On September 4, SanDisk (SNDK) closed at $1,740.00, up roughly +555% year to date (measured at the September 3 close), the steepest curve in the memory complex this year. The endorsement behind the move is the FY26 10-K filed on August 17 (fiscal year ended 2026-07-03): revenue of $20,248M, up 175% year over year; GAAP gross margin of 71.5% against 30.1% a year earlier; net income of $11,433M for a 56.5% net margin; and operating cash flow of $11,671M, versus just $84M in the prior fiscal year. In the space of a year, this company went from a spun-off asset trading below book and writing off $1.8B of goodwill to a security priced like an annuity.
Our core judgment is this: SanDisk is running the first experiment in NAND's history to put roughly two-thirds of output into take-or-pay contracts, the floor has been genuinely raised with audited-statement-grade evidence, and directionally we continue to accept that. But the assumptions embedded in today's price are far more demanding — they require treating a peak free-cash-flow margin as a perpetual steady state. And NAND happens to be the one among memory's three sub-industries with the worst supply discipline and the earliest clock on supply loosening: TrendForce's baseline call is that the sufficiency ratio turns positive in 2H27 with prices under pressure in the second half, and the first year of SanDisk's FY28 target window runs straight into that window. This note unpacks the pricing, weighs the evidence, and lays out the criteria.
RPO of $59.8B and a $93.9B Floor Minimum: Contractualization Is Already on the Balance Sheet
Start with the strongest part of the bull case. Ten long-term agreements across eight datacenter and Edge customers; minimum expected revenue of $93.9B under floor pricing; RPO of $59.8B, of which $58.7B is unbilled and roughly 19% is to be recognized in the next twelve months; contract liabilities of $1,242M and customer deposits of $1,500M; two further agreements signed after the balance sheet date totaling $31.3B; financial guarantees of $16.5B; terms running up to five years with a weighted average above four years; pricing structured as fixed plus floating, with the floating component carrying both a floor and a ceiling. On the company's own definition, more than half of FY27 bits and roughly two-thirds of FY28 bits fall inside the contract framework. One apparent tension deserves a pre-emptive answer: the ~19% of RPO recognizable in the next twelve months (roughly $11.2B, a derived figure) looks lower than the annualized run rate implied by FY27 revenue guidance, but this does not contradict "more than half of bits under contract" — RPO captures only the fixed minimum consideration in the contracts, while the floating-price component and the bit-volume commitments sit outside that number. The same structure is being replicated across the industry: Micron has sixteen take-or-pay agreements with cumulative minimum-price revenue of roughly $100B, and Kioxia's long-term contract target covers about half of its CY28 shipments.
In our view, this is an institutional innovation in which the supply side collectively sells its cyclical volatility to the demand side, with the bill picked up by cloud vendors anxious to lock in supply. Its durability hinges on a question that has never been tested: once supply loosens in 2H27, will buyers still perform at the floor? Take-or-pay has never been stress-tested through a NAND oversupply phase. And one critical figure remains suspended: the gross margin associated with the floor has still not been disclosed. The "roughly 80%" the CFO offered on the August 5 earnings call is the transaction gross margin under the current contracting environment; what it becomes under the floor case is the pricing core of the "raised floor" story. That disclosure has no fixed date, but it is the single highest-weighted item on the watch list.
Company-Wide Bits Grew Only Mid-Teens: Two-Thirds of FY26's Earnings Improvement Came from Price
The 10-K's volume-price decomposition is worth reading line by line. Datacenter revenue of $5,153M (+437%) broke down into roughly +120% bits and roughly +150% price per unit of capacity; Edge grew bits by a high single digit with unit price up roughly 180%; Consumer bits fell mid-teens. Company-wide bits grew only mid-teens in total. Roughly two-thirds of the full-year earnings improvement came from price and one-third from volume and mix (on the 10-K's volume-price decomposition); Q4 revenue of $8,965M (+51% QoQ) showed the same split on the sequential increment: one-third volume, two-thirds price. This is still a price cycle, and volume expansion is concentrated in a single segment.
Capital discipline is the most solid stretch of the bull narrative: cash capex for the full year was just $177M, with a further $275M of net funding into Flash Ventures, together under 3% of $20.2B in revenue; bit growth comes from the node migration from BiCS8 to BiCS10 rather than added wafers; FY26 buybacks totaled $4.5B, with a further $14B authorization granted in August. Not adding capital during a high-return period is a textbook move. But be clear about the subject: SanDisk's discipline does not bind the industry. YMTC sits outside the discipline framework, and TrendForce expects its global bit share to rise toward 19% in 2027; migration to ultra-high layer counts is itself implicit capacity addition, and "not building new fabs" is not the same as "not adding supply." The implicit costs are written into the 10-K as well: the 15% corporate alternative minimum tax (CAMT) applies from FY27, and Malaysian tax holidays expire in stages across 2028-2031; inventory days stood at 178, up 43 days year over year, and the high-cost inventory staged for contract performance will amplify gross margin volatility in reverse during a price downturn.
What the Market Has Believed: ~9% Revenue CAGR, ~57% Margin, Forever
Solving backward from the early-September closing price (reverse DCF: WACC 9.5%, terminal growth 3%, ten-year window, with growth fading linearly to the terminal value across years 6-10), today's price implies a five-year revenue CAGR of just 8.98% and a steady-state free cash flow margin of 56.77% (the solved values are sensitive to assumptions; read them as roughly 9% and roughly 57%). Fixing the revenue growth rate and solving instead for margin, or for the competitive advantage period, yields no solution. This set of mathematics is compatible with only one reading: the entire weight of the current price rests on the single assumption of margin permanence, with growth assumptions barely participating in the pricing at all.
There are two reference points for weighing that assumption. First, within the roughly $11.5B of TTM free cash flow (operating cash flow of $11,671M less capex of $177M; deducting the $275M of net Flash Ventures funding as well gives roughly $11.2B), the $1,242M of contract liabilities plus $1,500M of deposits, roughly $2.7B in total, are deferred revenue and deposits. Strictly speaking these two items are fiscal-year-end balances; but the contract framework launched during this fiscal year and the balances were essentially formed within the period, so they can be treated approximately as in-period net inflows — the company's own adjusted FCF definition likewise strips out contract prepayments. On that approximation, the post-strip margin is roughly 43%. Second, the FY25 full-year FCF margin was roughly 1%, and the mid-cycle FCF margin through NAND's history has been in the single digits. In our view, the market has capitalized the cyclical peak, together with a one-off contractual cash inflow, into perpetuity.
July's price action already demonstrated the fragility of this pricing once: an all-time high of $2,354 on June 22, $1,016 on July 29, a 57% drawdown, followed by the rebound to today. Here is the ruler we use to measure the divergence (framework-derived calibration, not a price target): the upper bound from a cycle mean-reversion framework is roughly $1,000, which lines up with the July low; the lower bound from a contract-cash-flow framework is roughly $2,400. At $1,740, the current price sits about 53% of the way between those two points, while the August 13 investor day close of $1,548 corresponded to about 39% — September 4 alone advanced roughly 8 points. Half of the cheap part of the framework switch has already been consumed; the remaining upside requires three things to land in sequence: FY27Q1 delivering, 4Q26 contract prices not turning negative, and the floor being disclosed at a level high enough.
The Bear Evidence: 2H27 Loosening Is the Industry Research Baseline, and the Crack on the Consumer Side Is in the Company's Own Statements
NAND's historical character is that of a repeat offender on supply discipline: six groups of players (Samsung, SK hynix plus Solidigm, the Kioxia-SanDisk joint venture, Micron, YMTC), a layer-count race with stratified technical barriers and high exit barriers, and every prior round of "this time there's discipline" ending in the prisoner's dilemma. TrendForce's bifurcated call of July 30 deserves attention: DRAM stays tight through all of 2027, while NAND's sufficiency ratio turns positive in 2H27 with prices under pressure in the second half. The spot market has already moved first, with 512Gb TLC wafer spot prices turning down by 1.47% in a single day on August 24.
The crack on the demand side requires no outside sourcing. The company's own Consumer bits fell mid-teens and Q4 Consumer revenue fell 32% sequentially, while consumer applications account for close to 40% of NAND bit demand, and enterprise SSD cannot fully offset this. Enterprise has a ceiling of its own: the per-GB price gap between QLC eSSD and HDD has widened to 15x, the economics of cold and warm data flowing back to HDD are strengthening, and customer cost discipline is the cap on penetration.
HBF: The Spec Is Set, Volume Production Is About a Year Out, and the Most-Told Use Case Is the One It Suits Least
HBF (high bandwidth flash), treated by bulls as the second growth curve, has made real progress at the factual level: the first OCP technical specification was jointly released with SK hynix on August 3 (512GB/stack, three bandwidth tiers of 0.4-3.0TB/s, UCIe interface, with Google and Tenstorrent joining), and the ecosystem position has indeed been staked out. But volume production is about a year away, with no customer qualification, no revenue, and no pricing. More to the point is workload fit: a full-stack empirical study (arXiv 2608.11668) swapped HBF directly into the KV-cache offload stack and found end-to-end latency degrading by 2-5.5x — transient KV is a write-intensive workload, and flash's write thermal wall and wear characteristics mean a faster device produces a slower system. HBF's sensible ecological niche is read-only weights and precomputed cache layers. Translating "agentic inference drives a KV-cache explosion" directly into "HBF revenue" breaks the transmission chain at the technical layer. If no mainstream xPU platform writes HBF into its product roadmap during 2027, treat the second curve as an option expiring at zero.
Verification Calendar
| Timing | Event | Interpretation |
|---|---|---|
| Oct-Nov | 4Q26 NAND contract prices settle | ≥+10% keeps the supply-discipline narrative alive / +3~8% is a neutral blunting / ≤0% breaks the first pillar and returns the cycle grammar |
| Early Nov | FY27Q1 results | Whether non-GAAP gross margin lands in the 83-85% guidance band; contract coverage progress (>50% of FY27 bits); the marginal terms of newly signed contracts (whether the floor moves down, whether tenors shorten) matter more than the stock figures |
| Timing uncertain | Floor gross margin disclosure | ≥55% turns "raised floor" from a directional judgment into a calculable floor / 40-55% means the floor is real but insufficient to make the 80% medium-term target a steady state, treat as the blunting scenario / continued non-disclosure or <40% confirms the cyclical exposure in the 80% target |
| Ongoing | Supply discipline | Any major supplier announcing a new NAND wafer fab (not a node migration), or a material upward revision to 2027 NAND capex, is the first signal of discipline breaking; an upward revision to YMTC's share estimate means supply outside the framework is accelerating; the earlier observation is on a bit basis: if TrendForce's supply-demand balance shows the NAND sufficiency ratio turning positive earlier than 2H27, or if consensus on 2027 bit supply growth is revised up, implicit capacity addition (node migration) is already working |
| Each quarterly 10-Q | Movement in refund liabilities and contract liabilities | Any customer renegotiation, default, or guarantee drawdown is direct falsification of the contractualized-annuity narrative |
| During 2027 | Whether HBF enters mainstream xPU product roadmaps | Not entering = second curve treated as an option expiring at zero; entering = upgraded to a trackable proposition |
To draw this together: contractualization has genuinely raised NAND's floor, and we accept that part; the evidence is on the balance sheet. But what the current price embeds goes beyond the floor: it treats a roughly 57% peak margin as perpetual, and roughly $2.7B of that is unearned deferred revenue and deposits. NAND's supply-loosening clock (2H27) runs earlier than DRAM's, and the first year of SanDisk's medium-term target window falls exactly where that clock goes off.
The next observation points are 4Q26 NAND contract prices in October-November and FY27Q1 results in early November; when the floor gross margin is disclosed, and at what level, is the single highest-weighted variable in the whole case. Under the same capital cycle framework, our deep read on Micron's DRAM/HBM is the companion piece to this note, and the two should be read against each other.
This note represents the research views of LunarTulip Research, compiled and analyzed from public information. It does not constitute investment advice. Markets carry risk; decisions must be made independently.