Sandisk Puts a Number on the NAND Shortage, and the Memory Tape Reprices Again

by Karl Lindström · August 14, 2026

Semiconductors & Hardware Reporter, Frontier Tech Wire

The memory complex did most of the heavy lifting in Thursday's session. Sandisk (Nasdaq: SNDK) closed up 13.7%, at $1,528.11, and Micron Technology (Nasdaq: MU) closed up 4.2%; the percentage moves are the ones carried in the market wrap published by Yahoo Finance at 4:07 p.m. Eastern on Aug. 13. Those are closing figures. The four-figure share price is not a misprint — Sandisk earned $43.97 per diluted share on a GAAP basis in the June quarter alone, and every per-share figure in this article is on that same, unsplit basis. Several intraday reports circulating through the day put Sandisk as high as roughly 15% to 16.7% and Micron near 6%, and those were snapshots taken while the tape was still moving — a distinction worth keeping straight before anyone builds a thesis on the larger number.

The catalyst was Sandisk's investor day. Investing.com, reporting at 11:29 a.m. Eastern, said management delivered a set of long-term financial targets that ran ahead of what analysts had modelled: revenue growth in the mid-to-high teens annually across fiscal 2028 through fiscal 2030, non-GAAP gross margins of approximately 80%, and non-GAAP operating margins near 75%. Every one of those is a company target on a non-GAAP basis, and none of them is guidance for the current fiscal year.

The nearer-term number matters more, and it did not come from Thursday's presentation. With its fiscal fourth-quarter results on Aug. 5, Sandisk guided fiscal first-quarter revenue to $10.3 billion to $10.8 billion with non-GAAP gross margin of 83% to 85%, the midpoint sitting roughly 60 basis points below the record 84.6% posted in the June quarter. That is the shape of a business at or very near a cyclical peak on margin, still growing volume. Memory has run this movie before in both directions.

Sandisk's chief financial officer, Luis Visoso, has said the total NAND market should pass $300 billion in 2026 and $500 billion in 2027, and that supply stays tight into 2028 — comments made before Thursday and carried into the investor-day framing, not fresh disclosure from the stage, per FinanceFeeds and Investing.com, which describes the market-size remarks as previously stated. That is the load-bearing claim in the entire rally. Strip the 2028 tightness assumption out and the 80% gross margin target through fiscal 2030 does not survive contact with normal industry behaviour, in which shortage pricing pulls in capital expenditure that eventually becomes oversupply.

The company's answer to that objection is contract coverage. Management has pointed to a book of long-term customer supply agreements — Sandisk calls them New Business Model contracts — carrying roughly $93.9 billion of minimum contracted revenue calculated at floor pricing. The scope matters: that figure is cumulative across the entire book, not annual. The agreements run up to five years with a weighted average duration of more than four, so the $93.9 billion is spread across a multi-year period rather than landing in any single fiscal year. It was disclosed in early August, around the Aug. 5 fiscal fourth-quarter results, per summaries published by ad-hoc-news and TrendForce, and coverage is described as more than half of expected fiscal 2027 bit shipments and roughly two-thirds of fiscal 2028. Sources differ on how many separate contracts and customers make up the book — one account describes more than ten agreements with eight strategic customers, another eight contracts with six — so Frontier Tech Wire is reporting the dollar figure, the term and the coverage claim, and not the contract count.

It is also worth being precise about what that number is and is not. Per TrendForce, the contracts pair minimum volume guarantees with floor pricing, and the variable component is structured with floor and ceiling mechanisms, with management indicating the agreements are being signed at around an 80% margin level and retaining upside if market prices rise. So $93.9 billion is a contracted minimum, not a forecast. What it is not is revenue that has been recognised, or a guarantee of collection: a floor only binds for as long as the counterparties perform, and the arithmetic assumes the full multi-year term runs its course.

On the product side, Sandisk and Kioxia announced on Aug. 12 a ninth-generation, high-performance 2-terabit QLC 3D flash technology aimed at AI and data-intensive workloads. Per the companies' release and Benzinga's account of it, the part uses a CMOS directly bonded to array architecture with a six-plane design and a 4.8 gigabit-per-second NAND interface, a roughly 33% speed improvement over the prior generation. At the investor day Sandisk presented BiCS10, its tenth-generation NAND platform, claiming a 59% improvement in bit density alongside interface speeds up to 4.8 gigabits per second. The start of tenth-generation production at the Kitakami Fab2 plant is older news, announced with Kioxia on July 2 — it is not a Thursday item and should not be counted as one.

The other piece of the story is High Bandwidth Flash. Sandisk and SK hynix released the first Open Compute Project technical specification for HBF on Aug. 3 — earlier in the month, not Thursday — and Sandisk used the investor day to lay out a commercialisation path for the architecture, which it frames as necessary for AI inference workloads that need more memory capacity than current stacks economically provide. Whether HBF becomes a real product category or a standards-body artifact is unresolved, and no revenue in any Sandisk target depends on it yet in a way the company has quantified publicly.

The backdrop helped. The S&P 500 closed Thursday at 7,798.99, up 0.65% for a record close, the Nasdaq composite finished at 26,803.03, up 214.54 points or 0.81%, the Dow rose 0.13% to 53,839.99 and the Russell 2000 added 0.24% to 3,052.85, per the Associated Press tally of Thursday's closes; both the S&P 500 and the Russell 2000 set record closes. The S&P traded above 7,800 during the session but did not close there. 24/7 Wall St. reported that Korea's Kospi had risen about 4% overnight on AI infrastructure optimism, lifting the memory names before the U.S. open and before Sandisk presented.

Sandisk's own fiscal fourth quarter, reported Aug. 5, gives some sense of what peak-cycle memory economics look like: revenue of $8.97 billion, GAAP net income of $6.90 billion and GAAP earnings of $43.97 per diluted share, against non-GAAP net income of $6.16 billion and non-GAAP earnings of $39.25, on full-year revenue of $20.25 billion, per the company's earnings release. Note the direction of that gap. GAAP earnings above adjusted earnings is the reverse of the usual pattern, and the reason sits in the reconciliation: an $804 million net gain on equity securities runs through GAAP net income and is stripped out of the non-GAAP figure as non-core. The adjusted number is the more conservative one here, which is not how these two measures normally line up. Alongside those results the company added a $14 billion repurchase authorisation, taking remaining authorisation to $15.5 billion.

For the smaller names in the memory and storage supply chain, Thursday was a beta day rather than a fundamentals day. 24/7 Wall St., writing at 3:11 p.m. Eastern, had Western Digital up about 7% and SK hynix up about 9% while the session was still open; those are intraday marks, and neither figure should be treated as a close. Suppliers of test equipment, packaging and controller silicon tend to move on the same headline without any change to their own order books.

The risk that does not show up in a one-day chart is the one Sandisk itself is underwriting: a supply forecast extending two years out, in an industry whose defining characteristic is that supply forecasts two years out are usually wrong. The contracts and the buyback authorisation are real and disclosed. The 2030 margin structure is an aspiration, and the difference between the two is the whole question.