Broadcom Inc. AVGO Moat

Key data

The moat

A hyperscaler that commits a generation of custom accelerator to Broadcom is committing a multi-year program: the serializer/deserializer blocks, the high-bandwidth memory and SRAM integration, the advanced packaging, the mask sets, and then the Ethernet switch and optical layer that carry traffic between the parts. Walking away means re-qualifying the rack, not swapping a component. This is a proprietary technology moat, held with high confidence at the silicon layer, resting on roughly 19,000 issued patents and 2,170 pending applications as of November 2, 2025, with expirations spread from 2025 to 2044 and no material single-patent dependency. The secondary moat is switching costs in infrastructure software, where VMware Cloud Foundation sits under mission-critical private cloud at most of the Fortune 500 and migration is a project rather than a purchase.

Broadcom is the only merchant supplier that can hand one customer a custom accelerator, the interconnect IP feeding it, the 100-terabit switch connecting thousands of them and the optical DSP at the rack edge from a single library. Swap in Analog Devices, ASML or Advanced Micro Devices and that sentence collapses: each holds one layer at most. Third-party estimates put Broadcom near 55% of merchant data center switch silicon and about 60% of the custom AI ASIC market in 2026. What it produces is price: gross margin 44.9% in FY2016, 55.2% in FY2019, 66.5% in FY2022, 67.8% in FY2025.

Widening or narrowing

The ten-year gross margin sequence runs 44.9%, 48.2%, 51.5%, 55.2%, 56.6%, 61.4%, 66.5%, 68.9%, 63.0%, 67.8%. The FY2024 dip is acquisition accounting and software mix, not competitive loss, and it reversed in one year. Semiconductor revenue compounded from $17.27 billion in FY2020 to $36.86 billion in FY2025 without a down year.

The profit pool sits in three places: leading-edge wafers, memory and substrates upstream; design IP in the middle; box assembly downstream. Broadcom holds the middle slice at 67.8% gross margin against ASML at 52.8%, Analog Devices at 61.5% and Advanced Micro Devices at 49.5% in their latest fiscal years.

The overrated case is counterparty power. One customer went from 20% of revenue in FY2022 to 21% in FY2023 to 28% in FY2024, and the top five went from 35% to 35% to 40%. The claims of a 200-terabit tape-out, portfolio leadership and "very deep moats" come from the September 2, 2026 call and are asserted, not proven. Verdict: widening.

What breaks it, and who

Nvidia is moving down into Ethernet from the system side. Its data center Ethernet switch revenue reached $2.1 billion in calendar Q1 2026, up 192.7% year over year, for 21.5% of that market against a market growing 39.8% to $15.4 billion.

Marvell is the specific number two in custom silicon, estimated at 30 to 35% of custom XPU against Broadcom's 60%. More telling, MediaTek has secured design partnership on an inference-focused Google TPU generation, which is erosion inside the single largest relationship rather than at the edges.

The software half is under live regulatory attack. CISPE filed an EU antitrust complaint in March 2026 with a request for interim measures, the Commission sent cloud providers fresh questionnaires in July 2026 on how replaceable VMware products are, and the fourth ECCO report in September 2026 recorded renewal cost increases of ten times or more as widespread.

The read changes on three items: FY2026 and FY2027 disclosure of top-five and largest-customer concentration, whether the EU opens formal proceedings or accepts commitments on VMware licensing, and whether the Tomahawk 7 generation ships into customers already using Broadcom accelerators.

Closing thoughts

The moat is real and widening at the silicon layer, conditional in software. Margin, per-head revenue and semiconductor revenue all move the right way across multiple years, and the measured share positions at 55% of merchant switch silicon and 60% of custom ASIC come from outside the company rather than from its own slides. Against that, 40% of revenue sat with five customers in FY2024 and 28% with one, and those customers fund the alternatives. The checkable item is the largest-customer percentage in the next two annual filings read beside XPU unit growth, because a flat concentration figure with rising revenue means new programs, while a rising one means dependence. The moat strengthens if networking attaches to new accelerator programs at the rate management describes and weakens if a second hyperscaler awards a flagship generation to MediaTek, Alchip or Marvell.