If I had to pick one AI semiconductor outside Nvidia today, I'd probably lean Broadcom over AMD.

This isn't because I think AMD is a lesser company. Quite the opposite, AMD has executed incredibly well over the last few years. EPYC has taken real server share, Instinct GPUs are finally getting traction with hyperscalers, and Lisa Su has built one of the best turnaround stories in tech.

But Broadcom feels like it's playing a different game. AMD is still largely competing to sell more chips. Broadcom is helping build the AI infrastructure itself. It sits behind networking, switching, custom AI ASICs, high-speed connectivity and now VMware giving it another software layer on top. The more AI clusters get built, the more places Broadcom seems to show up.

Companies like Google, Meta and now even OpenAI are investing heavily in their own AI chips. Broadcom isn't competing against those projects, it's helping build many of them. The recent OpenAI custom inference chip announcement was another reminder of that.

That doesn't mean that AMD can't keep winning. If MI350 and future Instinct generations continue improving, there's still a massive AI accelerator market to grow into. AMD probably has more upside if it keeps taking share because it's coming from a much smaller base.

I just think Broadcom has more ways to win. AI networking, merchant silicon, custom ASICs, VMware, enterprise infrastructure, it doesn't feel like the company is betting on one product cycle. It's almost becoming the "behind-the-scenes" company for AI infrastructure.

Ironically, the biggest risk for Broadcom might just be expectations. The market already knows it's an AI winner, so every earnings report gets judged against an incredibly high bar. AMD feels more like it still has something to prove, while Broadcom has to keep proving it over and over again.

If I had to own only one for the next decade, I'd probably take Broadcom. Not because I think it'll build the best AI chip, but because it seems determined to make money no matter whose AI chip ends up winning.