market is valued at USD 5.95 billion in 2026 and is projected to reach USD 6.80 billion by 2035, up from USD 2.14 billion in 2025, a CAGR of 12.25% from 2025 to 2035 (DC Market Insights estimate). The market covers the Double Data Rate 5 (DDR5) dynamic random-access memory (DRAM) modules that serve as host memory in artificial intelligence (AI) servers, meaning servers built around graphics processing units (GPUs), custom application-specific integrated circuits (ASICs) or other accelerators: registered dual in-line memory modules (RDIMMs), multiplexed rank DIMMs (MRDIMMs) and DDR5 memory attached over Compute Express Link (CXL). The build receipt: about 460,000 DDR5-hosted AI servers deployed in 2025 × 1,200 GB of host memory each × USD 4.00 per GB, with 97% of that memory on DDR5, gives USD 2.14 billion. MRDIMMs are the fastest-growing module type at 32.29% a year because AI host processors move to 12 and 16 memory channels at 8,800 to 12,800 megatransfers per second (MT/s), and the Middle East and Africa is the fastest-growing region at 18.13% a year as sovereign AI campuses start from a small base.

Key Market Statistics: How Big Is the DDR5 AI Server Memory Market in 2026?

DC Market Insights values the DDR5 AI Server Memory Market at USD 5.95 billion in 2026, 177.9% above the USD 2.14 billion of 2025, because the average DDR5 server module price doubled while host memory per AI server rose by 30%.

• Market size 2025: USD 2.14 billion
• Market size 2026: USD 5.95 billion
• Market size 2030: USD 7.92 billion
• Market size 2035: USD 6.80 billion
• CAGR 2025-2035: 12.25% in the base case, 9.12% in the low case and 15.23% in the high case
• Largest module type: standard RDIMMs of up to 64 GB, 52.0% of 2025 value (USD 1.11 billion)
• Fastest-growing module type: MRDIMMs, 32.29% a year
• Leading region: North America, 50.0% of 2025 value (USD 1.07 billion)
• Fastest-growing region: Middle East and Africa, 18.13% a year
• Largest country: the United States, USD 985.21 million in 2025 (46.0% of global value)
• Market leaders: Samsung Electronics, SK hynix and Micron Technology (DC Market Insights assessment)

Executive Summary

The Global DDR5 AI Server Memory Market size was valued at USD 7.53 million in 2020, reached USD 2.14 billion in 2025, and is anticipated to reach USD 6.80 billion by 2035, at a CAGR of 12.25% during the forecast period.

Report attribute	Details
Historical Period	2020-2024
Base Year	2025
Forecast Period	2026-2035
DDR5 AI Server Memory Market Size 2025	USD 2.14 Billion
DDR5 AI Server Memory Market Size 2026	USD 5.95 Billion
DDR5 AI Server Memory Market, CAGR	12.25%
DDR5 AI Server Memory Market Size 2035	USD 6.80 Billion

The 2020 value of USD 7.53 million reflects validation modules only: JEDEC, the semiconductor standards body, published the JESD79-5 DDR5 standard on 14 July 2020, and AI servers ran on DDR4 until DDR5 server processors shipped in late 2022 and 2023. DDR5 then took 45% of AI server host memory in 2023, 85% in 2024 and 97% in 2025, so the 2020 to 2025 growth rate of 209.6% a year describes a technology switch rather than demand. From 2025 the drivers are volume and capacity. DDR5 host memory shipped into AI servers grows from about 552 petabytes (PB) in 2025 to about 3,490 PB in 2035, because AI capacity additions rise and each server carries more memory: NVIDIA’s DGX B200 ships with 2 TB of system memory, configurable to 4 TB. Value peaks at USD 7.96 billion in 2028 and eases to USD 6.80 billion by 2035, as the next memory generation takes 43% of AI server host memory and price per GB normalises from the 2026 high.

Market Scope: What Does the DDR5 AI Server Memory Market Cover?

The DDR5 AI Server Memory Market covers the DDR5 DRAM modules installed as host processor memory in accelerator-equipped servers, valued at module selling price in current US dollars; DC Market Insights sizes it at USD 2.14 billion in 2025 and USD 5.95 billion in 2026.

Included. Standard DDR5 RDIMMs of up to 64 GB, high-capacity RDIMMs of 128 GB and above, including three-dimensional stacked (3DS) modules, MRDIMMs, and CXL memory expansion modules built on DDR5 dies, when they are installed in servers that carry GPUs, custom ASICs, field-programmable gate arrays (FPGAs) or other AI accelerators. The scope covers new AI servers and AI servers bought to refresh older accelerator fleets.

Excluded. High-bandwidth memory (HBM) stacked next to the accelerator, low-power memory such as LPDDR5X used by Arm-based AI host processors and in SOCAMM modules, DDR4 modules, DDR5 in general-purpose servers without accelerators, client and workstation memory, and the registering clock driver (RCD) and data buffer chips sold to module makers, which are counted inside module prices.

How the segments fit together. The module type split adds standard RDIMMs, high-capacity RDIMMs, MRDIMMs and CXL modules to the USD 2.14 billion total. The memory capacity per server, accelerator type and end user splits re-cut the same total by server configuration, by the accelerator inside the server and by buyer.

Why Do DDR5 AI Server Memory Market Estimates Differ?

Public estimates under this market name differ because they draw the line in different places; DC Market Insights counts only DDR5 host memory in accelerator-equipped servers, which gives USD 5.95 billion in 2026.

Broader definitions add DDR5 in every server, including general-purpose servers that run databases, web services and virtual machines, which multiplies the total several times over. Some also add HBM, which is the largest memory pool in an AI server by value but sits on the accelerator package and is sized in our High Bandwidth Memory Market report. Others add LPDDR5X and SOCAMM modules on Arm host processors; NVIDIA’s Vera CPU is built on LPDDR5X memory, and Samsung Electronics named HBM4, DDR5 and SOCAMM2 as separate focus products in July 2026. Narrower definitions count only memory chips, not modules, or only training servers.

Element	In this report	Why
DDR5 RDIMMs and MRDIMMs in AI servers	Yes, USD 1.24 billion in 2025 for MRDIMMs and standard RDIMMs combined	Host memory feeding the accelerators
High-capacity and 3DS RDIMMs in AI servers	Yes, USD 813.87 million in 2025	Same function, larger modules
CXL memory expansion on DDR5	Yes, USD 85.66 million in 2025	Adds host memory capacity beyond DIMM slots
HBM on accelerators	No	Separate device on the accelerator package
LPDDR5X and SOCAMM on Arm hosts	No	Different memory standard
DDR5 in general-purpose servers	No	No AI accelerator in the server

Out-of-scope terms that broader estimates count include general-purpose server memory, HBM stacks, LPDDR5X, SOCAMM, DDR4, client memory and memory interface chips sold separately. In-scope buyer groups include hyperscale cloud providers, GPU cloud and colocation providers, enterprises, small and medium-sized enterprises (SMEs) that buy single AI servers, research institutions and sovereign AI programmes.

Market Drivers: What Drives the DDR5 AI Server Memory Market?

Three drivers lift DDR5 host memory spending in AI servers from USD 2.14 billion in 2025 to USD 6.80 billion by 2035: more AI capacity, more memory per server and more memory channels per processor.

AI Capacity Additions Rise from 4.2 GW to 17.8 GW a Year

New AI data center capacity rises from about 4.2 GW in 2025 to about 17.8 GW in 2035 in the DC Market Insights capacity model, and AI fleets installed five years earlier are refreshed on top. Spending signals confirm the pace. NVIDIA reported fiscal 2026 Data Center revenue of a record USD 193.7 billion, up 68%, and USD 89.0 billion in the quarter to July 2026, up 117%. Dell Technologies booked a record USD 60.9 billion of AI server orders in the same quarter. Micron Technology expects server unit growth in the high-teens percentage range in both 2026 and 2027. Together these lift DDR5-hosted AI server deployments from about 460,000 in 2025 to about 683,600 in 2030.

Host Memory per AI Server Grows from 1,200 GB to 4,500 GB

Average DDR5 host memory per AI server rises from about 1,200 GB in 2025 to 3,000 GB in 2030 and 4,500 GB in 2035 (DC Market Insights estimate), because agentic and long-context inference keeps model weights, key-value caches and data pipelines in host memory. Eight-GPU systems already ship with 2 TB: NVIDIA documents the DGX H100 with 2 TB using 32 DIMMs and the DGX B300 with 2 TB by default and up to 4 TB with 128 GB DIMMs. Dell’s PowerEdge XE9785 for AMD Instinct MI355X accelerators takes 24 DDR5 RDIMMs and up to 6 TB. Each extra terabyte at the 2025 price of USD 4.00 per GB adds about USD 4,000 of memory to a server.

Memory Channels per Processor Rise from 12 to 16

The number of DDR5 channels per host processor rises from 8 on the first DDR5 server processors to 12 on current AMD EPYC and Intel Xeon 6 parts and up to 16 on the next AMD generation, which raises the number of DIMM slots per server by up to 33%. AMD’s EPYC 9005 processors carry 12 channels of DDR5 memory per CPU (central processing unit), and AMD showed EPYC 9006 in July 2026 with up to 16 DDR5 channels and JEDEC-standard MRDIMM support up to 12,800 MT/s. Micron also noted in March 2026 that traditional server demand is robust, driven by agentic AI workloads, and SK hynix said in January 2026 that demand for conventional server memory increased sharply.

Market Trends: How Is DDR5 AI Server Memory Changing?

Module mix shifts to higher value per slot: high-capacity RDIMMs and MRDIMMs rise from 44.0% of 2025 value to 77.0% by 2035.

256 GB Modules Reach Volume Qualification in 2026

High-capacity and 3DS RDIMMs grow at 14.42% a year, from USD 813.87 million in 2025 to USD 3.13 billion by 2035, as 256 GB modules replace 64 GB and 96 GB parts. SK hynix completed Intel Xeon 6 validation of a 256 GB DDR5 RDIMM built on 32 Gb dies in December 2025, and Micron began sampling a 256 GB DDR5 server module at up to 9,200 MT/s in May 2026.

MRDIMMs Grow from 6.0% to 31.0% of Value

MRDIMMs grow from USD 128.51 million in 2025 to USD 2.11 billion by 2035, because they double the data rate of a DIMM slot without new processor pins. Intel’s Xeon 6 was the first server processor to support MRDIMMs in September 2024, and the Xeon 6980P supports DDR5-6400 and MRDIMM-8800. Each MRDIMM carries one MRCD chip and ten MDB chips (multiplexed registering clock driver and data buffer), which adds cost per module, and JEDEC announced in April 2026 that the second-generation MRCD02 standard was expected soon for 12,800 MT/s modules.

Price Swings Dominate 2026

The average DDR5 server module price rises from about USD 4.00 per GB in 2025 to USD 8.00 per GB in 2026 (DC Market Insights estimate). Micron reported DRAM price increases of about 20% in the quarter to November 2025, the mid-60s percentage range in the quarter to February 2026 and the high-teens percentage range in the quarter to August 2026.

Market Challenges: What Holds the DDR5 AI Server Memory Market Back?

Two challenges cap DDR5 spending in AI servers: Arm host processors that use low-power memory, and the move to the next DRAM generation after 2029.

Low-Power Memory Takes Half of AI Host Capacity by 2035

The share of AI capacity built on DDR5-hosted x86 servers falls from 65% in 2025 to 55% in 2030 and 50% in 2035 (DC Market Insights estimate), because rack-scale systems with Arm host processors use LPDDR5X. NVIDIA’s Vera CPU, launched in March 2026, is built on LPDDR5X and delivers up to 1.2 TB/s of bandwidth. Every 5 points of share lost to low-power memory removes about USD 680 million of 2035 value from this market.

The Next DRAM Generation Takes 43% of AI Host Memory by 2035

DDR5’s share of AI server host memory peaks at 99% in 2028 and falls to 92% in 2030 and 57% in 2035 as the next registered memory generation ramps on new processors. This transition turns 2028 into the value peak at USD 7.96 billion. Price also works against value: after the 2026 high, the model assumes DDR5 falls to USD 4.20 per GB by 2030 and USD 3.42 per GB by 2035 as new DRAM capacity comes online, which would cut 2035 value by 57% against 2026 prices.

Market Opportunities: Where Are the New Revenue Pools?

Two pools add the most new value: memory-rich inference servers above 2 TB and CXL memory expansion.

Servers Above 2 TB Reach USD 4.42 Billion by 2035

Servers with more than 2 TB of DDR5 grow from USD 428.35 million in 2025 to USD 4.42 billion by 2035, at 26.29% a year, and rise from 20.0% to 65.0% of value. Platforms already allow it: Supermicro’s 8U systems for AMD Instinct MI350X and MI355X take 24 DIMM slots for up to 9 TB of DDR5-6000 RDIMMs, and Dell’s XE9780 and XE9680 take 32 DDR5 RDIMMs and up to 4 TB.

CXL Memory Expansion Reaches USD 748.32 Million by 2035

CXL memory expansion modules grow from USD 85.66 million in 2025 to USD 748.32 million by 2035, at 24.20% a year, because they add host memory beyond the DIMM slots. The CXL Consortium released the CXL 3.2 specification in December 2024. Micron’s CZ120 modules come in 128 GB and 256 GB and add up to 2 TB per server, and Samsung’s CMM-D module uses DDR5 behind a CXL interface over PCIe Gen5 (Peripheral Component Interconnect Express).

Pricing: How Much Does DDR5 Memory for an AI Server Cost?

DDR5 memory for an average AI server costs about USD 4,800 in 2025 (1,200 GB at USD 4.00 per GB) and about USD 15,390 in 2035 (4,500 GB at USD 3.42 per GB), according to DC Market Insights estimates.

Year	Average price (USD per GB)	Host memory per AI server (GB)	DDR5 content per AI server (USD)
2020	8.00	768	6,144 (validation modules only)
2025	4.00	1,200	4,800
2030	4.20	3,000	12,600
2035	3.42	4,500	15,390

The 2026 average of USD 8.00 per GB puts a 2 TB server at about USD 16,400 of DDR5, against about USD 8,200 at 2025 prices. SK hynix reported in July 2026 that high-performance products for AI servers led price increases. MRDIMMs carry a premium of about 20% to 30% per GB over RDIMMs of the same capacity, and 256 GB 3DS modules about 15% (DC Market Insights estimates), because of extra buffer chips and stacked dies. A 64 GB RDIMM, the most common standard part, cost about USD 256 in 2025 at the average price.

Standards and Regulation: Which Rules Shape DDR5 AI Server Memory?

Four JEDEC and industry standards and one export-control regime shape the USD 2.14 billion market.

• JEDEC JESD79-5 (DDR5 SDRAM, synchronous DRAM). Published on 14 July 2020; sets the DDR5 device standard used by every module in scope.
• JEDEC JESD79-5C. Adds Per-Row Activation Counting (PRAC) for data integrity and extends speed grades to 8,800 MT/s (April 2024).
• JEDEC MRDIMM standards. MRCD and MDB logic for multiplexed rank modules; the MRCD02 generation for 12,800 MT/s was announced as expected in April 2026.
• CXL 3.2. The CXL 3.2 specification (December 2024) defines memory expansion and pooling for CXL modules.
• US export controls. The US Bureau of Industry and Security (BIS) tightened controls on advanced semiconductors including HBM in December 2024; controls on accelerators decide where high-end AI servers, and the DDR5 inside them, can be shipped.

Platform qualification works like a standard in practice: a module must pass validation on each host processor before hyperscale buyers deploy it, as SK hynix’s December 2025 Intel Xeon 6 validation shows.

Market Segmentation: Which Segment Leads the DDR5 AI Server Memory Market?

Standard RDIMMs lead the DDR5 AI Server Memory Market with 52.0% of 2025 value, servers with 1 TB to 2 TB of memory lead by configuration with 50.0%, GPU servers lead by accelerator with 80.0% and hyperscale cloud providers lead by buyer with 62.0%.

By Module Type: Standard RDIMMs Lead with 52.0%

Module type	2025 (USD million)	Share 2025	2035 (USD million)	CAGR 2025-2035
Standard RDIMM (up to 64 GB)	1,113.72	52.0%	816.35	-3.06%
High-Capacity and 3DS RDIMM (128 GB and above)	813.87	38.0%	3,129.34	14.42%
MRDIMM	128.51	6.0%	2,108.91	32.29%
CXL Memory Expansion Modules	85.66	4.0%	748.32	24.20%
Total	2,141.76	100.0%	6,802.92	12.25%

Standard RDIMM spending shrinks at 3.06% a year because 256 GB modules and MRDIMMs fill the slots of new servers; MRDIMMs are the fastest-growing segment.

By Memory Capacity per Server: 1 TB to 2 TB Leads with 50.0%

Capacity per server	2025 (USD million)	Share 2025	2035 (USD million)	CAGR 2025-2035
Up to 1 TB	642.53	30.0%	476.20	-2.95%
1 TB to 2 TB	1,070.88	50.0%	1,904.82	5.93%
Above 2 TB	428.35	20.0%	4,421.90	26.29%
Total	2,141.76	100.0%	6,802.92	12.25%

By Accelerator Type: GPU Servers Lead with 80.0%

Accelerator type	2025 (USD million)	Share 2025	2035 (USD million)	CAGR 2025-2035
GPU Servers	1,713.41	80.0%	4,353.87	9.77%
Custom ASIC Servers	364.10	17.0%	2,244.96	19.95%
Other Accelerator Servers	64.25	3.0%	204.09	12.25%
Total	2,141.76	100.0%	6,802.92	12.25%

Custom ASIC servers grow fastest at 19.95% a year as hyperscale operators deploy their own training and inference chips, most of them hosted by x86 processors with DDR5.

By End User: Hyperscale Cloud Providers Lead with 62.0%

End user	2025 (USD million)	Share 2025	2035 (USD million)	CAGR 2025-2035
Hyperscale Cloud Providers	1,327.89	62.0%	3,673.58	10.71%
GPU Cloud and Colocation Providers	385.52	18.0%	1,360.58	13.44%
Enterprise and Sovereign AI	428.35	20.0%	1,768.76	15.24%
Total	2,141.76	100.0%	6,802.92	12.25%

Enterprise and sovereign AI buyers grow fastest at 15.24% a year, because on-premises inference servers are configured with more host memory per GPU than cloud training clusters.

DC Exclusive: The Host Memory per AI Server Model and Module Type × Accelerator Matrix

At USD 862.59 million in 2025, standard RDIMMs in GPU servers are the largest cell of the matrix and 40.3% of all DDR5 AI server memory spending. The matrix below cross-tabulates what is bought against the accelerator it serves. It is a DC Market Insights model, not a shipment count: it is built from 34 inputs, namely 16 capacity cohorts, 5 annual curves (AI share, DDR5-hosted share, kW per server, GB per server and DDR5 share of host memory), a price curve, 4 mix tables and 8 regional and country weights, and then tested against 3 bound checks.

2025, USD million	GPU Servers	Custom ASIC Servers	Other Accelerator Servers
Standard RDIMM (up to 64 GB)	862.59	212.44	38.69
High-Capacity and 3DS RDIMM (128 GB and above)	659.47	132.89	21.51
MRDIMM	116.07	10.21	2.23
CXL Memory Expansion Modules	75.28	8.56	1.82
Total	1,713.41	364.10	64.25

GPU servers spend 6.8% of their DDR5 budget on MRDIMMs and 4.4% on CXL modules, the highest shares of any accelerator type, because eight-GPU systems are the first to use 12-channel processors at full speed. Custom ASIC servers spend 58.3% of theirs on standard RDIMMs, because many hyperscale ASIC hosts run one DIMM per channel at 64 GB.

The host memory model has inputs any reader can check: AI capacity added each year (4.2 GW in 2025), IT power per AI server (about 6.5 kW in 2025, 20 kW in 2035) and memory per server (1,200 GB in 2025). Dividing DDR5-hosted AI megawatts by kW per server gives servers; multiplying by GB per server, DDR5 share and price per GB gives the market. The model and its assumptions are explained on our research methodology page.

Regional Insights: Which Region Leads the DDR5 AI Server Memory Market?

North America leads the DDR5 AI Server Memory Market with 50.0% of 2025 value, and the Middle East and Africa is the fastest-growing region at 18.13% a year, because sovereign AI campuses in Saudi Arabia and the United Arab Emirates (UAE) add large GPU clusters to a small base.

North America Leads with 50.0% Share

North America accounts for USD 1.07 billion in 2025 and USD 3.06 billion by 2035, growing at 11.08% a year. Hyperscale budgets drive it: Alphabet guided 2026 capital expenditure of USD 175 billion to USD 185 billion and Meta USD 130 billion to USD 145 billion, and Supermicro reported fiscal 2026 net sales of USD 39.1 billion.

Asia Pacific Holds 28.0% Share

Asia Pacific accounts for USD 599.69 million in 2025 and USD 2.04 billion by 2035, growing at 13.03% a year. The region also builds most of the hardware: Taiwan, where most AI servers are assembled, reported record 2025 exports of USD 640.7 billion, and South Korea is home to two of the three main DRAM makers. China accounts for USD 203.89 million in 2025, held back by US accelerator export controls.

Europe Holds 14.0% Share

Europe accounts for USD 299.85 million in 2025 and USD 918.39 million by 2035, growing at 11.84% a year. Germany, the United Kingdom and France take 57.0% of European spending in 2025, mostly through cloud regions and national AI supercomputing programmes.

Middle East and Africa Holds 4.5% Share

The Middle East and Africa accounts for USD 96.38 million in 2025 and USD 510.22 million by 2035, the fastest growth of any region at 18.13% a year. Saudi Arabia and the UAE together take 64.0% of regional spending in 2025, at USD 28.91 million and USD 32.77 million.

Latin America Holds 3.5% Share

Latin America accounts for USD 74.96 million in 2025 and USD 272.12 million by 2035, growing at 13.76% a year. Brazil takes 45.0% of the region at USD 33.73 million in 2025.

Country Analysis: Which Countries Spend the Most on DDR5 AI Server Memory?

The United States is the largest country market for DDR5 AI server memory at USD 985.21 million in 2025, 46.0% of global value, followed by China at USD 203.89 million (9.5%) and Japan at USD 83.96 million (3.9%). The 12 largest countries account for 82.6% of 2025 spending.

Rank	Country	Region	2025 (USD million)	Share 2025	2035 (USD million)	CAGR 2025-2035
1	United States	North America	985.21	46.0%	2,785.79	10.95%
2	China	Asia Pacific	203.89	9.5%	571.45	10.86%
3	Japan	Asia Pacific	83.96	3.9%	265.31	12.19%
4	Canada	North America	74.96	3.5%	229.60	11.84%
5	Germany	Europe	65.97	3.1%	192.86	11.32%
6	South Korea	Asia Pacific	65.97	3.1%	224.50	13.03%
7	United Kingdom	Europe	62.97	2.9%	183.68	11.30%
8	India	Asia Pacific	53.97	2.5%	346.95	20.45%
9	Australia	Asia Pacific	47.98	2.2%	163.27	13.03%
10	Singapore	Asia Pacific	47.98	2.2%	132.66	10.71%
11	France	Europe	41.98	2.0%	128.57	11.84%
12	Brazil	Latin America	33.73	1.6%	119.73	13.51%

India is the fastest-growing large country market at 20.45% a year, ahead of Brazil at 13.51% and South Korea at 13.03%, as national AI compute programmes and hyperscale regions add GPU clusters. The country figures are a DC Market Insights allocation of the regional totals, weighted by each country’s estimated share of installed and new AI server capacity.

Forecast Scenarios: How Could the 2035 Forecast Change?

The DDR5 AI Server Memory Market reaches USD 6.80 billion by 2035 in the base case, with a range of USD 5.13 billion to USD 8.84 billion across the low and high scenarios.

Scenario	Assumption for 2035	2035 (USD billion)	CAGR 2025-2035
Low	4,000 GB per AI server; USD 2.90 per GB	5.13	9.12%
Base	4,500 GB per AI server; USD 3.42 per GB; DDR5 57% of host memory	6.80	12.25%
High	5,000 GB per AI server; USD 4.00 per GB	8.84	15.23%

The low case reflects faster DRAM capacity additions and a quicker move of AI host memory to low-power modules. The high case assumes that long-context inference keeps raising memory per server and that DRAM supply stays tight well beyond 2026.

Competitive Insights: Who Leads the DDR5 AI Server Memory Market?

• Samsung Electronics
• SK hynix
• Micron Technology
• Kingston Technology
• Nanya Technology
• CXMT (ChangXin Memory Technologies)
• Winbond Electronics
• SMART Global (Penguin Solutions)
• Netlist
• ADATA Technology
• Montage Technology
• Rambus
• Renesas Electronics
• Astera Labs
• Microchip Technology

The DDR5 AI Server Memory Market is highly concentrated: DC Market Insights estimates that Samsung Electronics, SK hynix and Micron Technology together supply about 93% of the market, because server-grade 32 Gb DDR5 dies and platform qualification are concentrated in these three DRAM makers. Samsung said in July 2026 that growth in demand for server DRAM is expected to accelerate. SK hynix reported a record second quarter in which AI server products led price increases. Micron reported record fiscal third-quarter 2026 revenue of USD 41.46 billion, against USD 9.30 billion a year earlier, and has completed customer qualifications of 8,800 MT/s server RDIMMs. Interface chip suppliers set module speed: Rambus reported 2025 product revenue of USD 347.8 million, up 41%, Montage Technology’s interconnect chip line sold RMB 5,139 million (Chinese yuan) in 2025, and Renesas Electronics launched a Gen6 DDR5 RCD at 9,600 MT/s in November 2025.

Recent Developments

• In December 2025, SK hynix completed Intel Xeon 6 validation of a 256 GB DDR5 RDIMM built on 32 Gb dies (Source: SK hynix newsroom).
• In April 2026, JEDEC announced new MRDIMM interface logic standards, with MRCD02 for 12,800 MT/s expected soon (Source: JEDEC).
• In May 2026, Micron began sampling a 256 GB DDR5 server module at up to 9,200 MT/s for AI infrastructure (Source: Micron press release).
• In July 2026, AMD showed EPYC 9006 processors with up to 16 DDR5 channels and MRDIMM support up to 12,800 MT/s (Source: AMD).
• In September 2026, Dell Technologies reported record AI server orders of USD 60.9 billion and AI server revenue of USD 16.4 billion for its second fiscal quarter (Source: Dell Technologies).
• In September 2026, Micron reported DRAM price increases in the high-teens percentage range and completed qualifications of 8,800 MT/s server RDIMMs (Source: Micron prepared remarks).

We follow these launches, qualifications and price moves as they happen in our data center industry updates. Buyers planning AI server memory can compare host DDR5 with accelerator memory in the High Bandwidth Memory Market report and with the interface chips on each module in the DDR5 Server Memory RCD IC Market report.

Methodology: How We Built the DDR5 AI Server Memory Market Model

DC Market Insights built this market bottom-up from 16 annual capacity cohorts (2020 to 2035). AI capacity each year is new data center capacity times the AI share (12.0 GW × 35% = 4.2 GW in 2025), plus AI capacity installed five years earlier that is refreshed with new servers (0.4 GW in 2025). The DDR5-hosted share of that capacity (65% in 2025) divided by IT power per AI server (6.5 kW in 2025) gives about 460,000 servers; multiplying by 1,200 GB of host memory, a 97% DDR5 share and USD 4.00 per GB gives USD 2.14 billion. In 2035 the same chain reads 31.0 GW of new and refreshed AI capacity, a 50% DDR5-hosted share, 20 kW per server, 775,500 servers, 4,500 GB, a 57% DDR5 share and USD 3.42 per GB, for USD 6.80 billion. The total was split by 4 module types, 3 capacity bands, 3 accelerator types, 3 end users and 5 regions, and every split sums to the market total in every year.

Because no supplier discloses DDR5 sales to AI servers, the result was tested against three bounds. Ceiling one: the 2025 market of USD 2.14 billion is 1.1% of NVIDIA’s fiscal 2026 Data Center revenue of USD 193.7 billion, in line with about USD 8,200 of DDR5 in a 2 TB server that sells for several hundred thousand dollars. Ceiling two: the 2026 market of USD 5.95 billion is 14.4% of Micron’s revenue in a single quarter, USD 41.46 billion, while three suppliers share it. Floor: 460,000 servers with only 512 GB each at the 2023 low of USD 3.20 per GB would still spend USD 753.66 million, so the modeled USD 2.14 billion sits well above the minimum. The analyst named on this page built the model, and the reviewer checked the arithmetic, the sources and the dates before publication.