Broadcom edges Nvidia for 5-year AI chip buys: $115B target

The Motley Fool's comparison starts from a simple observation: Nvidia and Broadcom are two of the biggest names in AI investing, and both supply a massive amount of the computing power AI companies need. But they get there from different directions. Nvidia recently started shipping its latest AI architecture, Vera Rubin, which is built around Vera central processing units and Rubin graphics processing units and is described as far more advanced than the previous Blackwell platform. When tasked with inference, it reduced token cost by a factor of 10, and for training it cuts the number of GPUs required for a given workload by 75%.

Broadcom's client list includes hyperscalers Alphabet and Meta Platforms, as well as AI labs such as Anthropic and OpenAI. The analysis says these computing units can outperform GPUs in both performance and cost when running a properly configured workload that suits their design, but would fall flat outside their core competency. The author expects the future to be a mixture of both approaches, perhaps leaning more heavily toward Broadcom's style, which is why Broadcom gets a slight edge.

That edge shows up in the guidance each company has given. Nvidia expects 70% revenue growth in its upcoming fiscal year. Broadcom expects even more rapid AI semiconductor growth: it told investors it should double its AI semiconductor revenue to $115 billion in 2027, and then double it again to $230 billion in 2028. The author projects major AI data center growth over the next five years and concludes that splitting an investment between the two companies would make sense, while noting Nvidia processors can do plenty that Broadcom's cannot.