Broadcom's newest analyst coverage shows how quickly the investment case has shifted. BMO initiated the stock at Outperform with a $455 target, while RBC later maintained Sector Perform at $400. The gap is not enormous compared with some technology names, but the reasoning behind it is important: Broadcom is increasingly being valued as a strategic supplier to custom AI infrastructure rather than simply a diversified semiconductor company.
That change matters because custom accelerators can create large, concentrated programmes tied to a handful of hyperscale customers. The revenue can be substantial, but visibility depends on design wins and customer capital-expenditure plans that sit outside Broadcom's control.
Custom silicon changes the competitive map
Hyperscalers have strong incentives to design more of their own computing stack. Custom chips can reduce cost, optimise specific workloads and lower reliance on merchant accelerators. Broadcom benefits when it helps turn those designs into deployable silicon and supplies the networking around them.
This does not mean custom chips replace Nvidia broadly. It means part of AI spending can flow through a different channel, particularly for mature or specialised workloads.
Networking may be the less glamorous part of the bull case
Large AI clusters need switches, interconnects and networking components as much as they need accelerators. That gives Broadcom exposure to the growth in cluster size even when the company is not supplying the main compute device.
The networking opportunity can also be more diversified across architectures. Investors focused only on custom accelerator headlines may be underestimating how much value sits in the surrounding infrastructure.
The caution case is customer concentration
A small number of cloud companies account for a disproportionate share of AI capital expenditure. That concentration can produce rapid growth when budgets expand and rapid estimate revisions when one customer delays a programme.
The more aggressive Broadcom targets effectively assume that hyperscaler spending remains durable and that custom silicon gains share inside it. The more cautious targets do not require a collapse in AI demand. They simply assign less certainty to that outcome.