Let’s start with a number that probably deserves a second read.

Q3 2026 revenue: $5.95 billion — up 251% year over year, with GAAP net income of $3.62 billion and non-GAAP diluted EPS of $23.41. The Street consensus was sitting around $14.66. That’s a 59.69% earnings surprise. Not a beat. A blowout.

And the stock went up after hours. Shares surged 4.89% in aftermarket trading, reaching $1,116.25 — above the previous 52-week high. Which tells you something different than the sell-the-news reaction some expected given how much the stock had already run.

This is a company that, not long ago, was hemorrhaging money. In Q3 FY25, SanDisk took a $1.83 billion goodwill impairment, with shares trading around $35.60. Tonight it printed $23.41 in earnings per share. Twelve months. Same company. Different universe.

The Margin Story Is the Real Story

Revenue is one thing. Margins are another. And this is where tonight’s report gets genuinely strange — strange in the best possible way if you’re long the stock.

GAAP gross margin expanded to 78.4% from 22.5% a year earlier, driven by datacenter revenue up 645% year over year and higher pricing. That’s not incremental improvement. That’s a different business model operating at a different altitude entirely.

Here’s the part worth sitting with: SanDisk reported 78.4% gross margins this quarter, while Wall Street expects NVIDIA to report 75.1% gross margins when it reports its Fiscal Q1 earnings. A NAND flash memory company — historically one of the most commoditized, cyclical businesses in all of semiconductors — just printed higher margins than NVIDIA. That sentence shouldn’t make sense. And yet.

The guidance for Q4 keeps going. For fiscal Q4 2026, SanDisk guides revenue to $7.75–$8.25 billion and non-GAAP diluted EPS to $30.00–$33.00, with non-GAAP gross margin anticipated around 79–81%. Margins are expected to expand further. From already historic levels.

What’s Actually Driving This

The answer is datacenter. Full stop.

The data center segment emerged as the fastest-growing market, with a 233% sequential revenue increase. Not year over year — sequential. One quarter to the next. The AI infrastructure buildout has created a supply-demand gap in enterprise SSDs that SanDisk is sitting directly inside of, and the company has been deliberately steering mix toward those higher-value customers.

On the earnings call, CEO David Goeckeler was direct about where things are heading. Before what the company saw this week, they would raise the calendar year 2026 datacenter growth number to the mid-70s from the 60s just three months ago — which itself was up from the 40s three months before that and the 20s three months before that. That’s not a revised estimate. That’s a demand curve that is moving faster than anyone’s models can track.

Slight tangent, but it matters: NAND flash was never supposed to be a beneficiary of AI at this scale. The original AI infrastructure thesis was GPU compute, then HBM memory, then networking. NAND was a downstream, lower-priority item. What’s changed is inference — the deployment phase of AI that requires massive, fast, cheap storage to keep models accessible in real time. NAND flash is emerging as the only economically viable solution to deliver the capacity, performance, and efficiency required to keep models accessible for real-time inference at scale. That repositioning is what explains 78% gross margins in a business that used to operate in the 30s.

The New Business Model — This Is What’s Different

The thing that separates this cycle from prior NAND upcycles is the contract structure. SanDisk was engaged in discussions with customers on multiyear supply partnerships — what they refer to as new business models. Five multiyear partnerships have now been signed, structured to lock in committed supply for customers and committed financials for SanDisk.

That’s a fundamental change to the cyclicality argument. Memory has always been boom-bust because no one was locked in. Prices spike, supply floods in, prices crater. That’s the NAND cycle going back thirty years. Multi-year take-or-pay structures with hyperscalers don’t eliminate that risk entirely — but they substantially change the downside profile. The bear case just got more complicated.

Long-term debt was reduced to zero while cash and cash equivalents rose to $3.74 billion, and the Board approved a $6 billion share repurchase program expected to be funded by operating cash flows. In Q3 2026, SanDisk generated $3.04 billion of cash from operating activities, and after $45 million of capital expenditures, free cash flow was $2.99 billion. The buyback isn’t a gesture. It’s a statement that management believes this earnings power is durable enough to return capital at scale while still funding the business.

Where does the stock go from here? That’s the question everyone is asking and nobody really knows the answer to. The stock is up over 3,100% in the past year and over 275% just in 2026. Those are parabolic numbers. The multiple has to be justified by forward earnings, and forward earnings are being revised upward in real time — but at some point the math on valuation has to reckon with itself.

The honest answer is that SanDisk has become a genuinely hard stock to model because the results keep moving faster than the models. Every quarter the Street rebuilds the numbers, and every quarter the actual results land above the new ceiling. CEO Goeckeler called this quarter