Marvell vs Broadcom

Which ASIC player is worth your money

As we enter year three of the AI data center super cycle the market has started looking beyond just Nvidia’s CUDA GPU dominance and into what should be the next winner of the AI infrastructure semi space, ASICs. ASIC stands for Application Specific Integrated Circuit and is a chip designed for a specific use case instead of a general purpose use that can be used for different tasks. These custom designed silicon are built with the hope that they can preform tasks with less energy consumption, cost less to manufacture, and most importantly outside of huge upfront CapEx cost to design the initial chip cost significantly less every upgrade cycle because you’re keeping everything in house versus paying Nvidia.

The major players are all neck deep in trying to remove Nvidia from their supply chain as much as possible. Google already has theirs called TPUs, Amazon is building ‘Trainium’. Everyone wants to control the backbone of the silicon stack. Designed for the best latency, throughput, and efficiency that they can achieve, all at a longer term cheaper cost.

ASICs don’t just fall out of the sky, there are companies that help the big players design and create them. Then these contracts become ongoing as they continue to upgrade the silicon at a regular clip. To constantly stay on the cutting edge of tech. There are three major companies that focus on this space. Broadcom (AVGO), Marvell (MRVL) and Astera Labs, Inc (ALAB). In today’s missive we’re going to focus on the difference between Marvell vs Cap- I mean Broadcom AI offerings, what they do differently, their growth and valuation expectations and which one I like more.