NAND Goes Vertical: Sandisk’s AI-Fueled Earnings Shock Ignites a New Memory Supercycle

Sandisk's quarter was the kind of print that makes “beat and raise” feel like an understatement and makes the stock chart look like it was drawn by someone who just discovered leverage. In fiscal Q2 (December quarter), SNDK posted adjusted EPS of $6.20 versus roughly $3.50–$3.60 expected, on revenue of $3.025B versus about $2.64B–$2.69B expected. The core message was simple: NAND is tight, pricing is firming fast, product mix is improving, and AI-linked demand is pulling forward a multi-year cycle. The market’s reaction was equally simple: shares ripped more than 120 points (roughly 20%) as investors were forced to recalibrate what “normalized” earnings power looks like in an undersupplied NAND environment.

The biggest headline inside the numbers was profitability. Non-GAAP gross margin expanded to 51.1%, dramatically above the prior quarter and well above expectations, with management attributing the upside primarily to higher pricing and unit-cost improvements. That matters because NAND cycles are typically judged by margin inflection: once you see sustained pricing traction plus cost downs, the operating leverage can become nonlinear. SNDK also called out startup costs (roughly $24M), implying underlying margin power was even a touch better than the headline. On top of that, operating income (about $1.13B) crushed expectations (roughly $0.63B), reinforcing that this wasn’t a “one line got lucky” quarter—it was broad-based execution with a margin structure that’s suddenly acting like it belongs in a different industry.

On NAND pricing, the company didn’t just imply conditions improved—it effectively told you the market is being repriced. The quarter beat guidance on revenue and margins largely because realized pricing moved higher across segments, and management framed the backdrop as demand exceeding supply. The implication is important: if pricing is doing the heavy lifting now, the durability of the move hinges on whether supply remains constrained and whether customer behavior shifts from short-cycle bargaining to longer-cycle planning. Sandisk leaned hard into that second point, emphasizing that customers are increasingly moving away from quarter-to-quarter transactional discussions and toward multi-year supply agreements, in some cases with prepayment features. That’s a big deal for a NAND vendor: it can reduce the amplitude of the historical boom-bust cycle and give management more confidence not to overbuild into what’s currently a very profitable tape.

The demand commentary was equally direct and very “2026-coded”: AI and data centers are changing the storage math. Sandisk characterized NAND as becoming more “indispensable” as AI workloads scale, especially in enterprise SSD deployments where performance bottlenecks are expensive and data movement is relentless. Data center revenue was a smaller portion of total revenue in the quarter (about $440M), but it grew sharply sequentially and was framed as the demand engine that changes the planning horizon. Management also pointed to increasing NAND content per deployment as inference scales and as architectures evolve. They highlighted ongoing qualification progress for PCIe Gen 5 enterprise TLC products with hyperscalers and flagged that higher-capacity QLC “storage-class” products are advancing in qualification as well—important because QLC adoption can unlock very large bit demand if performance and endurance meet hyperscaler requirements.

Then came the part that lit the match under the stock: the outlook. Sandisk guided fiscal Q3 revenue to $4.4B–$4.8B versus consensus around $2.9B, and guided EPS to $12–$14 versus consensus in the $4–$5 range. Even more eye-popping, they guided non-GAAP gross margin to 65%–67%. That’s the kind of margin guide that forces the market to stop debating “is the cycle real” and start debating “how long can this last.” Management also said the market should be “more undersupplied” in Q3 than it was in Q2, while bits are expected to be down mid-single digits sequentially—an important nuance that reinforces the pricing/mix narrative: they’re not flooding the channel; they’re monetizing scarcity.

On execution and supply visibility, the Kioxia joint venture extension through 2034 added strategic weight to the story. The $1.165B manufacturing-services commitment (paid across 2026–2029) effectively buys longer runway and tighter alignment on advanced 3D NAND production capacity at a time when supply is the currency of the realm. It also helps explain why investors are getting more comfortable underwriting multi-year strength: the company is positioning itself to secure manufacturing support while keeping capex discipline intact. Management reiterated it is not changing capital spending plans and continues to target mid- to high-teens bit growth through its BiCS8 transition—signaling they’re trying very hard not to be the person who restarts the boom-bust cycle by “getting excited” with a shovel.

So why the +120-point move? First, the magnitude and quality of the beat: EPS and margins weren’t just above consensus; they were in a different zip code. Second, the guide reset the entire Street model—revenue +50% versus consensus at the midpoint and EPS roughly 3x consensus forces analysts and PMs to chase the numbers higher. Third, the call’s structural framing (multi-year agreements, prepayments, undersupply beyond 2026) supports a valuation re-rate because it argues for a longer and less volatile profit cycle. And finally, the narrative tailwinds are crowded but powerful: AI infrastructure spend is still accelerating, NAND is a direct beneficiary, and supply additions are limited near term—meaning the “pricing umbrella” can stay open longer than skeptics expected.

What to watch from here is less about whether demand is strong (it is) and more about whether Sandisk can execute without stepping on the classic NAND landmines. Key items: 1) sustainability of gross margins as volumes scale and as any incremental supply comes online; 2) evidence that long-term agreements are real and repeatable (how much of the book becomes contracted, whether prepayments expand, and whether pricing frameworks reduce quarterly volatility); 3) product-mix progress in enterprise SSD, especially hyperscaler qualifications and the ramp of next-gen TLC/QLC platforms; 4) capex discipline—staying rational even as pricing screams “expand”; and 5) competitive behavior across the NAND ecosystem. If those boxes get checked, the stock can keep working even after a move that makes “up a lot” look like a rounding error.

Bottom line: Sandisk didn’t just report a strong quarter—it provided a roadmap for why this NAND cycle may be structurally different, and the market repriced the stock accordingly. Now the company’s job is simple in theory and hard in practice: keep supply tight, keep mix improving, keep customers committed, and don’t get tempted into rebuilding the very cycle it’s currently enjoying.