#BernsteinSaysMemoryBullMarketToLastUntil2027

The Memory Bull Market: Why We're Only Halfway Through the Cycle

Bernstein just dropped a reality check on the semiconductor world.

The memory bull market isn't dying—it's maturing. According to their latest monthly report, this cycle has legs through 2027. But here's the kicker: the explosive price surge phase? That's behind us.

Q2 was absolutely bonkers. DRAM prices ripped 74% quarter-over-quarter. Server DRAM climbed over 60%, while mobile DRAM surged nearly 80%. NAND told a more nuanced story—wafer spot prices softened, but contract prices for mobile and SSD storage still managed a 60% lift.

The Q3 slowdown isn't a death sentence.

Bernstein expects DRAM growth to cool to 13-18% in Q3 as consumer electronics demand weakens. This isn't the cycle ending. It's the cycle normalizing. The wild price action was never sustainable. What comes next is steadier, more rational growth driven by something far more durable than panic buying: AI cloud infrastructure.

The real structural driver isn't your iPhone upgrade cycle or laptop refresh. It's the hyperscalers. Amazon, Microsoft, Google, Meta—these companies are signing long-term supply agreements that extend visibility deep into 2027. They're not buying on spot. They're committing capital to secure memory supply for AI training and inference workloads that are only accelerating.

This is a fundamentally different demand profile than previous cycles. Traditional memory was tied to consumer hardware refresh rates—volatile, unpredictable, seasonal. AI data center demand is capex-heavy, multi-year, and structurally expanding.

The winners and losers are already sorting themselves out.

Bernstein's ratings tell the story. Samsung, SK Hynix, Micron, and SanDisk all carry positive outlooks. Kioxia? Cautious. The divergence makes sense. The leaders have secured HBM (High Bandwidth Memory) supply agreements with NVIDIA and AMD. They're vertically integrated. They control advanced packaging capacity.

SK Hynix is the standout here. They overtook Samsung in DRAM market share in Q1 2025—first time ever—purely on the back of HBM dominance. When you're shipping HBM3E to NVIDIA for Blackwell GPUs, you don't need to win the commodity DRAM game.

The 2027-2028 reset is already priced into planning.

Smart money knows this cycle ends. New capacity comes online. Long-term contracts get fulfilled. Supply catches up. Bernstein's timeline—prices gradually normalizing from H2 2027 through 2028—isn't bearish. It's realistic.

If you're playing memory stocks, you're not betting on perpetual price increases. You're betting on earnings duration. The companies that locked in high-margin HBM supply agreements in 2024-2025 will print cash through 2027 even as spot prices soften. Their forward margins are already contracted.

The commodity DRAM and NAND players without HBM exposure? They're riding the spot wave while it lasts. When it breaks, they break harder.

This isn't 2018. The demand driver isn't smartphones peaking. It's AI infrastructure building out. That takes longer, costs more, and creates stickier supplier relationships. The memory cycle has evolved from a consumer hardware play to an enterprise infrastructure play.

The bull market lives. But the character has changed. Position accordingly.