Product1 distinct publisher3 min readPublished
The semiconductor unit tripled to $16.7bn and cleared the Street by $1.5bn, yet the guide for the current quarter landed 0.66% light, which is what happens when revenue arrives on someone else's gigawatt schedule.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Somewhere there is a capacity planner with a row in a spreadsheet that says TPU8i and a delivery date lifted from an earnings call rather than a contract. That row is what this story is actually about.
Start with the size of the disappointment. Broadcom guided the current quarter to $34.8bn at the midpoint against a Street number of $35.03bn [6], a shortfall of $230m, or 0.66% [1]. That same midpoint implies 17.6% growth on the $29.59bn just reported [2]. Semiconductor revenue, meanwhile, beat its target by $1.5bn, close to 10% [3]. A supplier lands that near consensus and still takes its stock down almost 2% [7] when the argument is about when revenue arrives rather than whether it does.
What a customer actually does with a multi-gigawatt commitment is re-time it as power and buildings move underneath the plan. What a quarterly guide implies is a smooth monthly quantity. Look at how CEO Hock Tan described the pipeline on the call, as reported by SiliconANGLE: about 5GW of Google's TPU8i at Anthropic next year, with line of sight to another 10GW [11], and about 1.3GW of Jalapeno at OpenAI in 2027, with line of sight past 5GW [13]. The named near-term figures total 6.3GW [4]. The line-of-sight figures add at least 15GW on top, roughly 2.4 times the named amount [5]. Line of sight is not an order, and the space between the two is the variable that produces a 0.66% miss.
Tan's forward numbers deserve a scale check rather than a nod. He told analysts AI revenue would double to $115bn next year and double again to $230bn after that [14]. Annualise the quarter just reported and the company is running at $118.4bn [6], so the AI line alone would be about 97% of the current whole-company run rate next year and roughly 1.9 times it the year following [7]. The published account does not reconcile that with a $16.7bn semiconductor quarter [1], and it names two segments whose sum, $25.45bn, sits $4.14bn under the $29.59bn total [8]. If you are modelling off this call, that reconciliation comes before the doubling does.
Then the part that moves risk between balance sheets. Tan said Broadcom is considering residual value guarantees, which are contingent liabilities, for the two AI labs [16], and CFO Amie Thuener described the company as empowering its two most strategic customers to bridge the gap between their current cash flow and the upfront investment their businesses require [15]. In plain terms, the supplier would be taking a position on what the hardware is worth if the lab stops wanting it.
For anyone whose roadmap sits downstream of this, sort your supply into four boxes using two questions: is the capacity contracted or line-of-sight, and who absorbs the cost of a six-month slip. If it is contracted and the supplier absorbs the slip, you can publish dates. If it is contracted and you absorb it, publish dates and hold a buffer you have priced. If it is line-of-sight and the supplier absorbs it, treat the number as useful colour and nothing more. If it is line-of-sight and you absorb it, you are in the box where product dates get invented, and that is where a lot of inference roadmaps are sitting this week.
Ranked by verification strength, evidence, and original report placement.
Broadcom's semiconductor business unit revenue tripled from a year earlier to $16.7 billion, surpassing the Street's $15.2 billion target.
Broadcom reported quarterly sales of $29.59 billion, up 86% from $15.95 billion in the same period a year earlier, against analyst models of $29.36 billion.
Broadcom reported adjusted earnings of $3.32 per share, against Wall Street models of $3.24 per share.
Tan said Broadcom is considering providing "residual value guarantees" that are contingent liabilities to the two AI labs.
Net income rose by more than triple to $13.09 billion, from $4.14 billion a year earlier.
Broadcom's infrastructure business added $8.75 billion in sales, just below the $8.82 billion consensus estimate.
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · September 2, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Six customers underwrite Broadcom's $230bn AI chip target for 20282 distinct publishers
invest
Broadcom's AI chip revenue climbed $5.9bn sequentially to $16.7bn5 distinct publishers
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
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.
One filing, one outlet, twice
The $16.7 billion semiconductor line, the $13.09 billion net income, the $34.8 billion guide and every gigawatt in the story trace to a single Broadcom release and call, relayed by SiliconANGLE and then posted again in near-identical form. The reported figures do reconcile internally — the two named segments plus a $4.14 billion unnamed remainder add to $29.59 billion — which tells us the transcription is careful and nothing about whether the forward claims hold. No analyst, customer or filing check appears anywhere in this coverage.
Cash collected, gigawatts promised
This is not a story about pilots. A semiconductor unit that tripled to $16.7 billion is chips shipped and invoiced, Jalapeno is a real part with a second spin heading to tape-out, and Apple is putting more money into Broadcom's U.S. production. What keeps this short of the top of the range is that everything past the quarter — 5 gigawatts at Anthropic, 1.3 at OpenAI in 2027, Meta's MTIA moving to production — is future tense, described by the supplier rather than the deployer.
The promise outgrows the quarter
Positive, and the gap is quantifiable using only the story's own numbers. Broadcom currently annualises around $118 billion in total revenue; Tan says AI alone reaches $115 billion next year and $230 billion the year after. Meanwhile the nearest thing management actually controls — the guide for the quarter it is already in — came in $230 million light, and the market marked the stock down. When a soft guide is answered with a doubling and then another doubling, and when the supplier is simultaneously weighing residual value guarantees to help its customers pay, the forward figures are carrying more weight than the evidence beneath them.
The forecast is the message
Almost every forward-looking sentence here belongs to a CEO whose stock is up 6% while the index is up 12%, delivered on the call where that gap gets addressed. The gigawatt figures describe customers' plans, not Broadcom's shipments, which is a convenient way to size a market you do not control. The CFO's line about bridging customers' cash-flow gap is corporate framing for vendor financing, and it went into print as strategy. On the publishing side, SiliconANGLE's pages carry its own community and marketplace solicitations — a general commercial posture rather than anything specific to Broadcom, but worth knowing when the coverage reproduces the call with so little friction.
Solid on the quarter, thin past it
Split the story in two and the confidence splits with it. The reported quarter is specific, arithmetically coherent and checkable against a filing anyone can pull, so those numbers should hold. Everything from the analyst call onward — customer gigawatts, tape-out timing, the 2027 trajectory, the guarantees under consideration — comes from one speaker, in one room, relayed by one outlet publishing the same text twice. That is enough to report; it is not enough to plan against.