Invest4 distinct publishers3 min readPublished
Revenue grew 86% and the fourth-quarter guide calls for 93%, which leaves the capex-digestion thesis with little to stand on, though the sequential dollar increment is already about 15% smaller than last quarter's.
The Investor · Invest desk
Compiled by The InvestorSomething wrong?How this is made
Do the sequential arithmetic by hand, because that is where this story sits. If the $16.7bn of AI semiconductor revenue was 54% above the immediately preceding quarter [4], that quarter was about $10.8bn [1], so Broadcom put on roughly $5.9bn of AI revenue in three months [2]. The fourth-quarter guide of $21.7bn adds $5.0bn to it, about 30% sequentially [5][3]. The annual rate accelerates from 221% to 236% while the increment producing it comes in about 15% smaller [4], and the 93% headline on total revenue is flattered by a prior-year quarter of roughly $18.0bn [3][13].
Strip the AI line out and very little else moves. Implied non-AI revenue goes from $12.9bn in the reported quarter to $13.1bn in the guide [5]. Broadcom designs these parts for Google, Meta and OpenAI [19], which is the mechanism that decides how much the print is worth as evidence: a fabless supplier reports orders placed several quarters ago being filled now, so this quarter's print is a trailing record of decisions those customers made a while back, arriving now as filled orders, not a look at what they're committing to this month.
The share reaction is arithmetically sane. Guidance landed $230m under the $35.03bn LSEG consensus that Reuters compiled, or 0.7% short [9][7], and a quarter that beat on revenue and on adjusted EPS at $3.32 against $3.24 [12] lost the day to that. Annualise the guide at the guided 66% non-GAAP operating margin and you get about $23.0bn a quarter, up from $20.1bn reported [6], call it $92bn a year [7][10], which puts a $1.8tn market value at roughly 19.6 times [15]; annualise this quarter's $13.7bn of free cash flow and the same value is about 33 times [8][12]. At those multiples, the acceleration is already priced in, which is why the shares are up 6% in 2026 against the S&P 500's 12% while revenue nearly doubled [15]. The company sent $3.1bn out as dividends during the quarter, 23% of its free cash flow [18][14].
This read could still be wrong, and the likeliest fault lines run through supply, customer concentration and demand itself. Broadcom has been supply-constrained and signed a Samsung manufacturing agreement valued above $200bn in July [17], so the guide may describe what it can build; Marvell's custom-chip partnership with Google, under which the customer could contribute $120bn of revenue through fiscal 2033 [16], makes the largest custom account contestable rather than annuitised; or digestion is real and a shrinking increment is what its first print looks like. My read is that the flat margin argues against the capacity story, since a supplier rationing scarce wafers tends to price them and this one guided 66% flat year over year [7], which makes the guide a demand signal and the best evidence available that the order book is still filling. The falsifier is specific: if the first-quarter sequential AI add lands under $5bn, the ramp has topped, whatever the year-over-year percentage says.
Ranked by verification strength, evidence, and original report placement.
Broadcom reported revenue of $29.6 billion for the third quarter of fiscal 2026, ended August 2, 2026, up 86 percent from the prior year period, a record.
Broadcom's revenue in the year-earlier quarter was $15.95 billion.
CEO Hock Tan said third quarter AI semiconductor revenue of $16.7 billion grew 221% year-over-year and 54% quarter-over-quarter, and that demand for custom AI accelerators and networking continues to be very strong.
Third quarter revenue of $29.6 billion came in ahead of the $29.36 billion analysts expected, and adjusted EPS of $3.32 beat estimates of $3.24 and was up 96% from a year earlier.
Broadcom guided fourth quarter fiscal 2026 revenue to approximately $34.8 billion, an increase of 93 percent from the prior year period.
Broadcom expects fourth quarter AI semiconductor revenue to accelerate to $21.7 billion, up 236% year-over-year.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 2, 2026
2 articles · September 2, 2026
1 article · September 2, 2026
1 article · September 2, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Broadcom's $100B Chip Raise Turns AI Capacity Into a Credit Question1 distinct publisher
invest
The AI moat is now a balance sheet, so price the financing and not the model1 distinct publisher
invest
Nvidia is now VOO's biggest holding, and no index owner voted for it1 distinct publisher
invest
Nvidia stops eating memory costs: AI server prices up more than 15% on early-2027 shipments1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Audited-quarter numbers, single origin
The financial spine here is as solid as reporting gets: a filed quarter, published by the company, with revenue, operating income, cash flow and segment detail that three independent publishers reproduce without contradiction. The soft joint is the yardstick, not the results. Consensus arrives as $35.03bn from LSEG and $35.05bn from Bloomberg, EPS expectations as $3.24 and $3.23, revenue expectations as $29.36bn and $29.45bn, and the AI comparison rests on a single StreetAccount figure only CNBC cites. Everything forward-looking, the $21.7bn AI guide and the 66% margin, is management's own estimate with no GAAP reconciliation offered.
Demand showing up as booked revenue
This is adoption you can count rather than infer. Custom accelerators and networking produced $16.7bn of revenue in thirteen weeks, more than tripling year over year and beating an outside estimate by $1.5bn, and the company has committed to $21.7bn next quarter. Supporting signs point the same way: OpenAI's Jalapeno part exists, Apple is raising US production spend, and Broadcom was constrained enough to put more than $200bn of manufacturing behind Samsung. The reservation is concentration. A handful of named buyers, Google, Meta, OpenAI, account for a line that is now generating essentially all of the company's growth, and Marvell's Google arrangement shows those buyers keep second suppliers.
Headlines undersell the growth, oversell the miss
The framing runs against the arithmetic in both directions. A shortfall of $230m, seven tenths of a percent, is described as 'disappointing' and 'weak' while the AI line beat its own estimate by six and a half times that amount, so the disappointment is amplified and the growth understated. The genuinely bearish detail is the one nobody printed: the sequential AI increment guided for next quarter, about $5.0bn, is roughly 15% smaller than the $5.9bn just delivered, even as the year-over-year rate climbs to 236%. Overstated tone, understated substance, and the one number that would sharpen either case left on the floor.
The issuer sets the frame it is judged in
Broadcom wrote the document everyone is quoting, chose which metric to headline, and picked the comparison that flatters it most: AI revenue 'accelerating' on a year-over-year basis, margin 'flat from a year ago', records in revenue, profit and cash flow. It also declines to reconcile its guided non-GAAP margin to GAAP, citing unreasonable effort, having excluded stock compensation, amortisation and restructuring from the $20.1bn figure. On the other side, the sell-side voice in this story is a broker analyst who follows a comment about selling pressure with 'it's just a matter of when do you enter' — a positioning view, not an independent check on the numbers.
Firm on the quarter, thin on the trajectory
What happened is not seriously in doubt: a filed quarter, four accounts, no material disagreement on any disclosed figure. Where confidence drops is everything the story is actually about. The next quarter is a company estimate, the consensus it was judged against differs by publisher, the AI comparison rests on one analyst survey, and the competitive and supply context that would let a reader judge durability appears in exactly one of the five accounts.