A licensing-fortified cash machine being paid ~20x forward NOT to lose a $7B Apple leg it is already losing — the rerate only comes if Snapdragon-X PCs + custom data-center silicon replace Apple faster than handsets fade, and the tape (rev −3% YoY, Q3 guide −7%) says it isn't there yet.
No Friday close is on the record for QCOM yet. The weekly job prices the covered universe; a name it cannot price is listed as missing rather than estimated.
| Date | Type | What happened | Source |
|---|---|---|---|
| 2026-08-10 | editorial note | Capex figure revised: $1,082M → $496MCapex moved from $1,082M (deep-dive-2026-06-21.md) to $496M (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note |
| Margin figure revised: 21.8% → 69%Margin moved from 21.8% (deep-dive-2026-06-21.md) to 69% (deep-dive-2026-08-10.md). |
| dossier |
| 2026-08-10 | editorial note | Revenue figure revised: $38,367M → $40BRevenue moved from $38,367M (deep-dive-2026-06-21.md) to $40B (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note | Verdict changed: A licensing-fortified cash machine being paid ~20x forward NOT to lose a $7B Apple leg it is already losing — the rerate only comes if Snapdragon-X PCs + custom data-center silicon r…Before (deep-dive-2026-06-21.md): A licensing-fortified cash machine being paid ~20x forward NOT to lose a $7B Apple leg it is already losing — the rerate only comes if Snapdragon-X PCs + custom data-center silicon replace Apple faster than handsets fade, and the tape (rev −3% YoY, Q3 guide −7%) says it isn't there yet. After (deep-dive-2026-08-10.md): In seven weeks the story got a number and the tape got worse — management put $5B of FY27 data-center revenue and a $40B FY29 non-handset target on the record, then disclosed handsets −20%, QTL negative for the first time, gross margin below its own range, and a Data Center unit losing $214M a quarter; the stock fell 25% through the prior $180 target, which is the market paying for the trough instead of the pivot. | dossier |
The verdict
In seven weeks the story got a number and the tape got worse — management put $5B of FY27 data-center revenue and a $40B FY29 non-handset target on the record, then disclosed handsets −20%, QTL negative for the first time, gross margin below its own range, and a Data Center unit losing $214M a quarter; the stock fell 25% through the prior $180 target, which is the market paying for the trough instead of the pivot.
Primary sources
SEC filings
Earnings calls
Source documents — open to read in full
The structural thesis from 2026-06-21 holds, and it held in the direction the bear side of it pointed. The prior dive's core claim was a timing mismatch — Apple's ~$7B exits before automotive and data center can fill it, revenue was already declining, and ~20x forward was full rather than cheap for a name guiding revenue down. Every element of that has been confirmed by the tape, and two of them got worse than the prior dive assumed (handsets −20% not −13%; Apple's exit accelerated). The prior dive's contrarian read — that the market was too optimistic, not too pessimistic — has been paid: the stock fell 25% and through the target.
What did not hold is the prior dive's characterisation of the data-center bet as narrative ("multiple billions," ~$0 today). It is now a dated, quantified, customer-named plan with wafers in production, purchase orders from two hyperscalers, a taped-out memory-integrated accelerator (HBC Gen 1), a named CPU customer (Meta), and an acquired software layer (Modular). That is a genuine upgrade in evidence quality. It is also now visible in the segment tables as a loss: the nonreportable segments (Data Center + QGOV) ran $165M revenue on −$214M EBT in Q3, and $353M on −$419M across nine months.
Two moat claims from the prior dive need amending: the QTL annuity is no longer flat-to-up (see delta 3), and the "fabless, capital-light" frame is stretching — Capital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. is running at 2x the prior year and inventory is a deliberate strategic build (Lens 10).
The two-business frame (QCT chips + QTL patents) still holds, but a third leg is now separately visible in the filings. Qualcomm reports three segments — QCT, QTL, QSI — plus nonreportable segments: QGOV (government) and Data Center. Data Center has stopped being a slide and started being a P&L line, and the line is negative.
Management's own restatement of the company, from the Q3 call, is three-dimensional: expand into the data center with four product lines; drive agentic and physical AI compute everywhere; and expand beyond silicon into full-stack software and platform solutions. The third dimension is new since the prior dive and is what Modular buys.
The data-center product ladder is now dated: connectivity in FY2026 (Alphawave SerDes), custom silicon and AI accelerators in FY2027, server-class CPUs in FY2028. The differentiated piece is HBC (high-bandwidth compute) — integrating compute directly with high-density memory to attack the memory-bandwidth bottleneck, Gen 1 tape-out complete, silicon demonstration "in the coming quarters," first solution launching mid-2027.
Plain terms, updated: Qualcomm still makes the brains for premium Android phones and taxes the smartphone industry through patents. What is new is that it has committed, publicly and with dates, to becoming an inference-infrastructure company — silicon, memory-integrated accelerators, server CPUs, and now an open software stack aimed squarely at the position CUDA occupies.
One amendment that matters, and it is a chokepoint the prior dive did not name: memory. Amon describes the industry as running at pandemic-style 100% utilisation across wafers, assembly, and test, with shortages and price increases everywhere. Memory is now a demand destroyer for Qualcomm's customers rather than a cost line for Qualcomm itself — OEMs are cutting builds because DRAM/NAND has eaten their bill of materials. It is also, via HBC, the bottleneck Qualcomm is trying to sell into. Same input, two opposite signs.
The QTL moat took its first visible dent. Licensing revenue −3.0% YoY and EBT margin down two points to 69%. More important than the quarter is the structure management disclosed: royalties are capped on device ASP, so when handset prices rise on memory inflation, QTL captures the increase only below the cap and nothing above it. That is an asymmetry the prior dive did not price: QTL participates fully in unit declines and only partially in price increases. The renewal cliff (FY2027–2031) is unchanged and still the load-bearing tail risk.
The purchased software moat is real but unproven. Modular gives Qualcomm Mojo, MAX and a hardware-agnostic inference runtime plus the engineer who built Swift and LLVM. The strategic logic is coherent: if inference disaggregates across data center, on-prem edge, and device — which is Amon's stated thesis — then the layer that abstracts heterogeneous hardware is worth owning, and it is the one place NVIDIA's moat is software rather than silicon. Whether Qualcomm can run an open-source developer platform is an entirely different competence from shipping SoCs, and there is no evidence yet either way.
The QCT product moat is where the prior dive left it, minus a customer. Samsung remains strong — Snapdragon powers ~70% of Samsung flagships. Apple is exiting faster than telegraphed. MediaTek is unchanged as a mid- and increasingly high-end threat.
Net moat verdict, amended: a licensing moat that has now been observed eroding rather than merely projected to; a narrowing product moat; and a newly purchased software position whose value is entirely forward-looking.
QCT end-markets, Q3 FY2026 (3 mo. ended 2026-06-28):
| QCT end-market | Q3 FY26 | Q3 FY25 | YoY | Prior quarter YoY |
|---|---|---|---|---|
| Handsets | $5,086M | $6,328M | −19.6% | −13.1% |
| Automotive | $1,588M | $984M | +61.4% | +38.3% |
| IoT | $1,830M | $1,681M | +8.9% | +9.2% |
| Total QCT | $8,504M | $8,993M | −5.4% | −4.2% |
| QCT EBT | $2,192M | $2,671M | −17.9% | — |
| QCT EBT margin | 26% | 30% | −4 pts | — |
QTL $1,278M vs $1,318M, −3.0%; EBT $881M, margin 69% vs 71%.
Nonreportable (Data Center + QGOV) $165M revenue vs $54M; EBT −$214M vs −$10M. Nine months: $353M revenue on −$419M EBT. The revenue growth here is Alphawave consolidation, not organic silicon sales — the filing attributes the +$88M YoY equipment-and-services increase directly to the acquisition.
QSI $0 revenue, EBT +$768M — entirely investment gains from IPOs of QSI equity holdings. This is the single most important line for reading the headline: pre-tax income of $2,462M includes $1,014M of investment and other income. Strip it and pre-tax income is ~$1,448M against ~$2,594M a year earlier — −44%, versus the −17% the headline pre-tax number shows.
Read: the two designated offsets are behaving very differently. Automotive is not just delivering, it is accelerating (+61%, a record, with the run-rate target raised from $6B to ~$7B exiting FY2026 and a decade-long BMW lead-supplier win). IoT is growing single digits and Q4 is guided flat YoY because memory constraints are hitting tablets and consumer devices. And the handset base fell $1.24B YoY in a single quarter — more than Automotive's entire quarterly revenue.
Headline (GAAP):
| Q3 FY26 | Q3 FY25 | Δ | |
|---|---|---|---|
| Revenue | $9,947M | $10,365M | −4.0% |
| Cost of revenues | $4,670M | $4,606M | +1.4% |
| Gross margin | 53% | 56% | −3 pts |
| R&D | $2,607M | $2,226M | +17.1% |
| SG&A | $976M | $771M | +26.6% |
| Other (restructuring/severance) | $68M | $0M | new |
| Operating income | $1,626M | $2,762M | −41.1% |
| Operating margin | 16.3% | 26.6% | −10.3 pts |
| Investment and other income, net | $1,014M | $358M | +$656M |
| Income before taxes | $2,462M | $2,952M | −16.6% |
| Effective tax rate | 19% | 10% | — |
| Net income | $2,002M | $2,666M | −24.9% |
| GAAP diluted EPS | $1.87 | $2.43 | −23.0% |
| Diluted shares | 1,069M | 1,099M | −2.7% |
Non-GAAP EPS $2.21 vs $2.23 consensus — a 2c miss on revenue of $9.95B versus ~$9.69B consensus, a beat. Management called revenue "at the high end of our guidance".
The number that matters is operating margin: 16.3%, down 10.3 points. That is not an Apple story or a tax story — it is a gross-margin story (−3 pts) compounded by opex growing 18% into a revenue decline. Amon conceded QCT gross margins "will be slightly below our historical range," which he separately defined as 48–50% for the baseline business.
Guidance — Q4 FY2026:
Balance-sheet flags — the prior dive's yellow flag went amber:
Market reaction: shares fell ~4.4% the day after the release on the light Q4 guide, and have continued lower to $164.54.
Unusual vs own history: the GAAP-distortion problem the prior dive flagged is resolved and it resolved differently than the prior dive recorded. The 10-Q states the fiscal-2025 valuation allowance and the fiscal-2026 Q2 release were both $5.7 billion (the tax reconciliation line reads $5,724M). The prior dive cited $5,138M and a −230% Q2 rate; the 10-Q is the authority and the prior figure should be treated as superseded. The nine-month FY2026 effective rate is −50%, full-year FY2026 is estimated at a 40% benefit, and the Q3 quarter itself ran a clean 19%. From here, GAAP is readable again.
The prior dive's Lens 6 was `` because transcripts/ was empty. It is not empty any more — the FY2026 Q3 call is on disk as transcripts/2026-q2.md, sourced from Qualcomm's own IR-hosted transcript.
Tracking the framing across four calls:
| Call | Centre of gravity | Tell |
|---|---|---|
| Q4 FY2025 / Q1 FY2026 | Record revenue; "data center multiple billions ahead"; edge AI wins the AI race | Peak optimism, no dates, no customers |
| Q2 FY2026 | "Handset bottom"; custom silicon with a hyperscaler, December-quarter shipments | Pivot begins; still anonymous customers |
| Investor Day 2026-06-24 | $40B non-handset FY29; $15B Data Center FY29; Meta named; Dragonfly C1000; AI200 | Narrative converts to dated, customer-named commitments |
| Q3 FY2026 (2026-07-29) | Execution phase. $5B DC revenue FY27. HBC Gen 1 taped out. Modular closed. BMW lead supplier. And: memory pressure, GM below range, Apple accelerating out | Confidence on the plan, candour on the trough |
The sentiment shift is not optimism-to-pessimism — it is abstraction-to-specificity in both directions. Management got more specific about the upside (dollar targets, named customers, tape-out milestones, ship dates) and simultaneously more specific about the damage (handset market down low teens in FY2027, QCT Android revenue −20%, a stated ">$1.50" EPS impact, gross margin below range, Apple below 20% share). That is a more honest call than the prior two, and it is why the stock fell on it.
The most self-aware line on the call is Amon acknowledging that investors want proof points before believing a new data-center entrant, and welcoming the challenge. He is right that this is the crux. What is being asked of the market is a two-year bridge on trust, and management has now put falsifiable numbers on both ends of it — December-quarter data-center revenue, HBC silicon demonstrations, $5B in FY2027.
What they stopped saying: the phrase "handset bottom" as a company-wide claim. It has narrowed to a specific, checkable one — Chinese OEM handset revenue bottomed in the June quarter and returns to double-digit sequential growth in September, because channel inventory has thinned to the point OEMs cannot draw down further. That is a far better claim than the earlier one: it names a mechanism and it can be checked in 90 days.
One sourced update, because the price moved 25%: QCOM trades at 19.53x trailing P/E and 18.01x forward P/E, market cap $175.74B, consensus PT $196.27 on a "Hold" from 37 analysts. Conflict surfaced, not resolved: that 18.01x forward multiple implies forward EPS of $9.14, which is below both the FY2026 non-GAAP path built in Lens 11 ($10.42) and any FY2027 estimate I can construct — the provider's "forward" basis is not disclosed and may be GAAP or a different fiscal year. On the FY2026 non-GAAP path, the multiple is ~15.8x. Peer multiples were not re-sourced this pass (web-search budget exhausted) — the prior dive's peer table is stale as of 2026-06 and should not be quoted as current.
Moves since the boundary, with causes:
| Date | Move | Cause |
|---|---|---|
| 2026-06-24 | +16% intraday/after-hours | Investor Day: $40B non-handset FY29, $15B Data Center, Meta named |
| 2026-07-29/30 | −4.4% | Q3 print: revenue beat, EPS 2c miss, Q4 guide light |
| 2026-07-31 – 2026-08-01 | −2.6% | UBS PT $190→$170; Susquehanna $190→$160 Neutral |
| 2026-08-07 | −3.2%, then +4.66% intraday recovery to $167.86 | No single sourced cause — low-quality outlets; treat the intraday attribution as unsourced |
| Net 2026-06-21 → 2026-08-10 | ~−25% | ~$220 → $164.54 |
What this reveals, and it is a change from the prior dive. The prior dive concluded QCOM trades on narrative transitions rather than quarterly EPS. The last seven weeks refine that: the stock got the biggest possible narrative upgrade (a doubled long-range target with a named hyperscaler customer) and gave the entire move back plus more within five weeks, on a quarter where revenue beat. The market is now trading the near-term earnings power and the gross-margin line, not the FY2029 target. That is a regime change worth noting — the AI-optionality bid that carried QCOM to $259.92 in the last twelve months has stopped paying for promises.
Live catalysts from here (all dated by management, which is what makes them useful):
Amon's record and archetype are as the prior dive left them. Three things changed.
Capital allocation is unchanged in policy and increasingly stretched in funding. New $20.0B repurchase authorisation announced 2026-03-17 on top of $2.1B remaining; $20.6B authorised at 2026-06-28; 42M shares repurchased for $6.8B over nine months, share count 1,074M → 1,057M; dividends $0.92/quarter. The honest critique is not the policy but the funding mix: nine-month returns of $9.67B against $6.83B of FCF, with $2.49B of new commercial paper appearing on the balance sheet. Buying back stock at $200+ during a quarter that ended with the stock at $165 is also, mechanically, poor timing — though that is hindsight, not a governance flag.
Verdict: a competent operator who has now staked his tenure on a specific, dated, capital-intensive bet, hired well for it, and told investors honestly that they should demand proof. The strategy risk the prior dive identified is unchanged and larger.
our figures is now populated (three quarters) and every figure below is cited to the filing.
CORRECTION TO THE PRIOR DIVE. The valuation-allowance figures were $5.7 billion in both directions — the FY2025 establishment and the Q2 FY2026 release — with the tax reconciliation line reading $5,724M. The prior dive's $5,138M and "−230% effective rate" are superseded. Nine-month FY2026 effective rate: −50%. Estimated full-year FY2026: 40% benefit. Q3 standalone: 19%. Cause of the release: IRS Notice 2026-07 allowed CAMT reduction by previously capitalised domestic R&D, so Qualcomm no longer expects to be subject to CAMT. Non-cash, cosmetic, now behind us — but note the cash consequence is real and negative in the near term: nine-month income tax payments exceeded the provision by $5,550M, including a final $663M repatriation-tax instalment.
Inventory: the flag escalates. $8,379M, +28.4% since FYE2025, against falling revenue. Raw materials $682M vs $336M (+103%), WIP $5,005M vs $3,985M, finished goods $2,692M vs $2,205M. Implied DIO ~163 days vs ~129 at FYE2025. Management's defence is explicit and worth weighing: Amon frames inventory as a strategic advantage in a shortage, comparing conditions to the pandemic when Qualcomm's supply position won share. That defence has a track record behind it — it is the specific thing Amon is credited for in 2021. It is also exactly what a company would say while accumulating obsolescence risk into a market management itself expects to be down low teens next year. Both can be true; the check is the Q4 and Q1 write-down lines.
NEW: capital returns exceed free cash flow, funded with commercial paper. Nine months: $9,674M returned versus $6,827M FCF (~1.42x); Q3 alone ~4.6x. Short-term debt went $0 → $2,489M. Net debt up ~50% to ~$6,966M. This is not distress — $8.3B of liquidity and $12.8B of long-term debt against $27.7B of equity is comfortable — but it is a change of financing character that the prior dive's "impeccable capital allocation" note did not contemplate.
NEW: goodwill concentration in a loss-making segment. Goodwill $14,274M, ~52% of book equity, with $2,210M of the Alphawave goodwill allocated specifically to the Data Center operating segment and a further $76M from the seven bolt-ons. Modular adds several billion more in Q4 FY2026 (purchase-price allocation not yet disclosed — "not practicable" given timing). That goodwill sits on a segment currently running −$214M EBT per quarter on $165M of revenue. Impairment is no longer a hypothetical to "watch" — it is the single balance-sheet item most exposed if the FY2027 $5B data-center number slips.
NEW: non-GAAP quality — cash bonus converted to equity. See Lens 9.3. Nine-month SBC +21.7%; the increase explicitly includes replacing FY2026 and FY2027 cash incentives with a two-year equity award. Additionally, ~$700M of Modular executive shares are subject to a four-year service requirement and will be recognised as compensation expense, as was a portion of the $746M Alphawave Exchangeable Shares. Non-GAAP EPS in FY2026–FY2027 is being measured against a cost base from which a bonus pool has been removed.
NEW: customer concentration shifted. Customers ≥10% of revenue, three months ended 2026-06-28: (x) 23%, (y) 20%, (z) below 10% — versus (x) 18%, (y) 21%, (z) 13% a year earlier. A third ≥10% customer dropped out of the disclosure, consistent with the Apple ramp-down, while the largest grew five points. Concentration in the remaining two increased.
Headline pre-tax income flattered by investment gains. $1,014M of investment and other income inside $2,462M pre-tax, driven by QSI IPO gains. Ex that, pre-tax fell ~44% YoY versus the −17% headline. Not a red flag in itself — these are real, realised gains — but it is a non-recurring, non-operating contribution to a quarter that looks better than it is.
Receivables: $4,668M vs $4,315M at FYE2025 on lower revenue — a mild build, not a channel-stuffing signal, but no longer clean of it.
Regulatory findings:
total_sec_findings: 0 — unchanged.Net: still clean on fraud and accounting (zero SEC findings, tax noise now resolved and readable). The forensic story has shifted from cosmetic to substantive: inventory, goodwill in a loss-making segment, returns exceeding cash flow, and a compensation change that improves the metric management is asking to be judged on. None of these is a scandal. Together they are the balance-sheet signature of a company spending hard through a trough.
MODEL INTEGRITY — read before using any number below. our model was re-run against the newly populated CSVs and wrote model.xlsx. our model reports:
Everything below is `` with the arithmetic shown.
FY2026 anchor — this is the biggest single change from the prior dive.
Non-GAAP EPS by quarter: Q1 $3.41 + Q2 $2.65 + Q3 $2.21 actual + Q4 guide midpoint $2.15 = ~$10.42.
The prior dive carried a base of ~$12.3 and cited a ~$12.6 consensus. The realised path is ~15% below that. Anyone still anchored on a $12+ FY2026 is using a number the company has now guided away from.
FY2026 revenue: 9M actual $32,798M + Q4 midpoint $10,100M = ~$42.9B.
FY2027 building blocks — all management-stated:
Reconciling the revenue bridge — and a tension worth naming. Building FY2027 from those components:
| Line | FY2026E | FY2027E | Basis |
|---|---|---|---|
| Non-handset QCT (Auto + IoT + DC) | ~$13.1B | ~$20.9B | +60% per guidance; FY26 build: 9M Auto $4,015M + Q4 ~$1,685M; 9M IoT $5,244M + Q4 ~$1,806M; DC ~$0.3B |
| Android handsets | ~$16.6B | ~$13.3B | FY26 handsets ~$24.1B less ~$7.5B Apple; then −20% |
| Apple product | ~$7.5B | ~$1.8B | management: below the prior "little over $2B" |
| QTL | ~$5.6B | ~$5.6B | flat; 9M $4,252M + Q4 mid $1.3B |
| Total | ~$42.9B | ~$41.6B | −3% |
Management says it expects top-line growth in fiscal 2027. My build of management's own stated components lands at roughly flat to −3%. The gap is not resolvable from the disclosed numbers — it turns on whether the Android −20% and the market "down low teens" are the same assumption or stacked, and on how much of the double-digit price increase flows to revenue. Surfaced, not resolved. It is also the single most checkable claim in the whole refresh.
Three-year non-GAAP EPS path:
| Scenario | FY2026E | FY2027E | FY2028E | Logic |
|---|---|---|---|---|
| Bear | ~$10.4 | ~$9.0 | ~$8.5 | Memory pressure runs through FY2027; price increases stick only partially; handset market worse than low teens; data center lands below $5B and its DilutionIssuing new shares, so each existing share owns a smaller slice of the same company. bites without the volume; opex already committed. Apple gone, QTL flat-to-down on a renewal. |
| Base | ~$10.4 | ~$10.5 | ~$12.0 | Gross-profit bridge: non-handset +$7.9B revenue at ~46% blended GM (Auto/IoT ~50%, DC ~35%) ≈ +$3.6B; Apple −$5.7B at ~40% GM ≈ −$2.3B; Android −$3.3B at ~50% ≈ −$1.6B; price increases recover ~$1.6B. Net gross profit ≈ +$1.3B, offset by ~$1.0–1.5B opex growth ⇒ pre-tax roughly flat; ~13–14% tax; |
| Bull | ~$10.4 | ~$12.0 | ~$15.5 | Price increases fully restore 48–50% GM inside two quarters; data center hits $5B and pulls merchant accelerator wins off the HBC silicon demo; Android recovers faster than low-teens-down as the agentic upgrade cycle lands; the >$1.50 memory EPS drag reverses. |
The call: base case is ~flat EPS at ~$10.5 in FY2027, one full turn below where the prior dive's base sat, and the first real re-acceleration is FY2028. This is unchanged in shape from the prior dive — a two-year transition trough — but the level dropped ~15% and the trough got deeper before it got better. The offsetting new fact is that FY2028+ upside now has dates, customers and purchase orders attached to it rather than adjectives.
Falsifier seeds (not logged — unattended run):
Bull case (materially stronger than at the prior dive). You are buying a 69–72%-margin patent annuity and a #1-track automotive franchise at ~15.8x trough earnings, with a management-quantified >$1.50 of EPS sitting on the other side of a memory cycle that is a cost shock, not a demand collapse. The diversification is no longer a slide: Automotive +61% YoY with the run-rate raised to ~$7B and a decade-long BMW lead-supplier win plus Stellantis; industrial design-win pipeline >$7B with $3.5B secured this year. The data-center bet has converted from narrative to schedule — two hyperscaler custom-silicon programmes with POs and wafers in production, revenue starting in the December quarter, $5B in FY2027, Meta signed for CPUs, HBC Gen 1 taped out attacking the memory-bandwidth bottleneck, and Modular + Chris Lattner giving Qualcomm the one asset that could contest CUDA's position — an open, hardware-agnostic inference stack. The prior dive's biggest criticism (data center is $0 and unproven) has been answered with more specificity than any competitor entering this market has offered.
Bear case — three permanent-impairment risks, updated.
Pre-mortem (18 months out, thesis broke). It is early 2028. Memory prices never normalised, so the handset market that was "down low teens" in FY2027 was down high teens and stayed there; Qualcomm's double-digit price increases were partly absorbed by OEMs trading down within the premium tier rather than passed to consumers, so gross margin never returned to 48–50%. Data center generated $2.5B in FY2027 instead of $5B — the two ASIC programmes shipped but the merchant accelerator did not convert off the HBC demo, and the CPU slipped from FY2028. Modular's open-stack ambition ran into the fact that developers do not switch runtimes because a chip vendor asks them to. A large QTL licensee renewed at a lower blended rate. Goodwill from Alphawave and Modular was written down. EPS is ~$8.50, the multiple compressed to 12x on "structural decliner with a failed pivot," and the stock is near the $121.99 twelve-month low.
Is the multiple too high? At ~15.8x FY2026E and ~15.7x the FY2027 base, this is no longer the "full, not cheap" setup the prior dive described. The 25% de-rating has moved the asymmetry. It is now a genuinely two-sided price: cheap if the memory shock is cyclical and data center hits half its target; expensive if either the QTL erosion is structural or the pivot slips a year.
Contrarian view (what the market refuses to see). The prior dive's contrarian read was that the market was too optimistic. That trade paid. The contrarian read now inverts, and the reason is specific rather than sentimental: the market is treating the memory shock and the Apple exit as one story when they are opposite in duration. Apple is permanent and fully telegraphed. The memory-driven handset cut is a cost shock with a stated, quantified reversal value of >$1.50 in EPS — and the market has repriced the stock as if both are permanent. Meanwhile it is assigning close to zero to a data-center business with signed hyperscaler POs, because it has been trained by Centriq. The asymmetry may have flipped to the upside — but only for someone willing to underwrite two years of flat earnings and a balance sheet spending through them.
Dismantling the newly strengthened bull.
The fifteen questions there remain live. Five to add, arising from this quarter:
Every dossier we have written on Qualcomm, newest first, including where a later one corrected an earlier one.
In seven weeks the story got a number and the tape got worse
CorrectionCORRECTION TO THE PRIOR DIVE. The valuation-allowance figures were $5.7 billion in both directions — the FY2025 establishment and the Q2 FY2026 release — with the tax reconciliation line reading $5,724M.
A licensing-fortified cash machine being paid ~20x forward NOT to lose a $7B Apple leg it is already losing
Covered in the Knowledge Base
Hardware — Memory & Compute Architecture
| Industry | Semiconductors |
| Size | Public Company |
Where Qualcomm sits against the other names we cover on this beat. Each line is that company’s verdict, not a summary of it.
The Q2 beat was not a peak — July revenue accelerated to +44.7% YoY, the board appropriated a single-tranche US$29.4B of capacity capital, and H1 cape…
The Q2 print did the one thing the bear case could not survive
Cash $4.7B
The thesis got WIDER and the price got WORSE.
Cash $19.6B
The de-rate the June dossier warned about arrived early and for the exact reason it named
Cash $1.4B
LONG — MEDIUM (conviction DOWN one notch from the prior MEDIUM-HIGH, on wider risk, not a worse business).
Cash $25.0B