The same growth, made of the opposite ingredients
A hard-drive quarter has exactly two inputs. You ship a number of exabytes, and you get paid a price per terabyte for them. Everything else — margin, operating leverage, free cash flow — is downstream of which of those two moved. Western Digital discloses both, which means the decomposition is not an estimate. It is arithmetic anyone can run in a line.
Run it on the fourth quarter. Revenue grew 44%. Exabytes grew 22%. The residual is price, and it is +18.0%. Now the third quarter, three months earlier: revenue 45%, exabytes 34%, and a price contribution Irving Tan gave as a number rather than leaving to the residual — asked directly, "Yes, pricing was up 9% year-on-year." The top line barely moved between those two quarters. What produced it swapped ends.
This is not my inference imposed on the company. It is the company's own account, in Kris Sennesael's words: the blended average year-on-year price increase per terabyte "improved from high single digits last quarter to high teens this quarter." High single digits is Tan's 9%. High teens is the 18.0% the fourth-quarter disclosures imply. Management and the arithmetic agree. What neither of them did was put that line next to the exabyte line, and that adjacency is where a datapoint becomes a finding.
Hold the sequential figures next to it, because they say the same thing without any year-on-year base effect at all. Exabytes went 222 to 231, up 4.05%. Revenue went $3,337M to $3,747M, up 12.29%. Roughly two-thirds of the sequential top line arrived without a single additional exabyte leaving a factory.
The cost side moved the same way, which matters because it is the other half of the margin. Cost per terabyte fell about 8% year on year in the fourth quarter, against 10% reported for the third and a stated long-run target of about 10%. So in the quarter that produced a 1,310bps margin expansion, volume growth slowed and cost reduction slowed, and price covered both gaps and then some. That is also why the margin beat was as large as it was: WDC guided this quarter to 51% to 52% and printed 54.4%, 240bps above the top of its own range.
The rest of this dive is for paid subscribers.
The decomposition is above. The seven sections below carry what management called the deceleration and what it actually is, the fixed-asset series that says this cannot be fixed with money, who has to elect the next 25% of exabytes, the mark that produced two-thirds of GAAP net income, the Seagate comparison put to the CEO on his own call, and the dated test that settles it.
- Mix is the softest margin to capitalize — and the softest miss to excuse
- The plant that earned the windfall predates it
- The next exabyte is the customer's decision, not the fab's
- Two-thirds of GAAP net income is a mark on a company they no longer own
- Seagate, and the question asked on WDC's own call
- The strongest case against this read
- Valuation, the trade, and what settles it
Already a subscriber? Sign in.
Mix is the softest margin to capitalize — and the softest miss to excuse
Amit Daryanani of Evercore asked the obvious question: exabytes up 22% is "below the 30% exabyte growth you folks have had in the last several quarters," so is this timing, supply, or the new normal? Irving Tan's answer was mix, and he was specific about the mechanism: "if there's a particular customer in a particular quarter that takes a bit more CMR products, obviously, that will mean we ship a fewer bits into the marketplace for the number of units that we have."
Take that seriously, because it is almost certainly true — and then notice what it concedes. This newsletter has a standing rule that mix is the softest margin to capitalize: a result explained by mix has no price increase to point to and no cost removed to verify, just the weighted average of what happened to ship, so you refuse to capitalize it as a baseline until it prints twice. The rule cuts in both directions, and here it is being used the other way — to explain a shortfall rather than a beat. The problem is that the explanation is structural, not a one-quarter accident. Tan is describing a business where the number of exabytes shipped off a fixed number of units is set by which format the customer chooses to take. That is not noise around a supply number. It is the admission that the supply number is not entirely WDC's to set.
Sennesael's framing of the same fact is the tell: "in Q4 exabyte shipments were up 22% year-over-year, and when you look at it on a full fiscal year '26 exabytes were up 25% year-over-year." Both figures are accurate. Reaching for the annual number to contextualise the quarterly one is what you do when the quarterly one is the question.
The plant that earned the windfall predates it
The natural next question is whether WDC is spending to fix this — building the units that would let exabytes grow without the customer's cooperation. The answer is a clean no, and the filing is unusually direct about it.
The standing model here is price does the work the capex hasn't done yet, and it was revised on this site two days ago after reading a Chinese DRAM maker's capex decline as a decision when it was a construction timetable. The correction matters, so I will apply it as corrected rather than as originally written: a single year-on-year capex move proves nothing, because capital programmes run in waves and cash capex is the lagging record of one. The leading indicators are construction-in-process and the commitment note. That is the discipline behind the other standing rule, a one-period change is a phase, not a decision, until you have the series. So here is the series, and this time the lagging and leading lines agree instead of contradicting each other.
Capital expenditure was $418M in fiscal 2026 against $412M in fiscal 2025 — flat, on revenue that grew 36%. Depreciation was $372M, $334M and $347M across 2026, 2025 and 2024, so capex ran at 1.12x depreciation, down from 1.23x. Gross property, plant and equipment grew 4.07% to $9,455M against accumulated depreciation of $6,979M — the base is 73.81% written off. And construction-in-process, the leading line the correction says to read, fell 14.10%, to $457M from $532M.
Then the 10-K states the decision in a sentence: "In 2026, the Company canceled certain facility projects and impaired related construction in progress assets as realigned resources to optimize capacity to support recent growth." Not deferred. Cancelled and impaired, in the year of the windfall.
None of this is a criticism of the decision. Earning a 54.4% gross margin and $3,511M of free cash flow off a plant that is three-quarters written off is an outstanding return on assets somebody else paid for, and paying down total debt from $4,711M to $1,052M while returning $3.1B is exactly what you should do with it. The point is narrower and it is about what the multiple is buying: this is a harvest, run deliberately, and a harvest has no capacity option inside it. Sennesael said so without hedging — the plan to deliver substantially more exabytes "does not require spending CapEx to add unit capacity, but we are making the necessary investments in our heads and media operations as well as in automation to increase our productivity." And this is a position held across two calls, not a line that slipped out once. Asked in April whether serving 25%-plus demand would eventually force a capacity build, Tan was flat: "at this juncture, we still do not see any need to increase unit capacity. So we have no plans for that." Asked in the same session about customer prepayments to fund capacity, he drew the line precisely where the fourth quarter would later test it — "we'll look to add head and media capacity in terms of investments, but not unit capacity investments."
The next exabyte is the customer's decision, not the fab's
If units are fixed, every additional exabyte has to come from areal density or from format mix. WDC has a credible roadmap for the first: 40-terabyte ePMR drives began shipping in the June quarter and are in volume production with two customers, targeted at over 50% of nearline bits by the third quarter of fiscal 2027; 44-terabyte HAMR arrives in the first half of calendar 2027 and 50-terabyte in the second half; high-bandwidth drives at up to 8x today's throughput are sampling with five customers.
The second is the exposed one. UltraSMR is expected to be about 60% of nearline exabyte shipments exiting fiscal 2027, carries a 20% capacity uplift "without the associated cost," and only delivers that uplift if the customer runs it — which is precisely the CMR-versus-UltraSMR choice Tan named as the cause of the 22%. And the easy half of that conversion is already spent. On the third-quarter call the company said three of its largest customers had adopted the technology, with "two already meeting nearly all of the exabyte demand with UltraSMR, while the third is rapidly ramping in that direction." When two of your top accounts are already near-fully converted, the mix lever at those accounts is largely pulled; what is left is the third customer and a Tier 2 base that has to be qualified. So the supply plan has a dependency that is not on WDC's balance sheet or in its fabs. It is in a purchasing decision made by a very small number of people. Cloud was 89% of fiscal 2026 revenue. Three customers were 16%, 15% and 13% — 44% between them. The top ten went 55% to 68% to 73% of revenue across three years. Concentration is not stable here; it is compounding.
That is what makes the Q4 exabyte number worth more than one quarter's attention. It is the first observation of the density engine underdelivering, and the stated cause is the one input the company does not control.
Two-thirds of GAAP net income is a mark on a company they no longer own
GAAP diluted EPS was $8.21 against non-GAAP $3.56; for the year, $24.28 against $10.22. That gap is $4.65 and $14.06, and it is almost entirely one item. GAAP net income of $3,195M sits on operating income of $1,563M, bridged by $1,684M of interest and other income containing a $2,050M gain on the retained interest in SanDisk — 64.2% of the quarter's GAAP net income. For the full year the gain was $6,498M, or 69.0% of $9,424M.
Anyone reading non-GAAP has already netted this out, and WDC did not hide it. What is worth writing down is that the position is now closed: the company "completed the monetization of the remaining 1.7 million shares of SanDisk, exchanging them for 4.8 million WD shares," and retained interest on the balance sheet went to zero from $354M. Fiscal 2027 does not get this line at all, so a reader anchoring on $24.28 is anchoring on something that has already stopped.
One smaller reconciliation, in the same spirit. The press release reports $672M of buybacks; the call says $1 billion and 2.3M shares. Both are correct, and Sennesael disclosed the bridge himself — the call figure includes "$328 million to settle the conversion premium for some of our converts in cash rather than in stock, avoiding the issuance of roughly 773,000 new shares." That is anti-dilution spending, and it is real value, but it is not the same act as retiring float and it belongs in a different column when you are counting capital returns.
Seagate, and the question asked on WDC's own call
I do not have to construct the peer comparison, because C.J. Muse of Cantor Fitzgerald opened the Q&A with it: "It's hard not to compare your results with your main competitor where they're seeing better sequential top line growth and targeting gross margins nearly 200 bps higher than your September guide. So curious what you make of this? Is that due to the earlier ramp of HAMR?"
The premise checks out on the reported figures. Seagate's June quarter carried a 52.7% non-GAAP gross margin, up 570bps sequentially in a thirteenth consecutive quarter of expansion, on data center revenue of $2.9B up 57% across 195 exabytes, with HAMR at roughly 40% of its nearline exabyte run rate. Both companies guided September to $4.1B. WDC's absolute margin is the higher of the two today at a guided 55% to 56%; the gap Muse is pointing at is the trajectory, and its cause is structural rather than commercial. HAMR raises capacity at the drive, so a HAMR exabyte ships whether or not the customer changes anything. UltraSMR raises capacity at the format, so a shingled exabyte ships only if the customer adopts shingled. Seagate has already converted 40% of its nearline bits to the mechanism that does not need permission. WDC's equivalent arrives in the first half of calendar 2027.
Tan's answer was timing — "there will always be timing differences of when existing LTAs expire and new LTAs kick in with different pricing regimes" — which is a good answer to a question about margin and not an answer to the question about exabytes.
The strongest case against this read
The best argument against everything above is the contract book, and it is a serious one. Tan disclosed that one large customer is under a long-term agreement running to calendar 2029 and that WDC is in active negotiation for calendar 2029, 2030 and 2031. If a business with 73% of revenue in ten accounts converts those accounts to multi-year contracted volume, then the price-versus-volume decomposition I have spent this piece on stops being diagnostic. Contracted exabytes are not elective. A three-to-five-year book of them turns a cyclical manufacturer into an annuity, and against an annuity 43.2x trailing and 27.6x the annualised September guide are not obviously wrong — they are the multiple you would expect. On this reading the fourth quarter was a scheduling artefact inside a book that is being locked down, and the correct response to a 13.03% drawdown was to buy it. That case rests on real disclosure, not on hope, and it may well be right.
Here is why I still land where I land. Tan told us which half of those agreements is settled and which is open, and it is the unhelpful way round: "we have very good line of sight in terms of the exabyte demand. What we're working through with them is more in terms of the pricing commercial construct of what those LTAs would look like going forward." A commitment is an allocation claim before it is a cost — but the claim being locked here is the volume, and volume was never the number in doubt for a business whose customers are out of storage. The variable carrying 18.0% of a 44% quarter is the one still on the negotiating table, with counterparties who have just watched their supplier print a 54.4% gross margin off drives they have to buy. Locking the exabytes and leaving the pricing construct open de-risks the number that was already safe.
Valuation, the trade, and what settles it
At $441.57, WDC is worth approximately $161.85B. That is 43.2x fiscal 2026 non-GAAP EPS of $10.22, 27.6x the September guide of $4.00 annualised, 46.1x fiscal 2026 free cash flow of $3,511M and 12.5x revenue of $12,919M — for a company that manufactures hard disk drives and has told you it is not adding unit capacity. Those multiples are payable if exabytes compound at 25%-plus for years. They are not payable on an ASP cycle, because an ASP cycle mean-reverts and the balance sheet has no volume lever to catch it with.
What is genuinely reassuring is the quality of the earnings themselves, and it deserves saying plainly because it kills the lazy version of this argument. Days sales outstanding compressed to 49.2 from 51.9 while revenue grew 44%. Inventories rose 17.0% against revenue up 36%. Operating cash flow was $1,389M against non-GAAP operating income of $1,655M, and free cash flow of $1,281M was a 34% margin. Nothing here was pulled forward or financed by the customer. The quarter is exactly as good as it looks. My argument is not that the earnings are low quality — it is that they are high-quality earnings from a source the multiple is mispricing.
On structure — analyst framing, not advice — the reaction already did most of the repricing, and the volatility went with it. Thirty-day implied volatility is 0.642 with an IV rank of 33.53, against 0.974 and 77.22 on the day of the print, in a name that just travelled through its 8.98% expected move to close down 13.03% and touch −21.51% intraday. For a holder, that combination argues for expressing the next print through defined-risk optionality into rather than through the shares, because the outcome is genuinely binary on one disclosed metric and the options are no longer pricing it as though it is.
Bottom line. Everything Western Digital reported is true and most of it is excellent. Revenue of $3,747M, up 44%. A 54.4% gross margin. $3,511M of free cash flow at a 27% margin. A net cash position after retiring $3,659M of debt. What the print does not establish is that the growth is still coming from the thing the multiple is paying for.
So the test is one line in the October release, and it is not revenue. WDC reports its first quarter on . If exabytes shipped come in at or above 25% year on year with revenue at or near the $4.1B guide, then Q4's 22% was the customer-mix lumpiness Tan described, the density roadmap is carrying the load on schedule, and this read was wrong. If exabytes print below 25% again and revenue still lands at $4.1B, then price has done the work in back-to-back quarters on a fixed asset base — and the company will have spent two quarters selling the same drives to the same three customers for more money, which is a fine business and a different one from the one trading at 43.2x.