Nvidia sells graphics processors, GPUs, to many buyers. Broadcom designs custom AI accelerators, a kind of application-specific chip or ASIC, with individual customers. The filings put numbers on both models: Broadcom booked $16.7 billion of AI semiconductor revenue in its quarter ended 2 August 2026, its release says, against $89,023 million of Data Center revenue at Nvidia in the quarter ended 26 July, per Nvidia’s 10-Q.
Here are five numbers from those filings and the customers’ own documents that explain the difference, with what each one measures and what it does not.
Business Pill · Vertical Integration
An evergreen Business Engineer explainer on vertical integration. It teaches the general idea only and does not discuss Broadcom, Nvidia or their filings.
The key insight: As we read the filings, the two models differ in who designs the chip and who carries the risk: Broadcom’s $16.7 billion of AI revenue is 34% of Nvidia’s $48,710 million Hyperscale line, by our arithmetic, and its 8-K puts its maximum potential Backstop liability on one custom-rack arrangement at approximately $43 billion.
The Two Business Models
Nvidia’s 10-Q names the kinds of direct customers it sells to, such as add-in board makers, distributors, original design and equipment manufacturers, cloud service providers, AI model makers and system integrators. Five direct customers accounted for 22%, 14%, 13%, 11% and 10% of its accounts receivable on 26 July 2026, the filing says.
Broadcom’s model is built around individual customers. Its 27 October 8-K refers to “AI racks based on custom AI accelerators designed by the Company”. OpenAI’s announcement of 13 October 2025 describes a plan to deploy “10 gigawatts of OpenAI-designed AI accelerators” with Broadcom, “co-developing next-generation systems and Ethernet solutions”.
Buyers can use both. Anthropic’s amended S-1 of 2 November says it trains and serves Claude “across four major chip families (AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs)”, and it discloses an equipment lease with Broadcom, entered in March 2026, “to procure specialized AI accelerators and associated components”.

Number 1: $16.7 Billion Against $89.0 Billion
Broadcom’s chief executive Hock Tan said in the release that “Q3 AI semiconductor revenue of $16.7 billion grew 221% year-over-year”. Nvidia’s Data Center revenue was $89,023 million, up 117% from a year ago, its 10-Q says. The dollar gap between the two lines was $72.3 billion, by our arithmetic. Our comparison of the two lines tracks the ratio over time.
Number 2: Nvidia’s $48.7 Billion Hyperscale Line
Nvidia now splits Data Center revenue in two. Hyperscale was $48,710 million and AI Clouds, Industrial, & Enterprise (ACIE) $40,313 million in the quarter ended 26 July, the 10-Q shows. Hyperscale was 55% of Data Center revenue, against 59% a year earlier, and ACIE 45%, against 41%, by our arithmetic.
The 10-Q says ACIE growth was driven by “end-demand from AI natives, enterprises, and sovereign customers, as well as hyperscalers utilizing AI clouds”. Both lines are recast: Nvidia changed the presentation in the first quarter of fiscal 2027 and moved one company from ACIE to Hyperscale in the second, its footnotes say.
Number 3: 34% of the Hyperscale Line
Broadcom’s $16.7 billion of AI revenue equals 34% of Nvidia’s Hyperscale revenue and 41% of its ACIE revenue, by our arithmetic. Over a year, Broadcom’s AI line grew 3.2 times, Nvidia’s Hyperscale line 2.0 times and ACIE 2.4 times, by our arithmetic from the stated figures.
The lines are not like for like. Broadcom’s AI figure includes networking as well as custom accelerators, as Tan’s comment on “custom AI accelerators and networking” describes, and the filings we read do not split the two.
Number 4: The $21.7 Billion Guide
For its fourth quarter, Broadcom expects “AI semiconductor revenue to accelerate to $21.7 billion, up 236% year-over-year”, the release says. That is $5.0 billion, or 30%, above the third quarter, by our arithmetic. Nvidia reports its own quarter on 17 November; our earnings preview sets out the consensus.
Number 5: The $43 Billion Backstop
Custom chips can come with financing. In its 27 October 8-K, Broadcom says it arranged for financial partners to take on part of certain agreements to buy AI racks and the related lease agreements with its customer, and that it provides residual value support to a new special purpose vehicle, the Backstop.
The filing says its potential liability equals any difference between “72.5% of the outstanding amounts owed on the lease” and the value of the racks on a sale, and that its maximum potential liability is “approximately $43 billion” on an undiscounted basis. That is 2.6 times its third-quarter AI revenue, by our arithmetic. Broadcom calls the Backstop “a contingent liability the Company believes has a low probability of occurring”. The 8-K does not name the customer. Our backstop comparison sets it against Nvidia’s.
The Structural Read
As we read Nvidia’s 10-Q, Hyperscale is the line where large cloud buyers show up, and it is the natural yardstick for a business whose AI line is described as custom AI accelerators and networking; the 34% is a revenue comparison, by our arithmetic, not a share of chips.
As we read Broadcom’s 8-K, the Backstop shows custom accelerators can come with financing attached: its liability is tied to 72.5% of the outstanding lease amounts, against the value of the racks on a sale.
Hock Tan, Broadcom, 2 September 2026
“In Q4 the momentum continues, and we expect AI semiconductor revenue to accelerate to $21.7 billion, up 236% year-over-year.”
Three Implications
FOR BROADCOM Its guided $21.7 billion of fourth-quarter AI semiconductor revenue is $5.0 billion above the third quarter, by our arithmetic from its release.
FOR NVIDIA Its 17 November results replace the $48,710 million Hyperscale and $40,313 million ACIE figures used here, per its release timing.
RELATED Our Broadcom and Nvidia comparison tracks the ratio of the two lines.
The Business Engineer Lens
This story maps onto the Business Engineer framework The AI Supercycle.
The framework argues: “Three competitive categories within one layer: generalist substrate (NVIDIA), hyperscaler custom (TPU, Trainium, MTIA), merchant custom challengers (Cerebras, Groq, SambaNova).”
As we read it, the filings here put revenue numbers on the first two categories: Nvidia’s $89,023 million Data Center line and Broadcom’s $16.7 billion of AI revenue, which its release describes as custom AI accelerators and networking. The framework predates these filings and does not discuss these figures.
What Is Not Established
The filings do not give a revenue figure for custom accelerators alone, so none of these numbers is a market share of chips. The fiscal quarters end on different dates, about a week apart for these two companies. Customer names here are only those in the customers’ own announcements and filings.
Broadcom’s guide is its own forecast, and Nvidia’s 17 November report will replace the July-quarter figures. The figures come from SEC filings and company releases. The companies were not contacted for this piece. Our valuation comparison, AMD comparison, frontier-lab deals piece and Nvidia hub cover the rest of the story.
The Bottom Line
Broadcom’s $16.7 billion of AI revenue is 34% of Nvidia’s $48,710 million Hyperscale line, by our arithmetic, it guides to $21.7 billion next, and one of its custom-rack arrangements carries a maximum potential liability of approximately $43 billion, its filings say.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Broadcom’s AI semiconductor revenue, guide and quotes are from its results releases for the third quarters of fiscal 2026 and 2025, filed with the SEC as 8-K exhibits, and the Backstop from its 8-K of 27 October 2026. Nvidia’s segment figures are from its 10-Q for the quarter ended 26 July 2026. OpenAI’s plan is from its own announcement and Anthropic’s chip use and Broadcom lease from its amended S-1. Shares, ratios and multiples not stated in the filings are our arithmetic. The companies were not contacted for this piece.
Sources: Broadcom Q3 FY2026 results, SEC, 2 Sep 2026 · Broadcom Q3 FY2025 results, SEC, 4 Sep 2025 · NVIDIA 10-Q, quarter ended 26 Jul 2026, SEC · Broadcom 8-K, SEC, 27 Oct 2026 · OpenAI and Broadcom announcement, 13 Oct 2025 · Anthropic S-1/A, SEC, 2 Nov 2026









