Bernstein Forecasts the Memory Bull Market Through 2027: AI Demand Continues to Reshape the Semiconductor Industry

Market Overview

The global memory semiconductor cycle may be far from over.

According to a landmark research report from Bernstein, the current memory bull market is expected to continue through 2027, supported by an unprecedented supply-demand imbalance driven by accelerating artificial intelligence infrastructure investment.

Rather than viewing the current cycle as a temporary recovery, Bernstein argues that structural AI demand is creating a fundamentally different environment from previous semiconductor upcycles.

AI Infrastructure Is Changing Memory Demand

The primary driver behind Bernstein's outlook is the rapid expansion of AI computing infrastructure.

A single Nvidia Vera Rubin NVL72 rack, now being deployed by major hyperscale cloud providers including Microsoft, Google, and Amazon, contains approximately 20.7 terabytes of HBM4 memory alongside 54 terabytes of LPDDR5X memory.

At current market prices, the memory components for one rack already exceed $2 million.

Based on Bernstein's 2027 pricing projections, that cost could nearly double, representing an estimated 435% increase in memory cost per rack within a single product generation.

Memory Prices Continue Rising

Recent pricing trends continue supporting Bernstein's bullish outlook.

The firm's Q2 2026 industry tracking indicates that both DRAM and NAND contract prices increased more than previously expected.

Enterprise SSD demand and stronger mobile NAND consumption are expected to push NAND contract prices up approximately 65–70% quarter over quarter.

Meanwhile, sustained demand for server DRAM and enterprise storage continues keeping supply constrained across multiple product categories.

HBM Pricing Is Entering a New Phase

One of the industry's most significant developments involves High-Bandwidth Memory (HBM).

Because HBM pricing has traditionally been negotiated through annual contracts, its prices have remained relatively stable while conventional DRAM prices have increased approximately 4.5 times since Q3 2025.

As a result, semiconductor manufacturers currently generate roughly twice the revenue and nearly three times the gross profit from wafer capacity allocated to standard DRAM compared with HBM production.

Both Samsung and Micron confirmed during their Q1 2026 earnings calls that non-HBM DRAM has become more profitable than HBM, with the profitability gap continuing to widen.

Bernstein Expects Significant HBM Price Increases

To restore profitability, Bernstein expects substantial pricing adjustments.

The firm forecasts that average HBM prices could increase between two and two-and-a-half times during 2027.

Even after those increases, HBM profitability is expected to remain below traditional DRAM, although the difference would narrow considerably.

SK Hynix has also indicated that it intends to pursue an optimal allocation strategy between HBM and conventional DRAM rather than simply maximizing short-term revenue, balancing profitability with the long-term expansion of AI infrastructure.

Financial Outlook for Memory Leaders

Bernstein's long-term financial projections are equally notable.

The firm forecasts exceptionally strong 2027 Return on Equity (ROE) levels across the industry's leading manufacturers:

- Samsung: 55%
- SK Hynix: 108%
- Micron: 85%

Cash generation is also expected to accelerate rapidly, with Bernstein estimating that cash balances could account for approximately 70–80% of book value at these companies by 2027.

Given these projections, the research firm believes traditional price-to-book valuation models are becoming less useful and increasingly favors forward price-to-earnings metrics.

Higher Earnings Expectations

Bernstein has raised price targets for all three major memory manufacturers.

Its projected 2027 earnings per share remain well above current market expectations:

- Samsung: 26% above consensus
- SK Hynix: 32% above consensus
- Micron: 38% above consensus

As annual HBM pricing negotiations conclude over the coming months, Bernstein expects broader analyst consensus estimates to move higher, potentially providing additional support for semiconductor valuations.

Rising Costs for AI Infrastructure

The report also highlights the growing financial burden facing hyperscale cloud providers.

Bernstein estimates that data center capital expenditure may need to increase by approximately 30% simply to absorb higher memory costs.

To preserve its targeted 75% gross margin, Nvidia is expected to pass a significant portion of rising HBM costs through to cloud infrastructure providers, further increasing the total cost of AI deployments.

Final Outlook

Bernstein's analysis suggests the current memory cycle represents more than a traditional semiconductor recovery.

Instead, artificial intelligence infrastructure is creating sustained structural demand that existing manufacturing capacity cannot quickly satisfy.

If these projections prove accurate, the memory bull market could remain one of the strongest long-term investment themes through 2027, with continued pricing strength, expanding profitability, and rising AI infrastructure spending supporting the industry's next phase of growth.

#BernsteinSaysMemoryBullMarketToLastUntil2027
@Gate_Square