AVGO · Q3 FY2026 THESIS: MONITORING
Broadcom Inc. · Nasdaq AVGO · earnings teardown

What actually changed in the third quarter

Fiscal Q3 2026, quarter ended 2 August 2026 · results released 2 September 2026 · Form 10-Q filed 10 September 2026
Companion to Broadcom_Earnings_Q3_FY2026.docx · every figure traced to a primary document · prepared 21 September 2026
Net revenue $29.6B ▲ 86% y/y Exceeded the whole of FY2019 in one quarter. 10-Q: $29,591M
Operating margin 67.9% ▲ 240 bp y/y A record, achieved through a 210 bp gross-margin fall. 2 Sep call · non-GAAP
Software growth, ex-upfront +8.6% vs. +29% reported Consumption revenue was flat sequentially. computed · 10-Q Note 2
Share repurchases $0 $10.1B authorised, unused Against record $13.7B of free cash flow. 10-Q Note 7
The 20% that moves understanding

Four findings

Ordered by how much each changes the long-term case, not by how much attention it got on the call. Every one of them is visible only by putting the transcript beside the filing.

High impact · earnings quality

All of the sequential software growth is recognition timing

Upfront licence revenue — recognised at signature on contracts without termination-for-convenience rights, and booked inside products revenue rather than subscriptions — was $3,465M, or 40% of the infrastructure software segment against 28% a year ago. Software revenue rose $1,530M sequentially. Upfront licence revenue rose $1,501M.

Infrastructure software revenue, split by how it is recognised
$ millions · fiscal quarters, period-end dates as filed
Revenue ex-upfront licence (consumption) — the base Upfront licence revenue — the cap

Upfront licence revenue per Form 10-Q, Note 2, “Reclassifications to Condensed Consolidated Statements of Operations”: $3,465M in Q3 FY2026, $1,964M in Q2 FY2026, $1,916M in Q3 FY2025. Segment revenue per the 10-Q segment note; Q2 FY2026 segment revenue of c.$7,200M per the 3 June 2026 call. Consumption line computed as the difference.

Annual recurring revenue growth, as disclosed on the call
per cent year on year · the consumption measure management still gives

The Q4 guide of $8.7B is described as software revenue that will “stabilize”, which is a sequential decline — and holding it needs roughly another $3.4B upfront quarter. The $8.9B Q3 guide came in at $8.8B, the first clear shortfall in this sequence.

The 94%-gross-margin, uncorrelated half of the business is growing at single digits on consumption and decelerating on ARR. The reported 29% is a signature calendar, not a demand signal — and it is the half of Broadcom that was supposed to survive a downturn.
High impact · disclosure

Four routine disclosures ended at once, in the new CFO’s first quarter

Amie Thuener, previously Alphabet’s chief accounting officer, replaced Kirsten Spears effective 12 June 2026. Q3 FY2026 is her first reported quarter. Four series that had appeared in every prior call are absent from both the results and the guidance.

Was it disclosed on the earnings call?
five recurring metrics × five quarters · figure shown where given
✓ Given, with a figure ◐ Mentioned, no figure ✕ Not given

Counted directly in the archived transcripts of 4 Sep 2025, 11 Dec 2025, 4 Mar 2026, 3 Jun 2026 and 2 Sep 2026.

Days sales outstanding is the useful complication: it had already lapsed after Q4 FY2025, two quarters before the CFO change. So the thinning is a trend that predates the new CFO rather than her doing — which makes it more, not less, worth watching.

The withdrawal is not obviously concealment. Both working-capital metrics improved when computed independently: days of inventory fell to about 59 from 86 at Q2, and DSO of about 42 days is unremarkable. What is gone is the ability to check.

Days of inventory on hand
days · solid markers as disclosed on each call; the hollow marker is computed

Disclosed figures per each quarter’s call. Q3 FY2026 computed on the company’s own basis — inventory ÷ cost of products sold × days — from the 10-Q: 4,523 ÷ 6,983 × 91.

With consolidated backlog, bookings and total contract value all gone, the only forward bridge left is the 10-Q’s remaining performance obligations — and its twelve-month share has fallen for two consecutive quarters. The FY2027 and FY2028 AI numbers now rest on management’s word, which management has said it will not refresh quarterly.
Monitoring · capital allocation

No buybacks in a record cash quarter, and the share count is rising

Repurchases went $7,850M in Q1 to $600M in Q2 to nothing in Q3, with $10.1B of authorisation available and $13.7B of free cash flow in hand. Shares outstanding rose to 4,774M from 4,741M at the year end — up 33M, or 0.7% — because 58M shares vested against 25M retired.

Fiscal 2026Q1Q2 Q3Nine months9M FY2025
Repurchases ($M)7,850600 08,4502,450
Shares retired (M)232 02516
Shares issued on vesting (M)1824 1658n/d
Shares outstanding, period end (M)4,736 4,7584,7744,774n/d

Form 10-Q, condensed consolidated statements of stockholders’ equity and Note 7. Cash went to debt instead: $5.6B of long-term debt repaid in the quarter plus $1.5B of senior notes after quarter end; aggregate indebtedness $61,079M against $66,720M at 3 May.

The reason is on the record, six days after the call — and so is a divergence between the two executives who run the balance sheet.

Give me three months. We usually do the capital allocation annually in the December board meeting. … The other likelihood is we’ll do a stock buyback. These are probably the only two choices we have. Hock Tan, CEO · Goldman Sachs Communacopia, 8 September 2026
To probably bring down debt makes sense, but our total debt of about $56 billion was taken on when interest rates were much lower. So it’s almost like, why would I want to prepay low cost interest or debt? Hock Tan, CEO · same session — six days after the CFO reported $7.1B of debt reduction
Per-share compounding in FY2026 is coming entirely from earnings growth, not from a shrinking share count, and the Q4 guide of 4.94B diluted shares assumes none. Against an indicated “mid-40s” of billions of FY2027 free cash flow, the December board decision matters more to a long-term owner than the difference between $115B and $130B of FY2027 AI revenue.
Monitoring · competitive position

A rival TPU design partner was named for the first time

Across all forty archived Broadcom transcripts going back to 2016, a competing custom-silicon design partner had never been named in an AI context. On 2 September, unprompted, it was.

Just to reiterate the major technology challenges of developing these complex accelerators, Broadcom is now shipping the TPU v8i ahead of the MediaTek version v8t, which in fact was initiated earlier. Hock Tan, CEO · Q3 FY2026 earnings call, 2 September 2026

The framing matters as much as the fact. Multi-sourcing on Google’s TPU line is no longer deflected; it is reframed as a time-to-market race Broadcom is winning, with the stated moat being first-pass silicon — “consistently … delivered the fastest time to market for TPUs from product definition to production without the need for respins.”

Google is the oldest and most profitable socket in the portfolio, and the FY2028 outlook rests on locks already set rather than new wins. Whether the next generation stays sole-source is now the single most valuable checkable fact in the coverage.
Forward visibility

The backlog grew; the near end of it shrank

Remaining performance obligations are the only committed-revenue figure left in the public record now that consolidated backlog has gone. The total is up more than six-fold in a year — and the slice arriving within twelve months has fallen in both share and dollars.

Remaining performance obligations, split by timing
$ billions · as filed at each period end · the 12-month dollar figure is shown where the segment fits it, and in every tooltip
Expected within 12 months Expected beyond 12 months

Form 10-Q Note 2 at each date: $27.5B with 34% due within 12 months (3 Aug 2025), $164.6B with c.30% (3 May 2026), $179.2B with c.25% (2 Aug 2026). The twelve-month dollar amounts are computed from those percentages. The company cautions the figures “are not indicative of revenue for future periods”.

The step from $45B to $164.6B came from a single contract — “obligations under a long-term contract for custom AI accelerators entered in the fiscal quarter ended May 3, 2026”. Against that, the supply side has not moved to match the FY2028 ambition.

Non-cancellable purchase commitmentsAt 3 May 2026 At 2 Aug 2026Change
FY2026 remainder221,110+1,088
FY202755,21452,674−2,540
FY202872,87072,952+82
FY2029485+81
Total ($M)128,110126,821−1,289

Form 10-Q Note 12. Purchase commitments are, by the company’s own definition, “enforceable and legally binding” and exclude anything “cancelable without penalty”.

FY2028 commitments of $72.95B sit against a stated ~$230B of FY2028 AI revenue, and total commitments fell 1.0% in the quarter. Management’s gigawatt arithmetic is internally consistent — 30 GW across FY2027–28 at $20–30B of content brackets the $345B guided — but the commitment table is where that number gets settled, and it has not moved.
Guidance drift

The beat was entirely AI; both misses were not

June’s guidance, scored against what was delivered. Two of the four revenue lines missed, and both are the parts of Broadcom that are supposed to be uncorrelated with the AI build.

LineGuided in JuneDelivered OutcomeWhat it says

Guidance per the 3 June 2026 call; results per the 2 September 2026 call and the Form 10-Q. No sell-side consensus exists in the archive, so performance is scored against company guidance only.

Facts only · no interpretation

KPI dashboard

Every figure as filed or as stated on the call. Rows marked computed are the analyst’s arithmetic from filed data; the method is in the companion document.

MetricQ3 FY26Q2 FY26 Q3 FY25Read
Customer concentration, as disclosed in the 10-Q
per cent of net revenue · both up about ten points in two quarters · axis truncated at 25%
One distributor Top five end customers

Form 10-Q, Management’s Discussion and Analysis, “Net Revenue” concentration paragraph, each period. Distributor sales were 56% of nine-month net revenue.

Risk emergence

What the risk factors now say that they did not

The risk-factor summary bullets are identical quarter on quarter. The body text is not. These six changes were found by diffing the Q2 and Q3 filings sentence by sentence — open each to see the wording either side.

Part II, Item 1A of the Q2 FY2026 (filed 9 June 2026) and Q3 FY2026 (filed 10 September 2026) Forms 10-Q. Forty-five sentences were added and forty-three removed; these are the six that change what an owner should believe.

Actionable close

What to watch, and what would change the call

Five signals to track

  • Upfront licence revenue in the FY2026 10-K against $3,465M — the only disclosed way to separate software consumption from signature timing.
  • FY2028 purchase commitments against $72,952M, every 10-Q — the supply test for the $230B AI outlook.
  • Distributor share against 50% and top five end customers against 55%.
  • Gross margin against the ~73% Q4 guide, read with the XPU share of AI revenue against 73%.
  • The December board capital-allocation decision: dividend increase, buyback size, or neither.

Risks building

  • Concentration — one distributor at 50% of revenue, top five at 55%, both up about ten points in two quarters.
  • Customer credit — roughly $71B of support behind revenue booked at full value: a $29B backstop plus a new $42B facility under which the customer may issue Broadcom convertible notes.
  • Software deceleration — ARR growth 15% from 19%, consumption flat sequentially, first clear guidance shortfall.
  • Disclosure thinning — four series ended this quarter, a fifth already lapsed.
  • Non-AI semiconductors — the recovery called in June has failed to arrive twice.

Confirmation signals

  • FY2028 purchase commitments rising materially toward what $230B would consume.
  • Buybacks resuming above the rate of dilution after the December review.
  • Software meeting the $8.7B Q4 guide with ARR stabilising at or above 15%.
  • Backstop exposure staying at or below $29B as gigawatts deploy, nothing drawn.
  • The next Google TPU generation staying sole-source; non-AI revenue above $4.3B with wireless recovering.

Conviction tripwires

  • Any disclosed loss of an XPU programme generation to MediaTek, Marvell or a customer’s in-house team.
  • Any amount paid, or provision taken, under the Backstop.
  • Any issuance of the customer convertible promissory notes — facility $42B, currently drawn nil.
  • A reduction or withdrawal of the FY2027 $115B or FY2028 $230B figures. Management has said it will not update them quarterly, so any change is itself the signal.
  • Software revenue ex-upfront licence declining year on year, or purchase commitments falling materially while the AI outlook is maintained.

Source ledger

Not retrieved: the Q3 FY2026 press release itself (Exhibit 99.1). Figures published only there — non-GAAP EPS, segment margins, free cash flow — are taken from the earnings call. No sell-side consensus exists in the archive, so nothing here is scored against consensus. Q4 FY2026 reports on 9 December 2026.

Analytical companion to the Q3 FY2026 earnings update · prepared for Kun Xia · not a recommendation and not a valuation.