Research Broadcom's 10-K as part of the AI bottleneck investment thesis.
Focus areas:
- Custom ASIC / XPU business (Google TPU, Meta MTIA, Apple neural engine contracts)
- VMware integration progress and margin profile
- AI networking (Tomahawk, Jericho) alongside Arista
- Semiconductor segment revenue breakdown and AI-specific growth
- Customer concentration risk (Apple ~20% of revenue)
- Capital allocation: debt load from VMware acquisition vs. AI capex tailwind
- Competitive position vs. Marvell in custom silicon
Key question: Is Broadcom's custom ASIC story (hyperscaler in-house chips) a threat to Nvidia or a different market? And how much of the AI revenue surge is durable vs. one-time design wins?