#BernsteinSaysMemoryBullMarketToLastUntil2027


The Memory Bull Market: Why the Party Isn't Over—Even If the Music Has Changed

Bernstein's latest monthly storage report just dropped a reality check on the semiconductor world: this memory bull market isn't going anywhere until 2027. But here's the twist—the steepest price climbs are already in the rearview mirror.

If you've been watching DRAM and NAND prices rocket upward over the past eighteen months, you already know the story. What started as a cyclical recovery has morphed into something far more structural, driven by an insatiable appetite from AI hyperscalers who are essentially hoovering up every wafer they can get their hands on.

The Numbers That Matter

Let's cut through the noise with the hard data. In Q2 2026, traditional DRAM prices surged approximately 74% quarter-over-quarter. Break that down and you see server DRAM climbing 60–67% while mobile DRAM nearly touched 80%. These aren't gentle adjustments—they're seismic shifts that have rewritten the economics of the entire electronics supply chain.

But the report signals a deceleration ahead. Q3 DRAM growth is expected to moderate to 13–18% as consumer electronics demand shows cracks. Smartphones and PCs are feeling the pinch as memory costs eat deeper into bill-of-materials budgets. Entry-level devices are getting squeezed hardest—some analysts forecast a 22% decline in sub-$400 smartphone shipments this year as OEMs struggle to absorb these costs.

The Great Divergence: NAND's Split Personality

NAND tells a more nuanced story. While wafer prices are softening—a sign that raw supply constraints are easing—mobile and SSD contract prices remain elevated, up roughly 60%. This divergence reveals the market's bifurcation: commodity NAND is finding some breathing room, but specialized high-performance segments tied to AI infrastructure remain stubbornly tight.

Enterprise SSDs, in particular, are seeing "dry-year" conditions with module makers running inventories down to weeks rather than months. When Micron quotes jump 50% in a single month, you know this isn't normal cyclical behavior.

The Real Engine: AI's Long-Term Hunger

Strip away the quarterly noise and one factor dominates: long-term orders from AI cloud providers. These aren't speculative purchases. Amazon, Microsoft, Google, and Meta are locking in multi-year supply agreements that effectively remove massive chunks of memory capacity from the spot market.

SK Hynix, which captured 57% of HBM sales in Q3 2025, expects the AI memory market to grow 30% annually through 2030. When your customers are committing $22 billion just to secure future supply—as Micron recently disclosed—you're no longer in a typical semiconductor cycle. You're in a structural scarcity regime.

High-bandwidth memory (HBM) is the critical bottleneck here. It takes advanced packaging, tighter qualification, and yield execution that lock in the best-positioned suppliers. And the buyers? They care more about guaranteed supply than spot price. That dynamic gives memory suppliers genuine pricing power for the first time in years.

The Street's Take: Who Wins, Who Doesn't

Bernstein isn't shy about its picks. The firm maintains positive ratings on Samsung, SK Hynix, Micron, and SanDisk—essentially the entire Western and Korean memory complex. SanDisk, in particular, has been a standout, with Bernstein tripling its price target to $3,000 and highlighting the company's contract protection at $0.29 per gigabyte.

The outlier? Kioxia. Bernstein remains cautious here, suggesting the Japanese memory maker is facing competitive pressures that its peers have managed to navigate more effectively.

Dan Ives at Wedbush has coined these memory leaders the "golden jewels" of the AI revolution—a characterization that captures their transformation from cyclical commodity plays to essential infrastructure assets.

Here's where it gets interesting. Bernstein models memory prices gradually peaking and normalizing from H2 2027 into 2028 as new capacity comes online and long-term contracts get executed. That doesn't mean a crash—it means a return to a more balanced market where supply and demand find equilibrium.

For investors, this creates a window. The memory trade has already delivered extraordinary returns—Micron is up roughly 900% from its April 2025 lows, SK Hynix has surged alongside it. But if Bernstein's timeline holds, there's still runway before the cycle turns.

The risk? Consumer demand weakness could spread. If smartphone and PC shipments deteriorate faster than expected, memory makers might face a double bind: AI demand stays strong but can't fully offset consumer softness. We're not there yet—the report suggests server demand remains robust—but it's the scenario keeping analysts up at night.

This memory bull market has already outlasted most cyclical predictions. Bernstein's call for extension through 2027, even with moderating price gains, reflects a fundamental reassessment of how AI infrastructure demand reshapes semiconductor economics.

The companies sitting at the center of this—Samsung, SK Hynix, Micron—have transformed from cyclical commodity producers into something closer to toll collectors on the AI highway. Their products are no longer interchangeable; they're mission-critical components with structural supply constraints.

For anyone building positions in this space, the message is clear: the easy money from the initial price surge may be behind us, but the structural bull case remains intact. The memory market isn't going back to 2024 pricing anytime soon. As one Lenovo executive bluntly put it: memory prices "will never be like last year again."

The question isn't whether the bull market ends—it's when, and whether you're positioned to ride the remaining wave.
post-image
post-image
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned