An 8-K is the form companies use to disclose material events between their regular quarterly and annual reports. When one introduces a defined financial metric, the definition deserves more scrutiny than any headline number, because the boundary of the metric determines what the number means. Broadcom's latest 8-K defines 'AI Revenue,' and the boundary it draws is the lesson.
The filing is the 8-K filed September 9, 2025 (sec.gov; accession 0001193125-25-199281). It defines AI Revenue as revenue with respect to the company's custom AI accelerators, XPUs, ASICs, and networking and connectivity solutions - a specific basket of products, named in the document.
Why does the definition matter more than the figure? Because a self-defined metric can be drawn narrowly or broadly, and only the definition tells you which. By including custom accelerators, XPUs, ASICs, and networking, Broadcom is counting both the compute silicon and the connectivity that ties AI systems together - a wide perimeter that reflects its two-sided AI franchise.
The inclusion of networking and connectivity is the most instructive part. Many observers think of AI revenue as just the accelerators; Broadcom's definition formally folds in the switching and interconnect that moves data between those accelerators. The definition encodes a thesis: in AI infrastructure, the network is part of the compute story.
For an explainer reader, the discipline is simple - when a company coins a metric, read its definition before its value. A number labeled 'AI Revenue' is only as meaningful as the product basket behind it, and that basket is spelled out in the 8-K, on the formal record.
The primary source is the sec.gov 8-K; it was surfaced through EdgarBeast, an SEC-filing evidence index. The takeaway: a defined metric is a disclosure choice, and the definition - accelerators, XPUs, ASICs, and networking - is where the real information sits.