Citi Raises Memory Price Forecasts Sharply as AI Infrastructure Demand Accelerates

The global memory market may be entering a far more aggressive upcycle than previously expected.

According to recent projections from Citigroup, average DRAM prices could surge by as much as 200% in 2026, while NAND flash prices may rise approximately 186% year-over-year. The revised outlook reflects growing expectations that AI infrastructure expansion will continue driving unprecedented demand across the entire memory and storage supply chain.

Among the biggest beneficiaries is SK hynix, which Citi views as one of the strongest positioned companies in the current cycle due to its leadership in High Bandwidth Memory (HBM).

HBM Becomes the Core Growth Driver

HBM remains the most important variable in the AI server market.

Citi expects HBM pricing to strengthen significantly during the second half of 2026, particularly in Q4, when average selling prices could increase by roughly 30% quarter-over-quarter. Strong demand for HBM3e, combined with the upcoming transition toward HBM4, is expected to further improve product mix and profitability for leading suppliers.

The continued expansion of AI infrastructure investments by companies such as Anthropic is reinforcing expectations that demand for high-performance memory will remain elevated for the foreseeable future.

As a result, Citi reportedly raised its target price for SK hynix from KRW 1.7 million to KRW 3.1 million, representing an increase of approximately 82%.

Server DDR5 Prices Expected to Surge

Beyond HBM, Citi also significantly increased its expectations for server DDR5 memory pricing.

The firm now forecasts server DDR5 DRAM average selling prices to rise approximately 329% year-over-year in 2026, driven largely by accelerating AI inference deployments and rising memory density requirements inside modern AI clusters.

One of the most notable projections involves 64GB DDR5 RDIMM modules. Citi estimates pricing could increase from approximately $873 in Q1 2026 to nearly $1,586 by Q4 2026.

The report suggests that memory is no longer simply a supporting component inside AI servers — it is becoming a strategic infrastructure resource.

SSD Pricing Also Enters a Strong Upcycle

Enterprise SSD pricing is also expected to move sharply higher.

AI workloads such as KV Cache, vector databases, retrieval-augmented generation (RAG), and large-scale inference systems are dramatically increasing demand for high-capacity, high-performance storage solutions.

Citi forecasts average SSD prices could rise approximately 267% year-over-year in 2026 as hyperscale and AI-focused data centers continue expanding storage-intensive architectures.

AI Infrastructure Costs Expanding Beyond GPUs

The broader implication is that AI infrastructure inflation is no longer limited to GPUs.

HBM, DDR5 RDIMM, SoCAMM2, enterprise SSDs, and NAND flash are increasingly moving together within the same pricing cycle. For cloud providers, AI server OEMs, and data center operators, procurement strategies for memory and storage may soon become just as critical as GPU allocation itself.

The market is beginning to shift toward a new phase of AI infrastructure investment — one increasingly driven by memory bandwidth, storage capacity, and supply chain control.

Industry Outlook

If current forecasts prove accurate, 2026 could become one of the most significant pricing cycles in memory market history.

For enterprise infrastructure buyers, this may accelerate:

• Long-term supply agreements
• Advance inventory planning
• Quarterly price-lock negotiations
• Strategic sourcing for AI server components

As AI deployments continue scaling globally, the competition for next-generation memory and storage resources is expected to intensify across the data center ecosystem.