Broadcom
Table of contents
Thesis
- The verdict: scale economies, dominant positioning, customer captivity.
- Business: a capital allocator that owns 1) custom silicon design + networking content, 2) infrastructure software.
- Peers: 1) Marvell, 2) Nutanix.
- Moat:
- Silicon: SerDes, optical and packaging IP, design wins, capacity priority driven by volume.
- VMWare installed base and certification dependency.
- Competition: networking is more contested.
- Growth drivers: custom AI accelerator volume, attached networking and optical content, repricing the VMWare installed base into subscription.
- Economics:
- ✅ capex light (fabless design) -> high and rising GM and OM (partially determined by mix).
- ⚠️ Backlog is lengthening, not converting faster.
- Risks:
- Customers own the design and are multi-sourcing.
- Concentrated customers with uncertain funding.
- Credit risk associated with customer backstop.
- the rack and leasing model degrades the economics.
Broadcom vs Nvidia
- Nvidia has faster growth from a larger base, driven by buyer breath.
- Nvidia has larger operating leverage.
- Nvidia has stronger pricing power and cost pass through.
- Nvidia is also more supplier-resilient.
- Broadcom's customer concentration is increasing while NVIDIA's is decreasing.
- Cost structures are similar.
- Downside beta: Broadcom breaks first, NVIDIA breaks deeper.
- Market share: Broadcom is taking share from merchant GPU in aggregate while donating share inside its own largest programme. NVIDIA is donating aggregate accelerator share while taking networking share.
Broadcom vs Marvell
Broadcom is the materially stronger business for the next five to ten years. Broadcom carries a 31.4bn dollar software annuity at 94% gross margin behind an AI franchise almost six times Marvell's entire data-centre business, while Marvell carries 2.3bn dollars of declining communications revenue behind an 81%-data-center business it is paying its largest customer 6.4% of its equity to expand.
