QCOMQUALCOMM Incorporated

QCOM — dossier, 2026-09-22

Slot names-ai-capex-2b. Model-drafted, unattended, by the cloud research routine following .claude/skills/stock-sourcing-pipeline/SKILL.md. Nothing here is human-verified. Every figure carries the document it was read from; anything from background knowledge is marked [background] and anything from a web search summary [search-summary]. Private notes are marked as such and are owner reasoning, not evidence. Never edited after today — corrections go in a later dossier.

Never edited after today. Process support, not investment advice.

---

0. The one-sentence version

Qualcomm is losing its largest customer on a disclosed schedule that just accelerated, is being squeezed by the memory cycle in the business that remains, and has answered both by buying its way into the data centre with stock — two all-share acquisitions inside nine months and, on 2026-09-03, a 25-million-share warrant handed to Amazon that vests against Amazon's own purchases.

1. S1 — Source

watchlist (owner, Apple Stocks, 2026-09-20). Displaced from slot names-ai-capex-2 on 2026-09-21 when the OMAB Tier 0 promotion took the third seat; queue.json promoted was empty when this slot was claimed, so no promotion applied. Cluster ai_semis, coverage sec_domestic, cik 804328.

2. The event this run exists to record: the Amazon warrant

Filed 2026-09-08 on an 8-K under Item 3.02, for an event dated 2026-09-03 — after the Q3 FY26 10-Q (2026-07-29) and after the Q3 FY26 earnings call of the same date. No periodic filing carries it yet and no accounting treatment has been disclosed.

"the Company issued a warrant (the “Warrant”) to Amazon.com NV Investment Holdings LLC, an affiliate of Amazon (the “Warrantholder”) to acquire up to an aggregate of 25,000,000 shares (the “Warrant Shares”) of the Company’s common stock at an exercise price of $161.26 per share." — 8-K filed 2026-09-08, Item 3.02, first paragraph

"The Warrant Shares vest in tranches tied to the execution of certain commercial arrangements, the placement of binding purchase orders and actual purchases of QTI’s server chip products, technology, systems and manufacturing services by Amazon during the term of the Warrant, up to a maximum amount of $60 billion in payments, with 3,750,000 shares being vested upon issuance of the Warrant based on initial purchase commitments." — same filing, Item 3.02

Mechanics, all from the filing: cashless exercise permitted; expires 2036-09-03; no voting rights while unexercised; registration rights, with a resale prospectus supplement expected off the existing S-3ASR.

Arithmetic (this run's, from filing inputs): 25.0m shares is 2.37% of the 1,057m outstanding at 2026-06-28. The $161.26 strike is below the 2026-09-03 close of $168.57 and 17% below the 2026-09-21 close of $194.23 (ROIC NASDAQ:QCOM), so the 3,750,000 shares that vested on issuance were in the money on day one — roughly $124m of intrinsic value at the 2026-09-21 price.

The CFO described the commercial side at Goldman Sachs Communacopia the same morning:

"we issue warrants against purchases of up to $60 billion from Amazon for data center products over the next 10 years. And there is an upfront westing of about 15% of those warrants that's associated with the $60 billion, and that is tied to upfront commitments that are being made by Amazon." — Akash Palkhiwala, Goldman Sachs Communacopia + Technology Conference, 2026-09-08, Equibles speaker-labelled transcript. ("westing" is the transcript's transcription of "vesting"; quoted as the tool rendered it rather than re-typed.)

Two product components, per the same answer: customised silicon over multiple generations, and optical connectivity starting at 1.6T — the latter being the Alphawave assets acquired in December. So the $2.3bn Alphawave deal is the technical precondition for this agreement.

Why it matters to this watchlist. names-ai-capex-1 (2026-09-21) recorded as its cross-cutting finding that all three of NVDA, AMD and AVGO now underwrite their own buyers. names-ai-capex-3 recorded Cerebras as the inversion, where the buyer (OpenAI) finances the seller. QCOM is the fourth seller, and the closest structural match to AMD: equity to the customer, vesting on the customer's purchases. The 2026-09-20 cluster file split the complex into self-funding sellers and debt-funded buyers and put Qualcomm in the first group. It is still self-funding — but it is no longer only a seller.

3. S2 — Kill test

Five-year income statement (SEC XBRL, CIK 804328, us-gaap annual concepts from the 10-Ks)

The FY25 net income line is not an operating event. Operating income rose 23% and net income fell 45%, because income tax expense went from $226m to $7,122m — a 56% effective rate:

"As we expect to perpetually be subject to CAMT, we no longer expect to realize substantially all of our existing federal deferred tax assets and recognized a charge of $5.7 billion to income tax expense to establish a valuation allowance in the fourth quarter of fiscal 2025." — FY25 10-K, Item 7, MD&A, Income Tax Expense

And it reversed three quarters later, after IRS Notice 2026-07:

"Accordingly, we released our valuation allowance on our federal deferred tax assets resulting in a $ 5.7 billion income tax benefit in the second quarter of fiscal 2026." — Q3 FY26 10-Q, Note 3, Income Taxes

Net effect on any mechanical five-year EPS series: FY25 is understated by about $5.18 a share and FY26 will be overstated by about $5.33. The company estimates a 40% tax benefit as its FY2026 annual effective rate. A screen that ranks on GAAP earnings growth will read this name backwards in both years.

Current trading (Q3 FY26 10-Q, quarter ended 2026-06-28)

Nine months: revenue $32,798m vs $33,013m (−0.7%); operating income $7,302m vs $9,437m (−22.6%).

QCT segment, Q3 FY26 vs Q3 FY25: handsets $5,086m vs $6,328m (−$1,242m, −19.6%); automotive $1,588m vs $984m (+61%); IoT $1,830m vs $1,681m (+9%). QCT EBT margin 26% vs 30%.

Management's stated cause of the handset decline is not share:

"lower handsets revenues, primarily due to lower chipset shipments to certain major OEMs (primarily driven by customers adjusting build plans to reduce their inventory levels as a result of the negative effects of recent memory supply constraints and related price increases)" — Q3 FY26 10-Q, Item 2, MD&A, Segment Results, QCT, "Third quarter 2026 vs. 2025"

Balance sheet and cash (Q3 FY26 10-Q)

Net debt $15,270m − $8,304m liquid = $6,966m. LTM operating income $10,220m (FY25 $12,355m less 9M FY25 $9,437m plus 9M FY26 $7,302m); adding LTM D&A of roughly $1.6bn (9M FY26 $1,202m) gives EBITDA near $11.8bn, so net debt/EBITDA ≈ 0.59x. Maturity wall: $1,991m current portion plus $498m of commercial paper. Interest rate swaps of $5.0bn notional convert fixed to floating on part of the long-term debt. There is no solvency question here.

Cash flow, nine months: OCF $8,405m vs $10,016m; capex $1,578m vs $785m (doubled); FCF $6,827m vs $9,231m (−26%). Buybacks $6,806m, dividends $2,868m — $9,674m returned against $6,827m of FCF, funded partly by drawing cash down $3,310m. Acquisitions took a further $1,573m of cash.

Share count: 1,125m (FY21) → 1,121m → 1,114m → 1,113m → 1,074m (FY25) → 1,057m at 2026-06-28. Nine-month movement: 25m issued (Alphawave, Modular), 42m repurchased at an average $162.05.

Verdict. Survives comfortably — 0.6x levered, $12.8bn of FY25 free cash flow, no wall, a shrinking share count. The kill test does not kill it. What the numbers do say is that FY26 is the year the trend turned: first revenue decline since FY23, operating income down 41% in the latest quarter, free cash flow down 26%, capex doubled, inventory up 28% into falling revenue, and the GAAP earnings line rendered uninterpretable in both directions by a tax item. Not a balance-sheet risk; an earnings-transition risk.

Analyst coverage count: not obtained (no source in this environment covers it) — thin, same as every prior slot.

4. S3 — Filings, gate-minimum

Read this run: FY25 10-K (Item 1A risk factors, Item 7 MD&A including the tax and segment discussion, Note 2 concentrations, PwC's report and critical audit matter); Q3 FY26 10-Q (MD&A delta, Notes 3, 4 and 8, balance sheet, cash flow); DEF 14A filed 2026-01-22 (ownership table, incentive design); the 8-Ks of 2026-09-08, 2026-07-31, 2026-06-24 and 2026-01-16; two transcripts; the 12-month Form 4 / Form 144 cadence; and the EDGAR full-text red-flag sweep.

The three biggest risks management itself discloses, with locators

1. The largest customer is designing Qualcomm out, and says so in Qualcomm's own words.

"In particular, we expect that Apple will increasingly use its own modem products, rather than our products, in its future devices, which will have a significant negative impact on our QCT revenues, results of operations and cash flows." — FY25 10-K, Item 1A, "Our business, particularly our semiconductor business, may suffer as a result of our customers vertically integrating (i.e., developing their own integrated circuit products)."

The same risk factor names Samsung and Xiaomi alongside Apple. It is not a hypothetical any more:

"Finally, as a result of our supply constraint, we now expect an acceleration in the step-down of Apple product revenues starting in the fourth fiscal quarter as our share for upcoming iPhone launch is expected to be materially lower than our prior estimate of 20%." — Akash Palkhiwala, Q3 FY26 earnings call, 2026-07-29, prepared remarks

and later in the Q&A, on the shape of it: "we are forecasting approximately 50% decline from September to December quarter. This obviously accelerates kind of the exit of Apple revenue out of our model." Bernstein's Stacy Rasgon put FY26 Apple revenue near $7.5bn from the guided non-handset growth rate; the CFO's reply was "I think that's a fair range of estimate."

2. Customer concentration, unnamed and rising at the top. From the issuer's own XBRL (srt:MajorCustomersAxis, via the 10-Q's Note 2 concentrations table):

Two customers were 43% of Q3 FY26 revenue. Customer (z) — 13% of FY25 — has fallen below the disclosure threshold in two of the last three quarters. The filings never name any of the three, so which counterparty left the table is not on the public record, and the pass does not guess.

3. China, named by management as an exacerbated concentration.

"A significant portion of our business is concentrated in China, and the risks of such concentration are exacerbated by U.S./China trade and national security tensions." — FY25 10-K, Item 1A, risk-factor heading

Asked at Communacopia whether US/China restrictions threaten the new data-centre business, the CFO answered: "For the engagements we have today, we're not restricted. And it's within the category of other companies. They're also providing solutions for China. So that is not a concern. this moment." That last phrase is the tell, and it is the company's own wording.

The 10-K's summary bullet list elevates five risks in total; the other two are "There are risks associated with our debt" and "Tax liabilities could adversely affect our results of operations" — the second of which the $5.7bn round trip has now demonstrated twice in three quarters.

Red-flag sweep — the one previous slots skipped

EDGAR full-text search over CIK 0000804328 (efts.sec.gov):

No NT 10-K appears anywhere in the submissions index. Auditor:

"/s/ PricewaterhouseCoopers LLP San Diego, California November 5, 2025 We have served as the Company’s auditor since 1985." — FY25 10-K, Report of Independent Registered Public Accounting Firm

One critical audit matter, and its subject matters for the Amazon warrant:

"Certain amounts recorded as a reduction to revenues for customer incentive arrangements are considered variable consideration and are included in the transaction price primarily based on estimating the most likely amount expected to be provided to the customer/licensee." — FY25 10-K, Critical Audit Matters, "Revenue Recognition – Qualcomm CDMA Technologies (QCT) Segment"

Equibles' GetGoingConcernStatus was not called — the full-text sweep answered it for free, and the budget went to the second transcript instead.

Insider pattern

3spread insiders QCOM --days 365: 154 Form 4 filings since 2025-09-22, on a near-monthly cadence (roughly the 2nd-4th and the 11th-21st of each month), each paired with a Form 144 — 16 Form 144s since 2026-03-26. That pattern is 10b5-1 plan selling plus vest-and-withhold mechanics. Transaction codes were not itemised this run (3spread returns them only per-filing, and the Equibles GetInsiderTransactions fallback was not spent), so this pass cannot state whether any open-market purchase occurred. Thin, and named as thin. What is solid is the proxy's own total: 18 officers and directors hold 573,776 shares between them, marked "*" for under 1%.

Executive and board changes

Equibles GetExecutiveChanges returns nothing newer than 2025-09-02 (Zico Kolter to the board) and 2025-08-25 (Patricia Grech as Chief Accounting Officer), and says six older filings are queued for re-read. It lists the 2026-01-16 8-K among its covered filings but extracted no change from it. Read directly, that filing says a director elected eight months earlier is leaving:

"(b) On January 13, 2026, Christopher D. Young, a member of the Board of Directors (the “Board”) of QUALCOMM Incorporated (the “Company”), informed the Board that due to the significant time commitment required in his new role as Chief Executive Officer and a member of the board of directors of Vertex, Inc., he will be stepping down from the Board effective as of the Company’s 2026 Annual Meeting of Stockholders."

Immaterial in itself. Recorded because it is an extraction miss in a connector the skill routes this check through — worth knowing for the remaining 100-plus names.

What management committed to, on the record

From the Q3 FY26 call (2026-07-29) and the Communacopia fireside (2026-09-08):

That is an unusually dense set of dated, falsifiable claims, which is what makes S9 easy to write.

Not read at gate-minimum

The Q2 FY26 earnings call (the skill's default second transcript — substituted, deliberately, for the post-warrant conference); the Investor Day deck, which management referenced in almost every answer and which is the source of the FY29 targets this scorecard leans on; the non-GAAP reconciliation, so the EPS model below is built on a GAAP basis rather than the company's; the FY24 10-K; older 10-Qs; the second and third transcripts; comment letters (none outstanding).

5. S5 — Ownership

Proxy 5% table (DEF 14A filed 2026-01-22, as of 2025-12-15; 1,062,916,212 shares outstanding):

Both 5% rows are stale on the proxy's own admission — a 2026 proxy citing a 2024 and a 2023 filing. This is the same Vanguard reporting discontinuity the APH pass documented on 2026-09-21: Vanguard's internal realignment changed what its entities report, so the beneficial-ownership record across issuers is unreliable in this period. Carried forward, not re-derived.

13F by manager, report date 2026-06-30 (3spread; 2,155 filers, 595,003,486 shares reported):

⚠️ Two data-quality warnings, both carried into the scorecard.

1. The script's own: 2,155 filers for 2026-06-30 against 2,700 for 2026-03-31, so the quarter is incompletely ingested. Prior holders with no filing in it yet — not exits — include both Vanguard entities (138,682,624 and 34,192,863 shares), Geode (29,734,061), Morgan Stanley (19,991,313), both UBS entities, Northern Trust, Franklin, FMR and JPMorgan. This is the fourth consecutive slot to hit the same lag.
2. This run's own: Sixth Street Partners Management Company appears as a NEW position of 22,310,326 shares / $4,122.7m — byte-identical to Charles Schwab Investment Management's row on the line above it. A credit manager with a $4.1bn new common-equity position matching an index manager's to the share is a duplication artefact, not a holding, and it is not carried as a fact. Worth a note for the tool: 3spread has now produced three distinct failure modes across four slots (empty ADR CUSIPs at ARM/OMAB/ASML, incomplete latest quarters, and now duplicated rows).

13D/G census: five filings since 2026-03-27, all Schedule 13G or 13G/A — passive. No 13D on the record at any date in the returned history. No activist, no strategic holder, no founder block.

Short interest (FINRA via Equibles, restated onto today's split basis):

Shorts have covered 33% of the position in three and a half months, from 50.4m to 33.7m shares, while daily volume collapsed from ~32m to ~10m shares. 3.2% of shares outstanding short, 3.3 days to cover. Equibles also carries a model estimate of ~35.2m for the unpublished 2026-09-15 settlement — an estimate, not a FINRA figure, and not used here.

Shelf and buyback. The 2024 S-3ASR (File No. 333-283035) is live and has been used twice in two months: to register the resale of the 17,826,566 Modular consideration shares, and — per the Amazon 8-K — expected to be used again for the warrant shares. A $20.0bn repurchase programme was announced 2026-03-17; $20.6bn remained authorised at 2026-06-28. No ATM programme was searched for; QCOM is not an ATM issuer and the shelf plus buyback is the relevant pair.

Who is on the other side: index funds and market makers. There is no concentrated holder to agree or disagree with, no activist, no insider block, and the shorts have been leaving. That is the honest answer and it is a thin one — the register tells you nothing about this name.

6. S8 — Valuation and sizing (⏳ OPEN (user))

Price $194.23 at the 2026-09-21 close (ROIC NASDAQ:QCOM), on 18,395,928 shares against a 20-session mean of 14.5m. The month to that close: $158.53 (2026-08-24) → $194.23, +22.5%, with the step change on 2026-09-08 (the Amazon announcement: gapped open to $180.40 on 26.1m shares) and a further +9.3% on 2026-09-21. The 2026-09-18 session traded 54.4m shares and reversed from $192.12 to close $177.72 — a September index-expiry pattern [background], not read as information.

ROIC's TTM and quarterly multiples and its earnings calendar both need a paid plan, and Equibles' GetValuationMultiples was not spent. No peer cross-section is in this file. The model below is built from the filings and the guidance above, on a GAAP basis excluding the tax valuation-allowance round trip, because the non-GAAP reconciliation was not read.

Out-year model (this run's, from company-guided components)

Build notes, so the numbers can be checked: FY26 revenue is nine-month actual $32,798m plus the Q4 guidance midpoint $10,100m. FY26 segment splits are nine-month actuals plus the Q4 guidance (handsets ~$5.2bn, automotive +60% YoY, IoT flat YoY). FY27 non-handset applies the CFO's own ">60%" to the FY26 non-handset base and holds data centre at the stated $5bn; FY27 handsets remove ~$6.5bn of Apple and add ~8% to Android. FY28 straight-lines data centre between the stated $5bn FY27 and $15bn FY29. Operating income uses QCT EBT margins of 24% (FY27) and 26% (FY28) — inside the company's 48–50% baseline gross margin less its own stated 1.5–2.0 pt data-centre drag, with non-GAAP opex scaled from the ~$2.7bn quarterly run rate. Unallocated corporate cost is held near the FY25 implied −$3.4bn. The 19% tax rate is the nine-month FY26 rate excluding the valuation-allowance benefit ($1,588m on $8,241m), not the 13% FDII rate, because the 10-K says the FDDEI benefit falls in FY26 and the rate becomes a permanent 14% in FY27.

The model's own conclusion is that FY27 is flat. The Apple exit and the data-centre operating expense arrive in the same year, and the company's language concedes it: non-handset growth "will replace the entire Apple product revenue within the year" — replace, not exceed. The earnings inflection, if the plan works, is FY28.

Bear / base / bull on FY28

Base case ~$175 against $194.23 — about 10% below the traded price. Proposed buy price and proposed size are ⏳ OPEN (user) and always were. On liquidity: ADV ~$2.55bn/day (mean of close × volume over the 20 sessions to 2026-09-21; ~$2.17bn excluding the 2026-09-18 expiry session), mean 14.5m shares. No size the owner would take is constrained by this book.

Unquantified in the model: the Amazon warrant. If Qualcomm follows the accounting of two names already in this workspace — AMD's customer warrants and Cerebras' penny warrant, the latter amortised as a reduction of revenue — then some part of the 25m shares' grant-date fair value is charged against data-centre revenue as it vests. At $194.23 the full 25m carries roughly $4.9bn of market value and about $0.8bn of intrinsic value over the strike. Nothing in any filing says how it will be treated, and this file does not guess; it names the question and dates the answer (FY2026 10-K, due ~2026-11-05).

7. S9 — Monitoring

• Next earnings: 2026-11-04 18:45 UTC, Q4 FY26 and the FY26 close — scheduled, from Equibles ListInvestorEvents (event id 98ad8620-b551-4baa-b7e5-009dfaeddea7). This is a real calendar date rather than the filing-cadence derivation the previous four slots had to use.
• Filing feed: the Tier 0 EDGAR sweep already covers cik 804328. Two filing types deserve attention beyond the periodic reports: further Item 3.02 8-Ks (the warrant precedent is now set, and the CFO called Amazon "a first of many"), and 424B7 resale supplements off the 2024 shelf, which is how the Modular sellers and, prospectively, Amazon get liquidity.
• Leading indicator between quarters: the QCT gross margin against the stated 48–50% baseline. It is the single number that carries the two live questions at once — whether the double-digit price increases are actually sticking against memory-driven input inflation, and how much the low-margin custom-silicon revenue dilutes the mix as it ramps. Management has pre-committed to the range, to the direction and to a rough timetable, so each quarterly print scores itself.
• Trigger that converts this name to a buy: the first reported data-centre revenue, at a disclosed gross margin, in the Q1 FY27 release (early February 2027) — with QCT gross margin back above 46%. The whole thesis is one unproven claim: that a mobile chipmaker can enter the data centre. Management has made that claim falsifiable on a date, and paid Amazon 25m shares to make the first proof point happen. One quarter of actual, margin-disclosed data-centre revenue converts the story into a number; its absence, or a margin so thin that $15bn of FY29 revenue implies no profit, kills it.

---

Access notes for this run

• SEC EDGAR (rung 1): reachable throughout with User-Agent: ddboard-research gorkem@celebico.com. Submissions index, XBRL company concepts, Archives documents and efts.sec.gov full-text search all answered. The full-text sweep cost nothing and closed a gate three previous slots left open — worth doing by default.
• 3spread (rung 4b): reachable, key present, HTTP 200. Used for 13F, 13D/G, Form 4 and Form 144. Three caveats now recorded (above).
• Equibles (rung 4): reachable. 7 calls of the 25-per-run cap — Q3 FY26 transcript, customer concentration (3 filings), short interest, buybacks, executive changes, investor-event list, and the Communacopia conference transcript. Well inside budget because the slot held one name.
• ROIC.ai (rung 5): prices reachable on the free tier. get_valuation_multiples with period=ttm and list_earnings_calendar both require a paid plan — the fifth and sixth slots in a row to hit this. The earnings date came from Equibles instead, which is the better route.
• EDINET DB / eu_data: not needed; QCOM is sec_domestic.
• WebSearch: not used this run. S6 was carried forward, and the budget went to primary sources.