In this article

RAM and NVMe prices have surged as AI infrastructure competes for the same semiconductor capacity. We look at manufacturer timelines and capacity data to project when supply will catch up with demand.


The current memory shortage has created major price increases. OpenMetal’s buying data for high-end NVMe drives and high-end 64GB DDR5 shows that NVMe drive costs are up 223% from three years ago and 64GB DDR5 DIMMs are up 474% from two years ago. This is due to large orders of AI-related hardware that needs the same wafers as what DDR5 and NVMe drives need.

AI infrastructure demand competes directly with consumer and enterprise markets for the same semiconductor fabrication capacity. While manufacturers prioritize high-margin HBM and AI-specific products, concrete leading indicators suggest when this supply-demand imbalance might resolve.

The Supply Constraint Reality

The root cause is straightforward: AI data centers require massive amounts of high-bandwidth memory (HBM), and manufacturers have reallocated more than three times the wafer capacity to produce HBM chips compared to conventional DRAM. HBM requires a 3-to-1 trade ratio with DDR5, meaning every HBM wafer displaces three DDR5 wafers, and this ratio increases with future HBM generations.

As of Q3 2025, the top three manufacturers controlled 93% of the DRAM market: SK Hynix (34%), Samsung (33%), and Micron (26%), giving them significant pricing power. Samsung can fulfill only about 70% of current DRAM orders, as demand far outstrips output, while industry supply will remain substantially short of demand for the foreseeable future.

Manufacturer Analysis and Leading Indicators

Micron Technology

Concrete Timeline Milestones:

  • Idaho fab: DRAM output begins fiscal 2027 (September 2026-September 2027)
  • New York fabs: Production not expected until fiscal 2028 or later
  • HBM4 production: Ramping in Q2 2026
  • Singapore HBM packaging facility: Contributing meaningfully by calendar 2027

Financial Commitments:

  • $200 billion total U.S. expansion over 20+ years
  • $20 billion capex for fiscal 2026, up from prior $18 billion estimate
  • $6.4 billion in CHIPS Act funding confirmed
  • Target: 40% of DRAM production in U.S.

Key Constraint: Cleanroom space is necessary to address increased demand, and lead times for cleanroom build-out are lengthening across geographies.

Sources:

  • https://investors.micron.com/static-files/088991c5-a249-4f66-a0a6-258d9b66f3f9
  • https://investors.micron.com/news-releases/news-release-details/micron-and-trump-administration-announce-expanded-us-investments
  • https://investors.micron.com/news-releases/news-release-details/micron-ships-hbm4-key-customers-power-next-gen-ai-platforms
  • https://www.trendforce.com/news/2025/12/18/news-micron-reveals-three-culprits-behind-memory-crunch-and-why-it-wont-ease-soon/

Samsung Electronics

Production Capacity Expansion:

  • 1c DRAM capacity: Targeting 200,000 wafers per month by end 2026
  • 60,000 wafers/month by end 2025, expanding by 80,000 wafers by Q2 2026, and adding 60,000 more by Q4 2026
  • P4 factory scheduled to start production in 2026

Strategic Realignments:

  • Samsung recently shifted focus from HBM to DDR5 modules, freeing up around 80,000 DRAM wafers per month
  • Scaling back more than 20% of conventional DRAM production capacity in favor of advanced offerings like LPDDR5X and HBM
  • HBM4 samples delivered to NVIDIA, targeting Q2 2026 supply chain entry

Sources:

  • https://www.trendforce.com/news/2025/11/19/news-samsung-reportedly-plans-200k-1c-dram-wafersmonth-by-2026-about-one-third-of-its-total-output/

SK Hynix

Immediate Capacity Additions:

  • M15X fab construction completed November 2025 for early mass production
  • When M15X becomes fully operational, HBM capacity would increase by 20-30% compared to current levels
  • HBM capacity trajectory: 170,000 wafers per month by end 2025

Longer-term Infrastructure:

  • Yongin Semiconductor Cluster: First fab construction starts March 2025, completion May 2027
  • P&T7 advanced packaging fab: KRW 19 trillion investment, construction starts April 2026, operations by end 2027

Technology Leadership:

  • HBM4 development completed and ready for mass production
  • SK Hynix aims to ship HBM4 samples to NVIDIA as early as June 2025

Sources:

  • https://news.skhynix.com/sk-hynix-to-produce-dram-from-m15x-in-cheongju/
  • https://www.businesskorea.co.kr/news/articleView.html?idxno=258178

ChangXin Memory Technologies (CXMT)

Aggressive Expansion:

  • Monthly volume doubled to 200,000 wafers by Q1 2025, with forecasts pointing to 300,000 by 2026
  • 2025 DRAM output projected at 2.73 million wafers, a 68% jump from 1.62 million in 2024
  • Global market share could climb from 5% in 2023 to 12% by year-end 2025

Technology Progress:

  • DDR5 mass production delayed to late 2025 for quality improvements
  • HBM2E mass production planned for H1 2026, HBM3 samples by end 2025

Critical Constraints:

  • Added to U.S. Entity List January 6, 2025
  • Production capacity expansion heavily depends on tools made outside of China, with localization around 20%

Sources:

  • https://www.digitimes.com/news/a20250421PD218/cxmt-dram-samsung-sk-hynix-2025.html

Timeline Assessment for Supply Relief

2026: Modest Capacity Additions

  • Q2 2026: Samsung and SK Hynix HBM4 production ramp, Micron HBM4 production begins
  • H2 2026: Samsung’s 1c DRAM reaches 200,000 wafers/month target
  • End 2026: CXMT potentially reaches 300,000 wafers/month if not constrained by export controls

2027: Meaningful New Capacity

  • Early 2027: Micron Idaho fab begins production
  • May 2027: SK Hynix Yongin first fab completion
  • Calendar 2027: Micron Singapore HBM packaging contributes meaningfully
  • End 2027: SK Hynix P&T7 packaging facility operational

2028 and Beyond: Substantial Relief Potential

  • 2028 or later: Micron New York fabs begin production
  • 2028-2030: Full ramp of multiple new facilities

Market Signals and Risk Factors

Current Market Indicators:

  • DRAM contract prices increased 171.8% year-over-year as of Q3 2025, with spot prices nearly tripling from Q4 2024 to Q4 2025
  • DRAM spot prices reached $24.55 in late 2025, well above typical peak levels of $8-10
  • Every NAND manufacturer has reported 2026 capacity sold out
  • Server DRAM prices on track to double by late 2026

Potential Acceleration Factors:

  • CXMT expansion could provide significant additional capacity if geopolitical constraints ease
  • Technology transitions to advanced nodes could increase effective capacity through better yields

Potential Delay Factors:

  • Lengthening lead times for cleanroom build-out across geographies
  • Continued geopolitical tensions affecting Chinese capacity additions
  • AI demand growth potentially outpacing new capacity additions

Sources:

  • https://www.tomshardware.com/pc-components/dram/dram-prices-surge-171-percent-year-over-year-ai-demand-drives-a-higher-yoy-price-increase-than-gold
  • https://www.tomshardware.com/pc-components/ssds/phison-ceo-confirms-nand-prices-have-more-than-doubled-and-will-continue-to-rise-all-2026-production-already-sold-out-ssds-facing-pricing-apocalypse-throughout-2027
  • https://www.bacloud.com/en/blog/230/when-will-ram-prices-drop-global-memory-market-outlook-20242026.html
  • https://www.bacloud.com/en/blog/237/when-will-server-ram-prices-drop-december-2025-update—optimistic-and-pessimistic-scenarios.html
  • https://counterpointresearch.com/en/insights/advanced-memory-prices-likely-to-double-as-dram-crunch-spreads-on-nvidia-pivot-structural-factors
  • https://www.trendforce.com/news/2025/12/18/news-micron-reveals-three-culprits-behind-memory-crunch-and-why-it-wont-ease-soon/

Implications for Procurement and Planning

The data suggests 2027 represents the earliest realistic timeframe for meaningful supply relief, with more substantial improvement likely by 2028-2029. Industry analysts expect DRAM price peaks in 2026, with settlement beginning in 2027 before rising again in 2028.

Key procurement considerations:

  • Immediate term (2025-2026): Supply will remain substantially short of demand
  • Mid-term (2027-2028): New fab production should provide gradual relief
  • Long-term (2029+): Capacity additions may finally match demand growth if AI infrastructure buildout moderates

The leading indicators point to a sustained supply constraint period lasting through 2026, with gradual improvement beginning in 2027 as new fabrication capacity comes online. However, the ultimate timeline depends on whether AI demand growth can be satisfied by the substantial capacity investments now underway.

DRAM Production Capacity by Manufacturer (Thousands of Wafers per Month)

YearMicron TechnologySamsung ElectronicsSK HynixCXMTNotes
2022~280*~420*~350*~80*Post-pandemic capacity adjustments
2023~270*~400*~340*120Industry production cuts to stabilize pricing
2024~280*~410*~360*160-200AI demand begins driving capacity reallocation
2025~290*~430*170 (HBM focus)240-280*Major shift to HBM production, supply shortage intensifies
2026~320*200 (1c DRAM) + ~250* (other)~200* (total)300*New fabs coming online, but AI demand still exceeds supply
2027350* (Idaho online)~480*250* (Yongin fab)350*First meaningful capacity relief begins
2028420* (NY fabs start)~520*~280*400*Substantial new capacity additions
2029~480*~550*~320*450*Expected AI infrastructure buildout moderates

Key Notes by Year:

  • 2022: Industry operating at reduced utilization following pandemic demand normalization
  • 2023: Strategic production cuts implemented across all major manufacturers
  • 2024: CXMT doubled production from 100k to 200k wafers/month; AI demand surge begins
  • 2025: SK Hynix targeting 170k wafers/month HBM capacity; supply shortage peaks
  • 2026: Samsung targeting 200k wafers/month 1c DRAM; CXMT projected 300k wafers/month
  • 2027: Micron Idaho fab begins production; SK Hynix Yongin fab completed
  • 2028: Micron New York fabs start production; supply-demand balance begins to normalize
  • 2029: Expected AI infrastructure buildout moderates; industry reaches new equilibrium*

Important Caveats:

  • Capacity Mix: These numbers represent total DRAM wafer capacity. Actual available capacity for consumer DDR4/DDR5 is significantly lower due to HBM allocation.
  • HBM Trade Ratio: HBM requires a 3-to-1 trade ratio with DDR5, meaning every HBM wafer reduces DDR5 supply by 3 wafers worth of capacity.
  • CXMT Constraints: U.S. export controls and 20% tool localization rate may limit actual expansion compared to projections.
  • Technology Transitions: Newer process nodes (1c, 1b, 1γ) provide more capacity per wafer, but transition periods can temporarily reduce output.

*Asterisk indicates estimated values based on industry trends, company guidance, and capacity expansion announcements.

Wrapping Up

With AI infrastructure eating up three times the wafer capacity per gigabyte compared to regular DRAM, and manufacturers chasing the higher margins from HBM, consumer and enterprise markets are fighting for scraps.

The data points to 2027 as when we may start seeing real relief. That’s when Micron’s Idaho fab, SK Hynix’s Yongin facility, and new packaging capabilities come online. But getting back to normal pricing and availability? That’s more like 2028-2029, and only if AI infrastructure spending slows down as expected.

For anyone planning hardware purchases, expect tight supply and high prices through 2026. Things should start improving in 2027, but slowly.

Important Note

This is a work in progress. The numbers and timelines here reflect our current understanding based on public information from manufacturers, analysts, and market research. We’ll keep updating this as things develop and we get more clarity on:

  • When fabs actually come online and ramp production
  • How AI demand actually plays out
  • What happens with Chinese manufacturing capacity
  • How new process technologies perform in practice

Use these production estimates to get a sense of the overall market picture. They’re helpful for planning, but actual results will vary based on market conditions, technology developments, and global economic factors. Real-world outcomes rarely match projections perfectly.

Check the manufacturer investor relations pages and analyst reports linked throughout for the latest updates.


Chat With Our Team

We’re available to answer questions and provide information.

Reach Out

Schedule a Consultation

Get a deeper assessment and discuss your unique requirements.

Schedule Consultation

Try It Out

Take a peek under the hood of our cloud platform or launch a trial.

Trial Options

 

 

 Read More on the OpenMetal Blog

Leading Indicators of When RAM and NVMe Supply Will Match Demand

Feb 06, 2026

OpenMetal’s procurement data shows NVMe costs up 223% and DDR5 up 474% due to AI infrastructure demand. This analysis tracks leading indicators from Micron, Samsung, SK Hynix, and CXMT to project when supply relief begins in 2027, with substantial improvement expected by 2028-2029.

How to Build Multi-Region Infrastructure Across Three Continents

Feb 05, 2026

Complete guide to multi-region infrastructure across three continents. OpenMetal’s Los Angeles, Ashburn, Amsterdam, and Singapore locations enable disaster recovery, global performance, and data sovereignty compliance for 70% less than hyperscaler costs.

Why Singapore Outperforms Tokyo and Sydney for APAC Infrastructure

Feb 03, 2026

Companies expanding into Asia-Pacific choose Singapore for its central location providing 15-30ms latency to SEA’s major cities, infrastructure costs 50% below Tokyo, and generous bandwidth allocations. This article covers 10 ideal Singapore data center use cases from gaming to fintech with OpenMetal bare metal and Cloud Core pricing.

The 90-Day Cloud Cost Review Process for Engineering, FinOps, and Finance Teams

Feb 02, 2026

Stop treating cloud cost optimization as a one-time project. This 90-day operating rhythm gives Eng, FinOps, and Finance teams a repeating cadence — monthly variance reviews with top-3 actions, quarterly contract renegotiations, workload right-placement, and cost guardrails — that cuts waste without slowing releases.

Comparing Costs of Reserved Instances vs Bare Metal

Jan 30, 2026

AWS Reserved Instances offer 30-40% discounts through 1-3 year commitments, but the “savings” come with hidden costs: egress fees, support charges, and modification limitations. Bare metal infrastructure provides fixed monthly pricing with included bandwidth, support, and flexibility. We compare real configurations to show when each model makes sense.

When Underutilized VMs Don’t Actually Cost You More

Jan 29, 2026

Public cloud pricing creates constant pressure to optimize VM utilization, turning DevOps teams into full-time cost managers. But underutilization only wastes money when you’re paying per instance. With fixed-cost bare metal infrastructure, that idle capacity becomes operational headroom you’ve already paid for.

High-Bandwidth Use Cases Now Cost-Effective on Private Cloud

Jan 27, 2026

Ten bandwidth-intensive use cases with real cost comparisons. Video streaming, email infrastructure, game distribution, AI inference, and CDN workloads save millions annually on private cloud vs AWS per-GB egress pricing.

How to Calculate Total Cost of Ownership for Hosted Private Clouds

Jan 23, 2026

Learn to calculate hosted private cloud TCO with step-by-step methodology and real pricing data. Covers hidden costs like staff time, egress fees, and downtime. Real-world examples compare OpenMetal to AWS (70% savings) and on-premises (51% savings) over 5 years with break-even analysis.

Infrastructure Cost Audits: The Red Flags That Repeat Across SaaS Portfolios

Jan 16, 2026

Infrastructure cost audits uncover the same hidden risks across SaaS portfolios: spend volatility, networking blind spots, AI inference drift, and tool sprawl. This Runway Intelligence briefing shows how operating partners and VCs use audits to protect margins, runway, and valuation.

FinOps for AI Gets Easier with Fixed Monthly Infrastructure Costs

Dec 15, 2025

AI workload costs hit $85,521 monthly in 2025, up 36% year-over-year, while 94% of IT leaders struggle with cost optimization. Variable hyperscaler billing creates 30-40% monthly swings that make financial planning impossible. Fixed-cost infrastructure with dedicated GPUs eliminates this volatility.