QCOM · Analyze — Financebot

Full analysis · 2026-08-04 · $162.67

Verdict: HOLD — [5.5]. Do not add here. The business is sound and the balance sheet is fine, but this is a company whose revenue has not grown in four years, whose operating margin fell from 26.6% to 16.3% in four quarters, and whose largest customer is leaving faster than management modelled. The apparent cheapness — "15.8x forward" — is an artifact of a stale earnings base. On management's own Q4 guidance the stock trades at 18.9x the exit run-rate, above its entire FY2025 valuation range on both EV/EBIT and P/FCF. Fair value $120-160; spot sits just above the top.

The licensing annuity is more durable than the drawdown implies — Samsung is signed through 2030 with 6G explicitly pre-covered, and renewal contracts carry binding-arbitration clauses that make a royalty holiday structurally impossible. That is the floor, and it is why this is a 5.5 rather than a 4.5. It is also capped: the royalty applies only to the first $400 of device price, so the double-digit price increases coming in September help QCT and do nothing for QTL.

0. Knowledge check

python .mcp/kb.py find QCOM returned no matches — no prior note or report on this name. Three live notes governed the framing and are cited rather than re-derived:

Note	Bearing on this analysis
`Market/regime.md`	Act II (the AI-capex rout) reversed in four sessions, Jul 30 – Aug 4. SOX +16.5%, QCOM +7.32% on 8/04 alone. Act II drawdowns were sentiment de-ratings, "never priced-in value, so never durable entries either."
`Playbook/pattern-ai-levered-fields-trade-above-own-band`	Every AI-adjacent field swept in July still traded above its own historical band despite large drawdowns. QCOM confirms it again below.
`Playbook/pitfall-multiple-trim-inverts-on-peak-cycle-cyclicals`	Checked before setting the trim. QCOM is not in the trap — see §6.4.

Two new pitfall notes were written out of this analysis (§8).

1. Fundamentals

1.1 The four-quarter deterioration — the central fact

Everything else in this report is commentary on this table.

	FQ3 FY25	FQ4 FY25	FQ1 FY26	FQ2 FY26	FQ3 FY26	YoY
Revenue	$10,365M	$11,271M	$12,252M	$10,599M	$9,947M	−4.0%
Gross profit	$5,759M	$6,237M	$6,684M	$5,699M	$5,277M	−8.4%
Gross margin	55.6%	55.3%	54.6%	53.8%	53.1%	−2.5pt
Operating income	$2,762M	$2,918M	$3,366M	$2,309M	$1,626M	−41.1%
Operating margin	26.6%	25.9%	27.5%	21.8%	16.3%	−10.3pt
R&D	$2,226M	$2,370M	$2,453M	$2,463M	$2,607M	+17.1%
Operating expense	$2,997M	$3,280M	$3,318M	$3,390M	$3,651M	+21.8%
GAAP diluted EPS	$2.43	−$2.89	$2.78	$6.88	$1.87	−23%
Non-GAAP diluted EPS	$2.77	—	—	—	$2.21	−20%
Free cash flow	$2,581M	$3,589M	$4,416M	$1,916M	$495M	−80.8%

Revenue fell 4%. Operating expense rose 22%. That gap is the whole story: Qualcomm is paying for its diversification out of the current P&L while the business being diversified away from shrinks faster than the replacements grow. R&D at $2.6B/quarter is 26.2% of revenue, up from 21.5% a year ago.

1.2 Segments — the barbell

Segment (FQ3 FY26)	Revenue	YoY	EBT	EBT margin
QCT — Handsets	$5,086M	−20%
QCT — Automotive	$1,588M	+61% (record)
QCT — IoT	$1,830M	+9%
QCT total	$8,504M	−5%	$2,192M	26% (from 30%)
QTL (licensing)	$1,278M	−3%	$881M	69% (from 71%)

For the full fiscal year (FY2025 10-K): QCT $38,367M (+16%) — handsets $27,793M, IoT $6,617M (+22%), automotive $3,957M (+36%) — at a 30% QCT EBT margin; QTL $5,582M at a 72% EBT margin ($4,043M).

Automotive has now posted 23 consecutive quarters of double-digit YoY growth and the exit-FY26 run-rate target was raised to ~$7B (from $6B a quarter earlier). That is a real, compounding, design-win-backed business. It is also 16% of revenue against handsets' 51% — the good business is not yet big enough to carry the company.

One number in the FY2025 10-K's QCT discussion deserves to be pulled out because it is the durability tell for the handset franchise: of FY2025's handset revenue growth, $2.5B came from higher revenue per chipset and only $423M from higher shipments. Handsets grow by content, not by volume — which works beautifully until the mix trades down, which is exactly what Palkhiwala reported this quarter.

1.3 Four years of no growth

	FY2022	FY2023	FY2024	FY2025	TTM
Revenue	$44,200M	$35,820M	$38,960M	$44,280M	$44,069M
Gross margin	57.9%	55.7%	56.2%	55.4%	54.2%
Operating income	$15,860M	$8,650M	$10,250M	$12,390M	$10,219M
Operating margin	35.9%	24.1%	26.3%	28.0%	23.2%
Net income (GAAP)	$12,940M	$7,230M	$10,140M	$5,540M ⚠️	$9,259M ⚠️
Free cash flow	$6,830M	$9,850M	$11,160M	$12,820M	$10,416M
Diluted shares	1,140M	1,130M	1,130M	1,100M	1,069M
FCF / share	$5.99	$8.72	$9.88	$11.65	$9.74
Revenue / share	$38.77	$31.70	$34.48	$40.25	$41.22

Revenue CAGR FY22 → TTM: −0.1%. Four years, $44.2B to $44.1B.
Operating income CAGR FY22 → TTM: −10.4%/yr. This is the honest bottom-line trend.
FCF/share peaked in FY2025 at $11.65 and is now $9.74 — down 16%.
FCF CAGR reads +23.3% on a 3-year basis only because FY2022 was a working-capital trough. On FY2023 → TTM it is +1.9%/yr.

⚠️ Ignore the net-income line. FY2025 GAAP net income fell 45% on rising operating income because of a $5.7B non-cash tax charge ($5.29/share) booked in FQ4 FY25 when the One Big Beautiful Bill Act forced a valuation allowance against deferred tax assets — and it appears to have been substantially released in FQ2 FY26 (tax provision −$5,138M; deferred tax assets $743M → $5,968M). The two roughly cancel inside the TTM window, which is why TTM EPS of $8.75 looks sane — it is right by accident. See §8 and the new playbook note.

Data gap flagged: the framework asks for 5-8 year CAGRs. Yahoo returns only 4 annual periods and the roic.ai plan returned only 2 (FY24-25). Every CAGR above is 3-4 year. Longer history would need the 10-K archive.

1.4 Cash conversion is breaking down

This is the finding least visible in the headlines.

	Jun-2025	Jun-2026	Change
Inventory	$6,338M	$8,379M	+32.2%
Days inventory	125	163	+38 days
Receivables	$3,410M	$4,668M	+36.9%
Days sales outstanding	30	43	+13 days
Cash + short-term investments	$10,011M	$8,304M	−17.1%

Inventory up 32% and receivables up 37% while revenue fell 4%. The FQ3 working-capital drag was −$1,648M, which is why FCF collapsed to $495M — a quarter in which free cash flow came in below stock-based compensation ($830M) and below the dividend ($973M).

Management's defence is explicit and not unreasonable — Amon: "the industry now is probably operating very similarly to what was in the pandemic. Everything is at 100% utilization… Inventory is also, in times of shortage, a strategic advantage." Deliberate strategic inventory in a shortage is a defensible use of cash. But it is a bet, and it is the same bet that ends badly every time a shortage turns. One quarter is not a trend; two would be.

1.5 Balance sheet — genuinely fine

Metric	Value	Read
Total debt	$15.27B	Flat for three years
Cash + ST investments	$8.30B	Down from $11.82B two quarters ago
Net debt	$6.97B	0.5× TTM EBITDA
Debt / Assets	26.6%	Down from 29.5% at FY25 close
Interest coverage (TTM)	14.8×	9.1× at the FQ3 run-rate
Current ratio	2.02	Healthy
Goodwill + intangibles	$15.78B	vs $27.66B equity — Alphawave/Arduino/Modular M&A
Tangible book value	$11.87B	$11.23/share vs a $162.67 price

No solvency question, no refinancing wall, no covenant risk. The balance sheet is not why this is a hold.

1.6 Capital allocation — returning more than it earns

TTM, $M:

Use	Amount	% of TTM FCF
Buybacks	$9,250	89%
Dividends	$3,825	37%
Total returned	$13,075	126%
Capex	$1,985	(19% of OCF; 9-month capex doubled to $1,578M from $785M)
Acquisitions (cash)	~$1,573	Cash portion only — see the M&A table below
R&D (expensed)	$9,893	22.4% of revenue

The M&A is much larger than the cash-flow line shows, because the biggest deal was stock:

Deal	Size	Closed	Purpose
Alphawave Semi	$2.4B EV	Dec 2025 (a quarter early)	SerDes / high-speed interconnect for chiplet and rack-scale
Ventana Micro Systems	~$2.4B	Dec 2025	RISC-V server CPUs, parallel track to Oryon
Modular Inc.	~$3.9B, all stock	~July 2026	Chris Lattner's hardware-agnostic AI software stack — the anti-CUDA play
Arduino + Edge Impulse	undisclosed	2025	30M-developer funnel for edge AI

~$8.7B of data-centre capability bought in roughly eight months. That is a real, non-incremental strategic commitment, and it is coherent with the Investor Day plan. It also means management is simultaneously: funding an $8.7B transformation, running a $9.25B buyback, doubling capex, and building $1.8B of inventory — from a base where operating income just fell 41%.

Returning 126% of free cash flow is fine for a quarter or two out of a $10B cash pile. But note the quarterly buyback cadence: $2,444M → $2,650M → $2,792M → $1,364M. Management halved the repurchase in the quarter the margin broke. Read that as prudence, not weakness — but read it. It is the clearest signal in the filings that management is conserving cash.

Share count is doing its job: 1,099M → 1,069M diluted, −2.7% YoY, net of ~$3.2B of annual SBC. That is a genuine ~2.5-3%/yr tailwind to per-share value and it is the single best thing in the capital-allocation picture.

1.7 Dividend overlay

QCOM is a modest dividend payer, not a dividend grower thesis — but the overlay applies.

Metric	Value
Quarterly dividend	$0.92 (raised from $0.89, June 2026)
Forward annual	$3.68
Forward yield at $162.67	2.26%
5-year average yield	2.12%
10yr dividend CAGR (2015-2025)	6.6%
5yr dividend CAGR (2020-2025)	6.5%
Last two raises	+4.7% (2025), +3.4% (2026)
FCF payout ratio (TTM)	36.7% (from 29.7% in FY25)
FCF-less-SBC payout ratio	53.3%

Verdict on the dividend: covered and safe. Dividend growth is what is deteriorating. The payout consumes 37% of free cash flow and 53% of free cash flow after stock comp — both comfortable, and it survives a 40% cut to FCF without difficulty. But the raise cadence has halved, from 6.5%/yr over a decade to 3.4% this year. Management is telling you what it thinks of the earnings trajectory through the dividend, and it is saying "cautious."

⚠️ Data trap: Yahoo reports dividendYield as 2.43%. That is $3.68 / $151.57 — computed off yesterday's close on a day the stock rose 7.32%. The correct yield at spot is 2.26%. Against the 2.12% five-year average this is the difference between a 14.6% "cheap" reading and a 6.6% one. New note written: pitfall-yahoo-dividend-yield-uses-previous-close.

2. Moat & Competitive Advantage

2.1 Quantitative base — the moat is measurably eroding

ROIC (NOPAT at a 21% notional rate ÷ invested capital):

FY2022	FY2023	FY2024	FY2025	TTM	FQ3 FY26 annualised
37.4%	18.5%	19.8%	27.2%	18.8%	12.0%

Invested capital grew from $33.5B to $42.9B (+28%) while operating income fell from $15.9B to $10.2B (−36%). (roic.ai's FY25 figure of 13.7% uses net income and inherits the tax-charge distortion; the NOPAT series above is the usable one.)

Gross margin: 57.9% → 55.7% → 56.2% → 55.4% → 53.1% (FQ3 FY26).

On the Q3 call, CFO Palkhiwala put the baseline QCT gross margin range at 48-50% — and said data-centre revenue will be a further 1.5-2.0 point drag on QCT's weighted-average gross margin as it ramps.

Four forces are pushing on gross margin, and only two of them are moat erosion. This distinction matters and the first draft of this analysis got it too simple:

Force	Direction	Nature
Input-cost shock — wafer, assembly, test, packaging, memory all at 100% utilisation	↓	Cyclical. Reverses
Double-digit, broad-based price increases effective ~Sept 1, phasing over two quarters	↑	Cyclical. Management expects QCT back to the 48-50% baseline
Premium-tier trade-down + data-centre custom ASIC at 1.5-2.0pt dilution	↓	Structural. Real erosion
QTL shrinking as a share of the mix	↓	Structural but benign

That last row is the correction. Consolidated gross margin (~55%) sits above QCT's (48-50%) only because QTL's near-100%-gross-margin licensing revenue lifts it. As non-handset QCT grows toward $40B while QTL stays flat at ~$5.5B, consolidated gross margin mechanically compresses toward QCT's even if QCT is perfectly healthy. Consolidated GM drifting from 55% to the low 50s over five years is the expected arithmetic of the strategy working — not evidence it is failing.

Net read: the margin reset is real but it is roughly half mix arithmetic and half genuine erosion. The honest way to track it is QCT EBT margin — 30% in FY2025, 26% this quarter, guided 23-25% for Q4. That series has no mix artifact in it, and it is the one that has actually broken.

2.2 Revenue-stream map

Stream	~Size	Margin	Durability	Risk
QTL licensing	$5.6B (FY25), 13% of revenue	~70% EBT — ~32% of company operating profit	Standard-essential patents; paid regardless of whose modem is inside	Apple license expires March 2027. Huawei's expired and has produced $0 since FQ2 FY25
QCT handsets	~$21B, 51% of revenue	Falling with mix	Samsung strong (~70% of Galaxy flagships)	−20% YoY. Apple going from ~$7.5B (FY26) to <$2B (FY27)
QCT automotive	~$6B run-rate → ~$7B exit-FY26	Improving	BMW and Stellantis platform deals "well into the 2030s"	Auto design cycles are long — and so is the lag if a generation is lost
QCT IoT / industrial	~$7.3B	Mixed	>$7B design-win pipeline, >$3.5B won this year	Consumer IoT hit by memory costs; industrial the real engine
Data centre	~$0 today	Below corporate average by design	Two hyperscaler ASIC engagements, POs in hand, wafers started	Entirely unproven. First revenue December quarter

The concentration to hold in mind: 13% of revenue throws off roughly a third of operating profit, and its largest contract expires in seven months.

QTL's share of profit	FY2025	FQ3 FY26
% of total revenue	12.6%	12.8%
% of segment EBT (QCT+QTL+QSI)	25.4%	28.7%
% of consolidated pre-tax income	31.9%	35.8%
% of GAAP operating income	32.6%	54.2%

The last cell is the one to sit with. In a bad QCT quarter, licensing carries more than half of Qualcomm's operating income. QTL is not a growth engine — it is the profit floor.

2.2b The licensing moat is legal and structural, not technological

This is where the moat leg changed the verdict, so the mechanics are worth stating precisely.

How the royalty actually works (FY2025 10-K + the published rate card): royalties are a percentage of the licensee's wholesale selling price of the complete device — 2.275% (single-mode 5G, SEP-only) to 5% (multimode, full >130,000-patent portfolio) — but only on the first $400 of net selling price, reduced from $500. Effective ceiling ≈ $16.25 per handset. Connected vehicles are a flat $5/unit for 5G.

Two consequences fall straight out of the cap:

The September price increases do nothing for QTL. Palkhiwala, asked directly whether OEM price rises help licensing: "there is a cap on the total royalties based on the device ASP… to the extent that prices go up below the cap, there is some benefit… above the cap, it really doesn't make a difference." The pricing lever is a QCT lever only.
QTL's growth levers are unit volume and mid-tier ASP — both of which are shrinking. This is a ~$5.0-5.5B, ~70%-margin annuity that has stopped growing: $5,572M (FY24) → $5,582M (FY25) → tracking ~$5.2B (FY26).

Three things make it far harder to break than the drawdown implies:

Legal separation. The patents sit in QUALCOMM Incorporated; QTI — the subsidiary running QCT and all R&D — has, per the 10-K, "no right, power or authority to grant any licenses… under any patents owned by QUALCOMM Incorporated." A licensee buying zero Qualcomm silicon still owes royalties. This is why Apple leaving QCT does not touch QTL.
Binding-arbitration renewal clauses. The 10-K discloses that certain agreements provide that if the parties cannot agree renewal terms by a date, "either party may initiate binding arbitration… which would become effective immediately after the expiration of the prior agreement." A licensee cannot walk and simply stop paying. The realistic worst case at the March 2027 Apple expiry is arbitration-set terms, not zero — which converts a binary risk into a rate risk. It is also the ceiling: arbitrators may set terms "less favorable to us."
Samsung is signed through the end of 2030 — and the agreement explicitly covers "3G, 4G, 5G and upcoming 6G mobile technology." This is widely regarded as the first major license to pre-cover 6G, and it establishes a precedent rate structure before the standard exists.

On 6G specifically: 3GPP Release 21 freezes Stage 1 in March 2027, with commercial networks late 2029. 6G extends the annuity rather than ending it — it is an evolution of the OFDMA/NR foundation Qualcomm authored, and Samsung has already pre-priced it. The real erosion vector is Huawei, which holds roughly 6.5× more 5G SEPs than Qualcomm while earning only ~$630M of IP revenue against Qualcomm's ~$8B licensing haul. That 13×-less-revenue-from-6.5×- more-patents gap is a negotiating-posture gap, not a technical one. If Huawei ever chooses to monetise properly — and it has every geopolitical incentive to — the aggregate royalty stack rises and Qualcomm's share of it falls.

2.3 Adversarial stress-test — "how would a well-funded rival attack this?"

What is genuinely hard to replicate — five things:

The modem. Nobody has closed it on the first attempt. Apple needed a $1B Intel asset purchase, six years, and still ships without mmWave. Carrier certification across hundreds of operators worldwide is an unglamorous, compounding asset.
System integration inside a power envelope. CPU + GPU + NPU + ISP + modem + RF front-end on one thermal budget. Buying Arm CPU IP gets a rival one block out of six.
The patent estate and the legal architecture around it — the QTI separation and the binding-arbitration clauses above. Every party that has litigated it — Apple, the FTC, Transsion, Arm — has settled or lost.
Oryon custom cores (the Nuvia asset), now legally cleared. This is what breaks the "we all buy the same Arm IP" symmetry at the very top of the stack.
Automotive multi-generation platform lock-in. BMW and Stellantis-class agreements running "into the 2030s" are 7-10 year switching costs — far stickier than anything in handsets, and a genuinely new moat rather than a preserved one.

What is not defensible, and is being taken right now:

Apple has simply left. C1 (iPhone 16e), C1X (17e, Air), C2 (iPhone 18 family, Sept 2026). The only residual socket is mmWave, which Apple's silicon still does not support, so US iPhone 18 Pro variants are expected to keep Qualcomm parts. A narrow foothold with a visible expiry. Apple's licensing payments survive the modem transition — which is why the March 2027 licence expiry matters more than any modem milestone.
MediaTek up-market. Q1 2026 smartphone SoC share: MediaTek 32%, Qualcomm 23%, Apple 19% — though MediaTek fell from 38% a year ago as the low end collapsed. Its flagship Dimensity 9500 and Qualcomm's mid-tier parts use the same Arm C1-Ultra core, so at anything below the Oryon flagship the differentiation is integration and price. Palkhiwala named "a weaker mix within premium tier… OEMs making a choice on which chip to use and also using prior generation" as a gross-margin cause. Trading down within premium is precisely MediaTek's wedge.
Samsung LSI — bounded, and Qualcomm is winning it. Exynos 2600 is on 2nm while every rival flagship is on 3nm, and it is back in the global Galaxy S26/S26+. But Snapdragon still takes the US, China, and every Ultra model worldwide — ~70% of Samsung flagships — and Exynos 2600 has publicly reported thermal throttling under sustained load. A process-node lead did not buy the crown. Samsung has re-insourced and retreated repeatedly since Exynos 990; 70% is better than several prior cycles.
Chinese vertical integration — the largest single concentration in the business. 45.9% of FY2025 revenue is billed to China + Hong Kong, and industry estimates put Chinese OEMs at ~68% of QCT handset revenue. The 10-K names the scenario outright: "Chinese OEMs develop and use their own integrated circuit products… rather than our integrated circuit products." Palkhiwala confirmed reliance on domestic Chinese memory "will strengthen as we go forward." Memory today, application processors later, is the historical pattern.
Data centre is an attack Qualcomm is making, not defending. Against NVIDIA, AMD, Broadcom's custom-XPU business and the hyperscalers' own silicon teams, Qualcomm is a new entrant with no installed base and no software moat — which is why it paid ~$3.9B for Modular. Amon conceded it: "we recognize that investors want to see more proof points that we can successfully execute on our plans as a new entrant."

One rival test Qualcomm passed, and it is the cleanest datapoint available. Arm is now building and selling its own chips — competing with its own licensees — and it went directly after Meta's data-centre CPU socket against Qualcomm. Qualcomm won it, taking a multi-generation Meta agreement for the Dragonfly C1000 CPU, announced with Zuckerberg on stage at the June 2026 Investor Day. Whatever one thinks of the datacenter plan, that is a competitive result, not a slide.

Stress-test verdict: the licensing moat is wide, legally armoured and no longer growing; the handset chip moat is narrowing at the mid-tier and holding at the flagship; and a genuinely new moat is being built in automotive. The company is using a strong balance sheet and a capped annuity to buy time to enter three markets at once. That is the right strategy, competently executed so far. It is still not evidence that it arrives on the promised schedule.

2.4 Disruption forecast, 5-10 years

The handset royalty base shrinks with the handset market. Management guides the FY27 handset market down low-teens. 6G will renew the SEP cycle, but Chinese contributors hold far more essential 6G IP than they did at 5G, and every renewal is a renegotiation at a lower rate. The direction is knowable; the slope is not.
Automotive is the credible second leg. Multi-generation platform contracts with BMW and Stellantis, an order-of-magnitude increase in silicon content per vehicle, and 23 straight quarters of double-digit growth. This is the part of the story that deserves to be believed.
Data centre is a real option with a real price attached. Two hyperscaler POs with wafers already started, plus the Meta Dragonfly C1000 agreement, is more than a press release. But it is lower-margin custom silicon, it dilutes QCT gross margin by 1.5-2.0 points, the merchant accelerator (HBC) has no customer-validated silicon yet (Gen 1 tape-out complete, first product mid-2027), and the C1000 server CPU ships 2H 2028. The FY29 target was raised from $5B to $15B on a business with zero revenue today.
PC remains a rounding error. >10% of US retail Windows laptops above $800, but x86 still held 87% of PC revenue share in 2025. Folded into a $6B "personal AI and compute" bucket, so not individually load-bearing.
RISC-V is a hedge Qualcomm is buying, not a threat it is facing. It co-founded Quintauris (with Bosch, Infineon, NXP, Nordic) for automotive/IoT RISC-V, is co-developing a RISC-V Snapdragon Wear platform with Google, and paid ~$2.4B for Ventana Micro Systems (Dec 2025) to develop RISC-V server CPUs in parallel with Arm-based Oryon. Read Ventana as insurance against Arm royalty escalation and against an adverse appeal — see below. RISC-V commoditising the low-to-mid range hurts MediaTek more than it hurts Qualcomm.
⚠️ The Arm matter is not fully closed. Qualcomm won outright — Dec 2024 jury verdict, final judgment entered 2025-09-30 — but Arm noticed an appeal to the Third Circuit on 2025-10-01 and it is pending. Low probability, high impact on the Oryon roadmap. Ventana is the hedge.

2.5 Evergreen assessment

Not evergreen. Durable, but in transition. The 2015-2025 Qualcomm — a premium-modem monopoly with a royalty on every phone sold — is ending, and the ending is visible in the numbers now, not hypothetically. The 2030 Qualcomm is an automotive/industrial/edge-compute company with a flat licensing annuity attached. That company may be a fine business. It is not the same business, it will earn a lower gross margin by management's own guidance, and it should not be valued on the old one's multiple.

2.6 The diversification target moved — which is why the stock did what it did

This explains the $128 → $259 round trip better than anything else.

	Nov 2024 Investor Day	June 2026 Investor Day
FY2029 non-handset QCT revenue	$22B	$40B
— Automotive		$10B
— IoT		>$14B
— Data centre		>$15B
FY2029 non-GAAP diluted EPS	—	>$18.00
Non-handset as % of QCT	—	>50% by FY27, ~⅔ by FY29

Management nearly doubled the FY29 target in eighteen months, and the entire $18B increase is the data-centre line — a business with roughly zero revenue today. The stock rose ~15% on the raise and ran to $259 by late May. Then the FQ3 print showed the funding cost of that ambition landing in the current P&L, and it round-tripped.

Grading each leg honestly:

Automotive — ahead of plan. Design-win pipeline $45B (from $13B in 2021), guided toward $65B; exit run-rate raised $6B → $7B mid-year. Getting from ~$7B to the $10B FY29 target is ~13% CAGR on already-booked wins — the least heroic number in the entire plan.
Industrial/IoT — on plan. >$7B pipeline, >$3.5B of wins secured in FY26 alone against an $8B FY29 industrial target. But IoT grew only 9% this quarter and is guided flat for Q4; consumer is broken, industrial is working.
Data centre — the whole delta, and unproven. $5B in FY27 is partially underwritten by two hyperscaler POs with wafers started plus the Meta C1000 agreement — genuinely more concrete than a typical entrant's claim. Everything between $5B and $15B rests on a merchant accelerator with no demonstrated silicon and a CPU that ships in 2H 2028.

And a margin caveat that gets lost in the revenue headline: $15B of data-centre revenue at custom-ASIC margins is not worth $15B of handset revenue. This is a Broadcom/Marvell-shaped business — high revenue, structurally lower margin, customer-concentrated. Track QCT EBT margin, not QCT revenue.

3. Sentiment & Intelligence

3.1 The FQ3 FY26 print (2026-07-29) — a revenue beat that told you the wrong thing

Revenue $9,947M beat the ~$9.68B consensus and landed at the high end of guidance. Non-GAAP EPS of $2.21 missed by a cent or two. The stock fell 4.4%, then a further 4.7% after hours. Four reasons, stacked:

The Q4 guide. Non-GAAP EPS $2.05-2.25 against $2.36-2.38 consensus — the midpoint is ~9% below.
The Apple step-down accelerated. Palkhiwala: "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%." FY26 Apple product revenue ~$7.5B → FY27 <$2B, guided down from a prior "a little over $2B."
Gross margin missed (~54% vs ~55.6% expected) on premium-tier mix and input costs.
Handsets −20%, with Amon attributing the demand hit to the memory-cost shock: "even a double-digit price increase… is actually small when you compare it to the order of magnitude of the memory bill of materials."

Note the shape of the Apple answer. Bernstein's Rasgon asked directly whether Qualcomm was "starving Apple… and using that silicon to send it elsewhere." Palkhiwala did not deny it: "our supply constraints were a part of it and then where discussions ended up is that we'd have a share materially less than 20%." In an allocation-constrained market Qualcomm appears to have deprioritised a customer that is leaving anyway. Rational — and it pulls the revenue cliff forward into FY27.

3.2 Management's forward framing

The bull case, stated by management: - Non-handset revenue growth accelerates from 24% in FY26 to >60% in FY27, explicitly sized to "replace total Apple product revenues in '26" within the year. - FY29 non-handset target raised to $40B (from $22B set in Nov 2024). - Data centre: $5B FY27, $15B FY29; two global-scale hyperscaler custom-silicon engagements with POs in hand and wafers started, first revenue in the December quarter. - China handsets "reached a bottom in the third fiscal quarter" and return to double-digit sequential growth in Q4. - Double-digit price increases effective ~Sept 1, phasing in over two quarters. - Palkhiwala quantified the current damage as a future tailwind: QCT Android revenue down 20% YoY with an EPS impact "greater than $1.50" — which reverses if memory normalises.

That last point is the strongest thing in the bull case and it is worth taking seriously: a meaningful slice of the current earnings depression is a memory-cost shock in someone else's supply chain, not a Qualcomm problem. The China bottom, though, is a forecast, not a confirmed inflection — and the Q4 guide leans on it.

3.3 The sell side capitulated in one day

On 2026-07-30 every covering firm cut its price target and not one changed its rating.

Firm	Rating (unchanged)	Old → New
Wells Fargo	Equal-Weight	$265 → $170 (−$95)
RBC Capital	Sector Perform	$250 → $160 (−$90)
TD Cowen	Buy	$225 → $175
Cantor Fitzgerald	Neutral	$200 → $165
Susquehanna	Neutral	$190 → $160
Benchmark	Buy	$300 → $270
Citigroup / UBS / Evercore / Rosenblatt / Morgan Stanley	Neutral–Buy	all lowered

Post-print median target: $170. Yahoo's mean of $196.27 is contaminated by stale pre-earnings targets; Yahoo's own median of $175 is closer to honest. Consensus rating: Hold (23 of 37).

The five weeks before the print are the more revealing part. Barclays raised its target $150 → $245 on 2026-06-25 while keeping an Underweight rating; Bernstein $140 → $235; B of A $165 → $195 while keeping Underperform. Analysts raised targets 55-63% without upgrading, then cut them a month later. That is price-chasing, and it is the fingerprint of a narrative trade rather than an earnings trade.

3.4 Insiders: zero buying, at any price

Fourteen open-market sales in six months, ~$7.5M. No purchases — including none at $126 in April.

CEO Cristiano Amon sold 20,000 shares on 4-5 May 2026 at $180-185 ($3.65M), and 150,000 shares on 2025-10-01 at ~$165 ($24.8M) — the largest disposition on record here.
CFO/COO Palkhiwala sold in six of the last six months, a cadence that reads as a 10b5-1 plan and therefore carries little signal — except the 2026-07-14 sale at $178-189, two weeks before the print.
Yahoo reports insider ownership at ~0.0%.

The programmatic sales are noise. The absence of a single purchase during a 51% drawdown from $259 to $126 is not nothing, though it is weak evidence on its own for a megacap where compensation is equity-heavy.

3.5 Legal and geopolitical

Item	Status
Arm v. Qualcomm	✅ Won, but ⚠️ on appeal. Dec 2024 jury verdict for Qualcomm; final judgment entered 2025-09-30; Arm noticed an appeal to the Third Circuit on 2025-10-01, pending. Low probability, high impact on the Oryon roadmap
Qualcomm's countersuit v. Arm	⚠️ Unresolved. Trial was set for March 2026; no outcome found in search as of 2026-08-04. Check the FQ3 10-Q legal proceedings
Apple licence	⚠️ Runs through March 2027 on Apple's exercised two-year extension. The single most important date on this name — though the binding-arbitration clause (§2.2b) bounds the downside to a rate outcome, not a zero
Samsung licence	✅ Signed through end-2030, explicitly covering 6G
Transsion / two major Chinese OEMs	✅ New long-term licences executed FQ2 FY25; all Transsion litigation dismissed
Huawei licence	❌ Expired. $0 QTL revenue since FQ2 FY2025; previously ~$0.10-0.15 EPS/qtr. No substantive renewal update
China SAMR	⚠️ Antitrust probe opened 2025-10-10 over the Autotalks acquisition closing without notification. Qualcomm acknowledged the facts. Unresolved
EU (Icera)	✅ Closed. €242M reduced to €238.7M, Sept 2024, no appeal

China + Hong Kong = 45.9% of FY2025 revenue (ship-to basis, so it overstates true end-market exposure since Chinese OEMs sell globally — but the supply-chain and regulatory chokepoint is real). Apple, Samsung and Xiaomi were each ≥10% customers in FY2025.

4. The price chart is the argument

Date	Price
2026-03	$128.31	Act I / broad tech trough
2026-04	$178.92	+39% in a month
2026-05	$258.96 (weekly high)	Data-centre narrative peak — +102% off the March low
2026-06	$184.79
2026-07-29	$155.68	Post-print
2026-07-31	$147.61	Post-print low
2026-08-04	$162.67	+7.32% in one session on the semiconductor snapback

"−37% off the 52-week high" and "+27% off the March low" are the same stock on the same day. The high was a two-month narrative spike, not a valuation the business ever supported — and Market/regime.md records that the whole sector round-tripped with it (SOX 10,447 → 12,166, +16.5%, in four sessions).

This is principle-down-a-lot-is-not-cheap and pattern-ai-levered-fields-trade-above-own-band in one chart. The watchlist's framing of "−40% off high, drifting into interesting" was measuring against the spike.

5. Valuation

5.1 ⚠️ The forward P/E is measuring the wrong year

The watchlist carries QCOM at "14.2x fwd" and Yahoo reports 15.8x. Both come from forwardEps = $10.297. Check what that number is:


FQ3 FY26 non-GAAP EPS (actual)	$2.21
FQ4 FY26 non-GAAP EPS (guide midpoint)	$2.15
Implied FY2026 non-GAAP EPS (3 quarters actual + guide)	≈ $10.4
Yahoo `forwardEps`	$10.297

Yahoo's "forward" EPS is fiscal 2026 — a year that is three-quarters reported and ends in seven weeks. It is not a forecast of the future; it is an average of a year that included a record $12.25B December quarter, before the margin broke.

The honest forward number is the exit run-rate: $2.15 × 4 = $8.60.

Basis	EPS	P/E at $162.67
Yahoo "forward" (= FY2026, mostly historical)	$10.30	15.8×
FQ4 FY26 guide, annualised	$8.60	18.9×
FQ3 FY26 actual, annualised	$8.84	18.4×
TTM GAAP (tax-distorted)	$8.75	18.6×

QCOM is not a 15.8× stock. It is an 18.9× stock on the earnings its own management guided to six days ago.

5.2 Against its own historical band

Multiple	FY2024 range	FY2025 range	Today
Price / FCF	9.7 – 21.4×	9.5 – 14.4×	16.4×
EV / EBIT	11.7 – 25.6×	10.9 – 16.2×	17.4×
EV / EBITDA	10.0 – 21.9×	9.6 – 14.3×	13.2×

(roic.ai annual high/low; today computed on TTM.)

On both P/FCF and EV/EBIT the stock trades above the entire FY2025 range — after a 37% drawdown, on falling earnings. This is the third consecutive semiconductor name this year to show that shape.

Adjusting for stock comp makes it worse: TTM FCF less SBC is $7.17B, so the price is 23.8× owner cash.

5.3 The models

Model	Output	Weight	Why
FCF / owner earnings	$101-157	Primary	12-15× normalised FCF of $9.5-11.5B, less $6.97B net debt, ÷ 1,057M shares
Earnings multiple	$112-160	Primary	13-16× normalised non-GAAP EPS of $8.60-10.00
Bogle expected return	~4.5% (bear) to ~11% (bull)	Medium	2.26% yield + 2.5% buyback + 0-6% earnings recovery, ± multiple change. Base ~7%
Dividend Yield Theory	$173.6	Low	2.26% vs 2.12% 5yr avg. But the yield rose because the price fell, and dividend growth halved to 3.4% — `pitfall-dyt-inverts-when-price-caused-the-yield`
Graham IV	$71.77	Discard	√(22.5 × 8.75 × 26.17). On tangible BVPS of $11.23 it gives $44. Qualcomm's value is IP and design, not book — Graham says "this is not an asset-backed name," which we already knew
DDM (Gordon)	$47-96	Discard	D₁ $3.83, r 9-12.2% (β 1.66), g 4-5%. A 2.26% yielder cannot be valued on dividends. Reported for completeness

Converged fair value: $120-160. Midpoint ~$140.

At $162.67 the stock sits just above the top of the range — roughly fairly valued in the bull case, 16% expensive against the midpoint.

Cross-check: the post-print sell-side median is $170, and those are non-GAAP targets set by analysts who just cut their numbers ~9% and will likely cut FY27 again as the Apple reset flows through. My range sits below theirs, which is the intended conservative bias.

5.4 What has to be true to justify $162.67

Reverse-engineered: at 15× FCF, $162.67 requires ~$11.8B of normalised free cash flow — above the FY2025 record of $12.8B only if you also believe the FY25 working capital was normal. TTM is $10.4B and the current quarter annualises to $2.0B.

So the price already embeds: the working-capital drag fully reversing, the memory shock normalising, Chinese OEM revenue inflecting on schedule, non-handset revenue growing >60% to replace Apple, and the data-centre business arriving on time at scale. Each is plausible. All five together is the bull case being paid for in advance.

6. Synthesis

6.1 Where the analysts disagree

Fundamentals vs. Sentiment. Fundamentals sees a −41% collapse in operating income, a 32% inventory build on falling revenue, and FCF below the dividend. Sentiment sees a revenue beat, a record automotive quarter, 23 straight quarters of auto growth, and hyperscaler POs in hand.

Resolution — both are right, and the timing is the whole question. The deterioration is current and measured; the replacement is future and promised. Qualcomm has ~$8.3B of liquid assets, a 0.5× leverage ratio and a 70%-margin licensing annuity to fund the gap. It can afford the transition. The question is not survival, it is what you pay for an outcome that is several years from being demonstrable — and management itself said investors need "more proof points."

Moat vs. Valuation — a genuine debate round, and Moat won part of it. Valuation's opening position was that the licensing annuity is only 13% of revenue with its largest contract expiring in seven months, and that management guiding gross margin down five points is a moat resetting its own floor.

Moat rebutted on two specific points and both rebuttals hold:

The margin reset is roughly half arithmetic. Consolidated GM sits above QCT's only because QTL's ~100%-margin revenue lifts it. Growing non-handset QCT to $40B against a flat $5.5B QTL compresses consolidated GM toward QCT's even if everything goes right. The uncontaminated series is QCT EBT margin: 30% → 26% → guided 23-25%. Still broken, but less dramatically than the headline.
The Apple 2027 expiry is not binary. Binding-arbitration renewal clauses mean a licensee cannot walk and stop paying; the worst realistic case is arbitration-set terms. And Samsung — the other ≥10% licensee — is signed through 2030 with 6G pre-covered. The floor under QTL is firmer than a headline "biggest contract expires in seven months" implies.

Valuation's counter-rebuttal, which stands: the $400 royalty cap means the September price increases do nothing for QTL, and QTL revenue has gone $5,572M → $5,582M → ~$5.2B. A durable annuity that does not grow is a floor, not an engine — and floors do not justify premium multiples, they justify not going to zero.

Resolution: Moat raises the conviction, Valuation keeps the verdict. The business quality is better than the fundamentals leg alone showed — hence [5.5] rather than [4.5-5.0]. The price conclusion is untouched: 18.9× the exit run-rate, above the entire FY2025 range on P/FCF and EV/EBIT.

6.2 Verdict — HOLD, conviction [5.5]

The bull case, at full strength: - Licensing annuity: ~70% EBT margin, ~32% of pre-tax income (54% of operating income this quarter), armoured by legal separation and binding-arbitration renewal clauses, with Samsung signed through 2030 including 6G - Automotive: 23 consecutive double-digit quarters, +61%, ~$7B exit run-rate, $45B design-win pipeline heading to $65B, BMW lead-compute win across ADAS and cockpit, Stellantis into the 2030s. The FY29 $10B target is ~13% CAGR on already-booked wins - Two hyperscaler custom-silicon POs with wafers already started, December-quarter revenue, plus a multi-generation Meta agreement for the Dragonfly C1000 — won against Arm - Balance sheet: 0.5× net leverage, 26.6% debt/assets, 14.8× coverage - ~2.7%/yr share shrink plus a covered 2.26% yield ≈ 5% of annual return before growth - >$1.50 of EPS is currently suppressed by a memory-cost shock in someone else's supply chain - Double-digit, broad-based price increases from ~Sept 1, which management argues stick because they are small against the memory BOM increase OEMs are already absorbing - FY25's tax charge sets up a 13-14% forward effective tax rate — a real tailwind

The bear case: - 🚩 Four years of zero revenue growth ($44.2B FY22 → $44.1B TTM) and operating income compounding at −10.4%/yr - 🚩 Operating margin 26.6% → 16.3% in four quarters; opex +21.8% into revenue −4.0% - 🚩 QCT EBT margin 30% → 26% → guided 23-25% — the series with no mix artifact in it - 🚩 QTL has stopped growing ($5,572M → $5,582M → ~$5.2B) and the $400 royalty cap means the September price increases do nothing for it. A floor, not an engine - 🚩 ~68% of QCT handset revenue rides on Chinese OEMs against a stated national policy of semiconductor self-sufficiency — structural and unhedgeable - 🚩 ~$8.7B of M&A in eight months funded alongside a $9.25B buyback, doubled capex and a $1.8B inventory build, from a base where operating income fell 41% - 🚩 Arm's Third Circuit appeal is pending — low probability, high impact on Oryon - 🚩 Inventory +32% and receivables +37% on falling revenue; FQ3 FCF of $495M came in below both SBC and the dividend - 🚩 Apple: ~$7.5B → <$2B, accelerating, and the licence expires March 2027 - 🚩 China + HK 45.9% of revenue with an open SAMR probe and visible vertical integration - 🚩 Every covering analyst cut targets on one day; estimates still being revised down - 🚩 Zero insider buying at any price, including $126; CEO sold $28.5M over ten months - 🚩 The "15.8× forward" is FY2026, three-quarters reported. The real forward is 18.9× - 🚩 Above its entire FY2025 range on both P/FCF and EV/EBIT, after a 37% drawdown

[5.5] — a good business in a real transition, at a price that already assumes the transition works. Not a value trap: the balance sheet is sound, the dividend is covered, the licensing floor is legally armoured, and the automotive leg is genuinely compounding and ahead of plan. Not a buy either: at $162.67, after a 7.3% single-day sector-beta move, you are paying above fair value for an outcome management itself says needs proof points.

This slots between INTC [4.5] and AMAT [6.5] on the watchlist.

The single test that settles this thesis, and when: management claims FY27 non-handset growth of >60% will "replace total Apple product revenues in '26" — about $7.5B — within the year. That is a falsifiable, dated claim. The December quarter (FQ1 FY27) is the first read: data-centre revenue must actually appear against POs already placed and wafers already started, and automotive must sustain ~60%. If both land, this is a [7] at a lower price. If the data-centre line is absent in December after "wafers started," the whole $22B → $40B raise that drove the stock to $259 comes off the table.

A portfolio-specific passage was removed from the public build.

6.4 Trim convention — the multiple form is correct here

Checked against pitfall-multiple-trim-inverts-on-peak-cycle-cyclicals. The diagnostic for the trap is forward P/E far below trailing while margins sit at record highs — Micron at 5.7× forward vs 20.2× trailing on an 84.9% gross margin. QCOM fails that test in every respect: forward 15.8× vs trailing 18.6× is a narrow, ordinary gap, and margins are at multi-year lows, not highs. Forward EPS here is a consensus being cut, so as estimates fall the multiple-form trim tightens automatically — the correct direction.

Trim set at 18× fwd. At today's $10.30 consensus that renders ~$185; if consensus falls to $9.00 it renders $162, which is the intended behaviour.

7. Actions

Verdict: HOLD [5.5]. No add above ~$138. No trim of a $500 position.
Watchlist updated — conviction [5.5], thesis rewritten, trim converted from the stale $165 dollar level to 18× fwd, entry zone widened to $120-138 (was $120-132) to match the fair-value range.
Recheck: FQ4 FY2026 print, early November 2026 — the first clean quarter of the accelerated Apple step-down and the first test of the China-bottom claim.
Break triggers — any of these breaks the hold:
FQ4 gross margin below 52%, or QCT EBT margin below the guided 23%
A second consecutive quarter of inventory building on falling revenue
Chinese OEM revenue failing to grow sequentially in FQ4 (the guide depends on it)
Apple licence not renewed, or renewed at materially lower economics, by March 2027
Data-centre revenue absent from the December quarter after "POs in hand, wafers started"
FY27 non-handset growth tracking materially below the guided >60%
An adverse SAMR outcome on Autotalks, or Arm prevailing at the Third Circuit
Open items to resolve:
Outcome of Qualcomm's countersuit against Arm (trial set March 2026) — not findable in search; check the FQ3 FY26 10-Q legal-proceedings section
Arm's Third Circuit appeal of the Dec 2024 verdict — filed 2025-10-01, still pending
SAMR Autotalks resolution
Identity of the two hyperscaler custom-silicon customers

8. Data notes and new knowledge

Three vendor traps hit this analysis. Two produced new playbook notes:

pitfall-yahoo-dividend-yield-uses-previous-close (new) — Yahoo computes dividendYield off previousClose while computing trailingPE off currentPrice, inside the same payload. On a +7.32% day that reported QCOM's yield as 2.43% when it was 2.26%, halving the apparent DYT signal.
pitfall-tax-valuation-allowance-round-trip-breaks-eps (new) — the $5.7B OBBBA valuation allowance booked in FQ4 FY25 and apparently released in FQ2 FY26 makes two fiscal years of GAAP EPS, ROE and net-income CAGR unusable. QCOM's 3yr net-income CAGR reads −24.6% against a +23.3% FCF CAGR for the same window.
Yahoo's forwardEps can be the current fiscal year when that year is nearly complete. For a September-FY company analysed in August, "forward P/E" is three-quarters historical. Recorded in §5.1 rather than as a separate note — the general fix is to derive the forward multiple from the company's own next-quarter guidance, not from a vendor field.

Data gaps flagged: only 4 annual periods from Yahoo and 2 from the roic.ai plan, so every CAGR in §1.3 is 3-4 year against the framework's requested 5-8. The analysis_notes.md FCF CAGR requirement is therefore only partially satisfied.

Sources: Yahoo Finance MCP · roic.ai MCP · Public.com (real-time quote cross-check) · Q3 FY26 earnings release · Q3 FY26 earnings call transcript · FY2025 10-K · Qualcomm/Arm outcome · SAMR probe · HUMAIN data centre · FY25 tax charge