Broadcom remains one of the best AI infrastructure stocks to own.
Broadcom (NASDAQ: AVGO) continues to pile up the wins. Last week, the company extended a deal with Meta Platforms (NASDAQ: META) to help the social media giant develop new custom AI chips. The initial commitment is for 1 gigawatt worth of chips, and the agreement will go through 2029. However, this is expected to be just the start of a multi-generation custom chip roadmap. Given the scope of the deal, Broadcom CEO Hock Tan will also step down from Meta's board and move to an advisory role to help the company with its custom chip roadmap.
Meta also just revealed four generations of custom MTIA (Meta Training and Inference Accelerator) chips, developed with Broadcom's help, last month. The MTIA 300 is already being used for ranking and recommendation training, while the upcoming MTIA 400, 450, and 500 can handle all AI workloads but have been designed specifically with inference in mind.
Will AI create the world's first trillionaire? Our team just released a report on the one little-known company, called an "Indispensable Monopoly" providing the critical technology Nvidia and Intel both need. Continue »
Earlier this month, Broadcom also extended its partnership with Alphabet to help it develop future iterations of its highly popular Tensor Processing Units (TPUs). The duo then inked a deal to supply Anthropic with an additional 3.5 gigawatts worth of the chips starting in 2027. Broadcom already has $21 billion in TPU orders from Anthropic for this year.
Before these announcements, Broadcom had said it was on pace to deliver $100 billion worth of custom AI chips to customers in fiscal 2027. Given that the company had around $20 billion in total AI revenue, including networking in fiscal 2025, and under $64 billion in total revenue, that's a staggering amount. Meanwhile, these extended deals are only going to keep the growth coming in the years ahead.
Broadcom's custom chip business also directly feeds into its data center networking business. The company is a major player in networking, where its Tomahawk Ethernet is a leading switching solution. As AI chip clusters grow, this business is booming.
Hyperscalers (owners of large data centers) are increasingly looking for alternatives to Nvidia's high-priced graphics processing units (GPUs). Custom AI chips are becoming one of the best options for hyperscalers to pursue, especially for inference. Custom AI ASICs (application-specific integrated circuits) can provide both performance and cost advantages, as these hardwired chips are purpose-built and more energy-efficient. Meanwhile, given Broadcom's track record in the space, more and more companies are turning to it for help.