QCOM › analyze
QCOM · Analyze
Refresh analysis · 2026-09-10 · $176.88 · baseline analyze-2026-08-04 ($162.67, HOLD [5.5])
Verdict: HOLD — [5.5]. Do not add here. No trim of the ~$500 position. Nothing in the business changed since 8/04 — there has been no earnings print — but two things around it did: the stock ran +8.7% to $176.88, and on September 8 Qualcomm signed a named custom-AI-silicon deal with Amazon/AWS, converting one of the two unnamed hyperscaler engagements in the thesis into a real, warrant-backed contract. That is a genuine de-risking of the single weakest leg of the data-centre story — and it is also the reason the stock is now above the entire $120-160 fair-value range, ~$17 over the top.
The Amazon deal is structure, not revenue: no volumes, no delivery dates, no measured performance were disclosed, and it is built on a customer-inducement warrant — which is contra-revenue, i.e. Qualcomm is paying Amazon in equity to buy its chips (see [[pattern-customer-warrants-are-contra-revenue-that-buys-the-backlog]]). It raises the credibility of the December-quarter revenue test; it does not change the test. The improved data-centre credibility and the richer price roughly cancel — hence the verdict is unchanged at [5.5], and the price conclusion is firmer: at 20.6× the exit run-rate management guided to six weeks ago, you are paying above fair value for an option that has not yet printed a dollar.
0. What this is
This is a light refresh of the 2026-08-04 full analysis, not a re-derivation. QCOM has not reported since (FQ3 FY26 was 7/29; FQ4 FY26 lands ~Nov 11). The complete fundamentals, moat, and valuation work stands in the baseline report. This file records only what changed in the Aug 4 → Sep 10 window and re-tests the verdict at the higher price.
Knowledge check. kb.py find QCOM returned the 8/04 report as the live baseline. Three
playbook notes govern the framing and are cited, not re-derived:
| Note | Bearing |
|---|---|
| [[pattern-customer-warrants-are-contra-revenue-that-buys-the-backlog]] | The Amazon warrant is a textbook instance — the "$60B" headline overstates the economics; the warrant fair value is contra-revenue |
| [[pattern-ai-levered-fields-trade-above-own-band]] | QCOM confirms it again — a +9.3% run on a structure-not-revenue deal pushed it back above its own band |
| [[principle-down-a-lot-is-not-cheap]] · [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] | The "17.3× forward" is still the nearly-complete FY2026 base — the honest forward is the exit run-rate |
1. The one thing that changed — the Amazon/AWS deal (Sep 8, 2026)
What was signed: a multi-generational agreement for Qualcomm to supply Amazon/AWS with custom AI-inference silicon and optical networking (SerDes + optical DSP, links to 1.6 Tb/s). Structure disclosed:
| Term | Detail |
|---|---|
| Warrant to AWS | up to 25M QCOM shares at $161.26 strike (~$4B notional) |
| Vested at signing | 3.75M shares; remainder vests on binding orders / purchases |
| Purchase framework | eligible AWS purchases counted up to a $60B cap |
| Warrant expiry | Sep 3, 2036 |
| Disclosed quantities / dates / performance | none |
How to read it — three points:
- It names a hyperscaler. The 8/04 thesis rested on "two hyperscaler ASIC engagements, POs in hand, wafers started" — both unnamed. Amazon is now one of them, and the CFO says a second is still progressing. Naming the largest cloud provider is materially better evidence than two anonymous engagements. This is the real positive.
- It is still an option, not revenue. No volumes, no delivery schedule, no benchmarked performance. The warrant vests as Amazon buys — which is the tell that the $60B is a ceiling on a multi-year purchase framework, not a booked backlog. The December-quarter revenue test from the 8/04 report is unchanged: data-centre revenue must actually appear.
- The warrant is contra-revenue. A customer-inducement warrant is Qualcomm paying Amazon (in equity) to become a customer — [[pattern-customer-warrants-are-contra-revenue-that-buys-the-backlog]]. The fair value of the vested warrant is amortised against the revenue it generates, so reported data-centre revenue and margin will be lower than the gross purchase figures imply. The "$60B" headline is the least useful number in the release. Track the QCT data-centre EBT margin when it is first disclosed, not the purchase cap.
Market reaction: gapped +6.9% on 9/8, touched $183.49 intraday, closed $174.09 on ~26M shares (3× normal); held to $176.88 by 9/10. Combined with a broad AI/data-centre re-rating Aug 4 → Sep 4 (+4.3%, incl. Aug 31 +3.8%), the stock is up ~9.3% since the baseline. CFO Palkhiwala at Goldman Communacopia (9/9) called Amazon a "landmark deal" and — notably — said the shares are "well-priced." Management is not signalling the stock is cheap here.
2. Everything else — unchanged since 8/04
Confirmed no material change on every other thesis leg:
| Leg | Status vs 8/04 |
|---|---|
| Fundamentals | No new print. The −41% FQ3 operating-income collapse, the 32% inventory / 37% receivables build, FCF of $495M below both SBC and the dividend — all stand until FQ4 (~Nov 11) |
| Meta Dragonfly C1000 | No change — 250-core Oryon CPU, production not until 2H 2028 |
| HUMAIN / AI200 / AI250 | No change — 200 MW Saudi from 2026; AI200 commercial 2026, AI250 2027; no new named accelerator customers beyond Amazon |
| Arm Third Circuit appeal | No change — filed Oct 2025, no ruling/briefing milestone; reversal viewed as unlikely, ~1yr to resolve |
| China SAMR / Autotalks | No change — probe still open, no resolution |
| Qualcomm countersuit vs Arm (March 2026 trial) | ⚠️ Still unconfirmed — no reported verdict/settlement/docket update. Open item; check FQ4 10-K legal proceedings |
| Apple licence | No change — expires March 2027, no renewal news, C1 modem transition continuing |
| Automotive / IoT | No new design-win or run-rate disclosure since the 7/29 print (auto ~$1.6B record +61%, run-rate → ~$7B exit FY26; IoT ~$1.8B +9%) |
| Insiders | Still all sells, zero buys. Since 8/04: CAO Grech 625 @ $162.85 (8/21); Palkhiwala 2,500 @ $161.77-165.34 (8/12) + intent to sell ~3k more. No insider bought the AI narrative. |
Snapdragon Summit correction: the 8/04 report listed it as a September catalyst; it has not happened yet — scheduled Sep 22-24, 2026 (Maui), expected to reveal Snapdragon 8 Elite Gen 6 + a Gen 6 Pro tier on a 2nm process. Live forward catalyst, no reaction to score.
3. Valuation — the price ran past the news
The 8/04 report converged fair value at $120-160 (12-15× normalised FCF; 13-16× normalised non-GAAP EPS of $8.60-10.00). The Amazon deal improves the credibility of the data-centre option that sits on top of that base, so I nudge the range up modestly to $125-165 — but no more, because the deal is not yet revenue and the warrant dilutes the economics.
| Basis | EPS / FCF | Multiple at $176.88 |
|---|---|---|
| Yahoo "forward" (= FY2026, ¾ reported) | $10.22 | 17.3× — the wrong year, see [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] |
| FQ4 FY26 guide, annualised (exit run-rate) | $8.60 | 20.6× — up from 18.9× at the baseline |
| TTM GAAP (tax-distorted) | $8.75 | 20.2× |
| EV/EBITDA (TTM) | 16.0× — above the FY2025 9.6-14.3× band |
At $176.88 the stock is ~$17 above the top of fair value and 20.6× the exit run-rate — richer on every basis than at the baseline. Analyst targets have started nudging up on the deal (RBC $160→$180 on 9/9; Piper initiates Neutral $190 on 9/10; Rosenblatt maintains Buy $235), but the median target is still $175 — about 1% below spot. The mean of $193.90 is skewed by a bull tail (Benchmark $270, Barclays $245). Consensus rating remains Hold.
4. Verdict — HOLD [5.5], unchanged
The Amazon deal and the higher price cancel. Business quality ticked up (a named hyperscaler is real de-risking of the weakest leg); the valuation ticked down (20.6× the exit run-rate, above the FV range). The net is the same [5.5] — a good business in a real transition, at a price that now embeds not just the transition working but the data-centre option paying off, on a deal that is still structure rather than revenue.
The dated, falsifiable test is unchanged and now partly de-risked on the customer side: management says FY27 non-handset growth of >60% replaces all of FY26's ~$7.5B Apple revenue within the year. Amazon names a buyer; the December quarter (FQ1 FY27) must show the revenue. If it appears, this is a [7] — but ideally bought back inside $120-138, not chased at $177.
Actions: - HOLD [5.5]. No add above ~$138. No trim of the immaterial held position. - Fair value nudged to $125-165 on improved data-centre credibility; spot ~$17 above the top. - Trim re-set to 19× fwd (from 18×) — one turn wider to give the now-named data-centre option room; renders ~$194 at the current $10.22 forward field, ~the analyst mean. As consensus is cut further on the Apple reset, the multiple-form trim tightens automatically — the correct direction ([[pitfall-multiple-trim-inverts-on-peak-cycle-cyclicals]] checked; QCOM is not in the trap). - Recheck: FQ4 FY2026 print, ~Nov 11, 2026. Secondary read: Snapdragon Summit Sep 22-24. - Break triggers (unchanged from 8/04, plus one): FQ4 GM <52% or QCT EBT margin <23% · a 2nd consecutive quarter of inventory build on falling revenue · Chinese OEM revenue not up sequentially in FQ4 · Apple licence unrenewed by March 2027 · data-centre revenue absent from the December quarter after Amazon signed · FY27 non-handset growth tracking well below +60% · adverse SAMR ruling on Autotalks or Arm prevailing at the Third Circuit. - Open items: the March-2026 Qualcomm-vs-Arm countersuit outcome (still unconfirmed); the exact FQ4 date; identity of the second hyperscaler.
Sources: Yahoo Finance MCP (quote, recommendations, insider transactions) · Bloomberg 9/8 Amazon deal · WinBuzzer warrant terms 9/9 · Stocktwits — CFO at Goldman Communacopia 9/9 · Snapdragon Summit 2026 dates · baseline analyze-2026-08-04.