WDC

The call in plain words. Western Digital now makes only hard drives. Its flash business became SanDisk in 2025. AI has created a real shortage of drive capacity, and the company is earning more than it ever has. In the September quarter, the price it got per terabyte rose 27% from a year earlier while its cost per terabyte fell 5%. Gross margin reached 57.8%. Before 2025, the industry's best year was 31%.

We think the shortage lasts about two more years. Contract prices are set for most of calendar 2027 and less of each year after. Both big makers are adding density, not drive counts. After that, price per terabyte falls back toward cost and the industry settles at a margin above its old normal. We use 42% in the central case, against an old normal of about 26%.

At $354 the market is paying for most of today's margin to last. Our central value is $201. The optimistic case is $426; it keeps prices rising through fiscal 2029 and settles at a 55% margin. We rate the stock Avoid.

How to read the cases. The central case is our best estimate and alone decides the call. The pessimistic (bear) case is a severe but plausible bad outcome, about one time in five to ten. The optimistic (bull) case is the matching good one.

The exit value is the mean of two estimates:

• Fade: lets the company's return on new investment drift to a mature level over fifteen years.
• Outside view: what comparable companies have traded at, as a multiple of operating profit.

Western Digital's fiscal year ends near 30 June. FY2027 runs from July 2026 to June 2027.

1. What the September quarter showed

We read the following:

• the Q1 FY2027 release (8-K, 2026-10-22);
• the full call, prepared remarks and all twelve analysts' questions;
• the FY2026 annual report (10-K, filed 2026-08-14);
• the 8-Ks since it;
• the two previous calls;
• the CFO's talk at Citi's conference on 2026-09-09.

We did not read the slides; they are not on EDGAR. The quarterly report (10-Q) was not yet filed. The five things that matter most:

1. The beat came from price, not volume.
   • Revenue was $4.381B: up 55% from a year earlier and 6.9% above the $4.10B guidance midpoint.
   • Non-GAAP gross margin was 58.1%, 2.6 points above guidance.
   • Exabytes shipped (millions of terabytes) were 248, up 22%.
   • Price per terabyte rose 28% from a year earlier, after 18% the quarter before (CFO, call).
   • Our cost of revenue per terabyte, from the release and the exabytes, was down 4.8%. Management blames DRAM and NAND inside the drive (CFO, call). Its long-run aim is about 10% a year.
2. The guidance accelerates again.
   • Q2 FY2027: revenue $5.1B ± $0.1B (+69% from a year earlier at the midpoint), non-GAAP gross margin 60-61%, EPS $5.80 ± $0.20 on about 380M shares (release).
   • The added margin on each added dollar of revenue is guided at about 75%, after 80% in Q1 (analyst's question, not disputed).
   • Over the last six quarters, revenue beat the guidance midpoint by 2.7% to 6.9% and gross margin by 0.8 to 3.0 points (output.txt).
3. The agreements fix price for a while, and less the further out.
   • The 10-K says the long-term agreements commit customers to buy, and Western Digital to sell, "agreed volumes ... at predetermined or formula-based prices for the duration of the contract terms" (Item 1A). The CEO describes contractual price increases at each agreement's anniversary (call).
   • The CFO on pricing: "The further you go out, the less of it is covered under LTAs" (call). The agreements carry base and upside volumes. At Citi he said agreements cover the vast majority of calendar 2027, some of 2028 and less of 2029.
   • The CEO has "good visibility up to 2030, 2031" of demand from his top seven customers (call). Some customers ask for agreements running to 2031 (Citi). Visibility of demand is not a price.
   • The 10-K calls remaining performance obligations "not material" at 2026-07-03. That disclosure leaves out arrangements of a year or less, so it does not show the agreements lack prices. It does show no large disclosed multi-year backlog. SanDisk's agreements, by contrast, carry $93.9B of minimum revenue.
4. Growth comes from density, and HAMR is still ahead.
   • The new up-to-40TB conventional drive (energy-assisted recording, called ePMR) will carry about 30% of nearline terabytes in Q2 and 50% in Q3. Nearline drives are the high-capacity drives sold to data centres.
   • First-generation HAMR (heat-assisted recording) finishes qualification with two cloud customers in the March 2027 quarter. Volume shipments start in the first half of calendar 2027. Second generation, 5TB per platter, starts qualifying in late 2027 (call). HAMR is priced per terabyte like conventional drives, with no discount (CEO, call).
   • Management calls Q1's 22% exabyte growth "the floor" (CEO, call).
   • Capital spending runs above its 4-6% of revenue range for two years (call). It goes into recording heads, because HAMR heads take longer to make, not into more drives.
5. The cash goes to shareholders.
   • Free cash flow was $1.377B in the quarter: $1.604B from operations less $227M of capital spending (release).
   • The dividend rises 17% to $0.175 a quarter.
   • Net cash was about $0.5B at 2026-10-02.
   • The last $109.5M of 2028 convertible notes is called for 2026-11-16 (8-K 2026-09-14).
   • Management's policy is to return all free cash flow (Q4 FY2026 call; Citi).

2. Is this cycle different? Partly, and the price needs most of it

The case for "different" is real:

• Three makers. Western Digital about 48% of hard drives, Seagate 42%, Toshiba 10% (TrendForce, 2026-10-06).
• AI-driven demand. About 89% of revenue goes to cloud buyers (10-K). Their AI build-out stores training data and inference logs on the cheapest bulk media, which is still the hard drive.
• Supply discipline. Both big makers say they are adding no drive-making capacity, only density.
• A new kind of agreement. Customers now sign year-ahead agreements with set prices instead of buying quarter to quarter. Seagate describes the same change: by July 2026 most of its nearline exabytes were allocated into calendar 2028, with prices set for calendar 2027 (Seagate's calls of 2026-04-28 and 2026-07-28, cited by the advocate reviewer).

History argues for a return toward normal (market data statements; Seagate 10-Ks; output.txt):

Our central case:

• The shortage runs through FY2028: price per TB is 8% above the FY2027 average, about flat on the FY2027 exit.
• Price falls in a straight line from FY2029 and reaches cost over one less a 42% normal gross margin in FY2031.
• That normal sits 11 points above any year before 2025, which is what we credit to the three-maker structure and the agreements.
• Exabytes grow 25% in FY2027 and 22% a year after, between history and management's "over 25%".
• Cost per TB falls 9% a year.

The fall in price per TB from FY2028 to FY2031 is steep: $21.31 to $8.73, 26% a year, against Seagate's 15.5% a year over FY2012-FY2022. The margin reset and the cost decline compound.

Because the normal price is cost over one less the margin, a faster cost decline lowers value in this model: 6% a year gives $222, 12% gives $182. That is deliberate. In past cycles cost savings went to buyers once supply caught up. It is also the rule's weak point if makers keep the savings.

The pessimistic case puts the reset in FY2029, with a digestion year and a 30% normal. The optimistic case holds price through FY2029 and settles at a 55% normal.

What would prove us wrong.

• Agreements that fix price later. Agreements fixing price per TB over most of calendar 2029 and beyond would show that price, not just demand, is locked in.
• A margin that holds. Non-GAAP gross margin of 55% or more for FY2029 as a whole.
• Volume we underrate. Exabyte growth above 30% a year through FY2029 at a steady price.

Each would move us toward the optimistic case. The other side:

• The bust begins. Revenue per TB falling from a year earlier before June 2028.
• A digestion year. Exabyte growth below 10% in any FY2028 or FY2029 quarter.

Either would say the pessimistic path has begun.

3. The forecast and the value

The model is model.py, with its printout in output.txt and every input sourced in assumptions.md. assumptions.md is the record as committed before the market pass; the reviewers' fixes are in section 9.

FY2027 by quarter:

• Q1 is the actual result.
• Q2 is the guidance plus the median past beat: revenue +4.3%, gross margin +2.5 points.
• Q3 and Q4 grow 7% a quarter, with margin up 1 point a quarter.

FY2028-FY2031: exabytes times price per TB, with cost per TB falling. Operating costs grow 6% a year. Tax is 17-18%. Capital spending is 6.5% of revenue in FY2027-FY2028, then 5%, scaled with cost of revenue. Free cash is paid out as earned.

The FY2028 gross margin of 68.5% is our forecast, not guidance. It follows from holding price while cost per TB keeps falling.

Where the value sits.

• Of the $201, $96 is cash paid out over FY2027-FY2031 and $108 is the exit value, both discounted.
• Today's claims cost $2.59 a share: cash less debt, unrecognised tax benefits, the global minimum tax, the pension deficit and $1B of operating cash.
• The discount rate is 12.53% = 5.29% risk-free + 1.72 × 4.20%. The 1.72 is the peers' median beta after Blume's adjustment; the peers are Seagate, Micron and NetApp. On Western Digital's own beta the value is $198.

The two exit estimates disagree by 49%: the fade gives $150 and the outside view $224. On the same FY2031 after-tax profit, the outside view pays 16.2 times and the fade 10.9 times. The outside view's 13.3x operating profit comes from Western Digital's own ten-year median and its peers'. That history includes trough years priced on depressed profit (FY2016, 37x; FY2020, 48x). A normal-year FY2031 profit is not a trough, so we trust the fade more. We keep the method's mean, which adds $21 a share over the fade alone.

What moves the central value most (output.txt):

Management's framing is the gap in one line. If price per terabyte never falls again while cost keeps falling 10% a year, the business is worth far more than the price. Our view is that no commodity component has done that through a full build-out.

4. What the market expects

At $354.19 the market capitalisation is $132.6B (market data), or $134.6B on our 380M diluted shares. The price fell 6.4% on the day of the results, before the release, and 4.7% more after hours. It is 56% under its 52-week high of $799.87.

Alone, with everything else central, the price needs one of these:

• a normal gross margin of 57.5%, about Q1's margin forever;
• price per TB held through FY2030 and a 55.3% normal after;
• exabytes growing 41% a year;
• an exit at 45 times FY2031 operating profit.

Combined, a smaller version of each will do. The advocate reviewer priced one coherent world in our model at $371 ($369 after the auditor's fixes):

• a 55% normal;
• exabytes +25% a year;
• price held through FY2029.

In that world price per TB still falls 19% a year from FY2028 to FY2031. So the price does not need peak pricing forever. It needs most of today's margin kept after HAMR and Toshiba's capacity arrive.

The Street against our central (market data, averages). These estimates may not yet carry the Q1 results:

We are above the Street on the next two years and far below it after. The difference is not the shortage; it is whether it ends.

The decisive calculation (market.txt). We ran the Street's revenue and operating profit through our model, year by year, keeping our tax, capital spending and exit method:

Both reviewers reproduced this. Using the Street's EPS in place of its operating profit gives $228 through FY2030 and $684 with FY2030 held. The Street's next four years are worth about what ours are. Nearly all of the gap to its $635 target comes from treating FY2030's peak profit as the exit's run rate. FY2030's operating margin is 71% on five analysts' revenue, and its EPS comes from one analyst.

We cannot see the analysts' models, so this does not prove their targets assume exactly that. It does show the value turns on the exit year's profit, not on the next four years.

Alpha. Capfolio's quantitative ranking model ranks Western Digital 521st of 702 on 2026-10-22 ("Below Average"):

Its integrity check warns on six inputs. The growth grade looks wrong for a company growing revenue 55%. The likely cause is the SanDisk separation in the history; we did not check it. Alpha and this view agree on the call for different reasons.

The chart. Our technical read grades the stock D, timing only and never an input to value:

• The stock is topping: the close is under a rising 30-week average of $468.94.
• It is below the 50-day average of $435.87.
• The September swing low of $407.12 is broken.
• The market and the semiconductor group are both weak.
• The grade rises on a weekly close above the 30-week average once that average is rising.

5. Where we differ, and what settles it

We agree with the bulls on the next two years. Our FY2027 and FY2028 revenue are above the Street's, and the Q2 guide was another raise. We differ on what happens once Toshiba's new capacity and HAMR's density gains arrive while buyers digest what they bought. Three releases settle it:

• the price per TB in each quarter of calendar 2028;
• how far ahead agreement prices are set, as reported on each call;
• the gross margin for FY2029.

6. Why the other side could be right

The strongest case for the price has four parts:

• Lasting AI storage demand. Inference keeps every token, the "KV cache", plus logs, video and physical-AI data. The CEO says the gap between supply and demand is not narrowing. Demand from AI labs and newer cloud providers broadens the buyer base.
• No glut is coming. With two makers holding 90% and both adding density rather than capacity, supply can only grow as fast as areal density. Areal density is how many bits fit on a platter.
• Price stays above cost. Price per TB stays above cost for longer than any past cycle: drives remain about five times cheaper per TB than flash in bulk storage, so buyers have nowhere else to go.
• Management already says so. The advocate reviewer found Western Digital's Innovation Day of 2026-02-03. There management changed its pricing outlook from yearly declines to stable pricing, with mid-20s nearline exabyte growth and gross margin above 50% over three to five years. Seagate's June-quarter GAAP gross margin was 52.3%. That is a dated claim made before this quarter, and we have not read the presentation.

Our optimistic case is built on this view and is worth $426, 20% above the price. We give the structural world (shortage to FY2030, 55% normal) a fifth of our odds weight. No agreement of this kind has yet met a glut. The advocate would give it two fifths, which puts the odds of the price being justified near 30%.

7. Next results preview: Q2 FY2027, about 2027-01-28

The date is tentative (Robinhood) and not confirmed by the company.

Each $100M of Q2 revenue above ours at our margin is about $0.14 of EPS. A quarter barely moves the value; the normal margin does. Two signals in Q2 matter more than the beat:

• whether price per TB growth from a year earlier still rises (Q1: 28%);
• what management says about agreement pricing for calendar 2028.

8. Catalysts and falsifiers

Catalysts:

• About 2027-01-28: Q2 FY2027 results.
• About 2027-02-05: the big four cloud buyers' 2027 capital spending guides. A 2028 cut of 20% or more is our pessimistic path.
• By 2028-03-31: Toshiba's doubled capacity is due.

Falsifiers:

• A FY2029 non-GAAP gross margin of 55% or more.
• Fixed-price agreements covering most of calendar 2029.
• Revenue per TB down from a year earlier in any quarter before June 2028.
• Exabyte growth below 10% from a year earlier in any FY2028 or FY2029 quarter.
• Toshiba's share above 15% by 2028.
• Exabyte growth above 30% a year through FY2029 at a steady price.

9. What the reviewers found, and what we did

Who reviewed. Two outside reviewers read the draft at commit 71cbaf35e, each separately and without the other's findings:

• a model auditor, who reran every script and checked each input against its filing;
• an advocate, who argued the Buy case from its own dated sources.

Both were OpenAI Codex agents, run read-only. Their reports and the scripts behind every figure are saved as data/review-auditor.md and data/review-advocate.md.

Before the reviews, research-check raised three flags:

• the value is 42% from the price;
• the fade and the outside view are 49% apart;
• the value is 0.36 times the Street's target discounted one year.

The auditor. It reproduced output.txt, market.txt and odds.json exactly. Effects are each alone, on the committed model; together they move the central from $203.50 to $200.63, and the interaction is under $0.01.

The auditor found clean:

• cash, debt and the convertible notes (principal in debt, premium in shares);
• the 380M shares and stock pay;
• financing income;
• the tax claims and payout timing;
• every reported input it traced.

The advocate. It argued that fair value could reach $354.19.

Net effect of the reviews: central $203.50 to $200.63 (-$2.87), bear $64.34 to $64.60, bull $428.60 to $426.37. The view is unchanged, so no second round.

10. Sources

• Western Digital filings (EDGAR). Q1 FY2027 release (8-K, 2026-10-22); FY2026 10-K (2026-08-14); 8-Ks of 2026-08-26, 2026-08-27 (amended), 2026-09-14; releases Q2 FY2025 to Q4 FY2026.
• Earnings calls. Q1 FY2027 (2026-10-22), Q4 FY2026 (2026-08-05) and Q3 FY2026 (2026-04-30); Citi Global TMT conference (2026-09-09). Quartr transcripts via TradingView.
• Seagate. 10-Ks FY2012-FY2026 (EDGAR).
• Financial statements. market data, read 2026-10-22.
• Toshiba. Nikkei via Evertiq, 2026-10-20.
• Market shares and the TDK head unit. TrendForce, 2026-10-06.
• Cited by the advocate reviewer, not read by us:
  • Western Digital Innovation Day presentation, 2026-02-03;
  • Seagate's calls of 2026-04-28 and 2026-07-28, and its FY2026 results;
  • TrendForce's forecast of 2026-10-07.
• Market data. market data quote, estimates and targets (2026-10-23 02:15 UTC); Robinhood quote (read-only, 2026-10-23 02:13 UTC).
• Our model. Alpha rank, 2026-10-22.
• Work files. Saved and pinned in data-manifest.json.