The Software Customers Are a Different CompanyWide moat
Broadcom (AVGO) — moat facet
Broadcom has two customer bases that share nothing, and the enterprise one is the actual argument for the multiple.
The single most underappreciated fact about Broadcom's customer base is that there are two of them, and they have essentially nothing in common.
The semiconductor business sells roughly $36.9 billion to a very small number of enormous buyers — hyperscalers commissioning custom accelerators, networking OEMs, handset makers — in long design cycles with lumpy, program-driven purchasing. The infrastructure software business sells roughly $27.0 billion, about 42% of revenue, to tens of thousands of enterprises and governments paying annual subscriptions to keep existing systems running1. One base is concentrated, technical and cyclical; the other is fragmented, contractual and recurring.
This is the actual argument for Broadcom's multiple, and it is better than the one usually made. The software base does not care about the AI capital cycle; an enterprise renews VMware because its applications run on it, whatever hyperscaler capex is doing. That is a genuine counterweight to a semiconductor customer list where one relationship was 32% of revenue last year and 50% in the latest quarter2.
The counterweight has a condition attached: it holds only while the software customers stay. They are being repriced aggressively — the subject of a root threat and of the Nutanix page under Competitors — and a customer base held by switching costs rather than affection is exactly the kind that leaves slowly and then all at once. Watch software revenue growth. If it stalls while semiconductor revenue swings with the AI cycle, Broadcom has one customer base again, not two.
The two-customer-base structure is intact and is Broadcom's best structural answer to semiconductor concentration: roughly $27 billion of subscription revenue that does not care what hyperscaler capital spending does. Stable rather than widening, because the repricing that makes this base so profitable is the same force pushing some of it out the door. Watch software revenue growth, not customer counts.
The software base is the part that does not move with the AI cycle. A falling share is arithmetic while chips triple - but if it falls because software stops growing, the ballast is gone.
- ReportedThe infrastructure software business sells roughly $27.0 billion, about 42% of revenue, to tens of thousands of enterprises and governments paying annual subscriptions to keep existing systems running.Broadcom, Form 10-K FY2025 + earnings (rev ~$64B +24%; AI semiconductor revenue ~$20B +65%; ~$73B multi-year AI backlog; VMware segment growth) — FY2025 (ended Nov 2025) · publ. Dec 2025 · source ↗
- ReportedThat is a genuine counterweight to a semiconductor customer list where one relationship was 32% of revenue last year and 50% in the latest quarter.Broadcom Inc., Form 10-Q for the quarter ended 2 August 2026 - remaining performance obligations approximately $179.2 billion, including a long-term contract for custom AI accelerators entered in the quarter ended 3 May 2026, about 25% expected within 12 months; one semiconductor customer, a distributor, 50% of net revenue in the quarter (32% a year earlier) and top five end customers about 55%; distributors 56% of net revenue for three quarters; purchase commitments $126,821M ($52,674M for fiscal 2027, $72,952M for fiscal 2028); operating income $15,955M (54% of revenue) against $5,887M (37%); infrastructure software revenue $8,752M (+29%) and operating income $7,325M (+40%), attributed to VMware Cloud Foundation including additional license revenue; subscriptions and services revenue $5,312M against $4,779M, products $24,279M against $11,173M — Quarter ended 2 August 2026 · publ. 2026-09-10 · source ↗