Re-Engineering RiskWide moat

Re-engineering risk is what makes a design win stick: to swap out one of Broadcom's chips, a customer must not merely buy a different part but redesign the product around it — new circuit boards, new software, new testing, new qualification. That work is expensive, time-consuming, and risky, and for a component that is functioning perfectly well, few customers can justify it. The cost of change, not any contract, is what keeps the customer buying.

The deeper the chip is woven into the product, the higher the wall. A Broadcom part that interfaces with custom software, sits at the heart of the system's function, and has been tuned over a product's life becomes something the customer's own engineers understand and depend on; replacing it means unpicking years of accumulated integration. The switching cost is not a single number but a tangle of engineering hours, schedule risk, and the ever-present fear that the new part will misbehave in some way the old one never did.

The limit is that re-engineering risk deters change during a product's life but does not prevent it at the next redesign, when the customer is rebuilding the product anyway and the incremental cost of switching suppliers is far lower. And a large enough incentive — a big price gap, a meaningful performance edge, a strategic desire to escape a supplier — can justify even a costly re-engineering. So this is a powerful source of stickiness within a product cycle, and a wall that comes down, at least partway, whenever the customer starts the next design with a clean sheet — a cycle the ~$20B AI franchise faces every generation¹.