Broadcom's custom silicon for AI is positioned as a multi-billion-dollar growth driver within the AI supercycle.

For years, graphics processing units (GPUs) have dominated artificial intelligence processing. They excel at training AI and executing a wide array of tasks.

Recently, however, a shift has emerged toward custom processors, known as application-specific integrated circuits (ASICs). These processors can be tailored to specific AI models or agentic tasks, offering greater cost-efficiency and speed than general-purpose GPUs.

According to Cathie Wood's Ark Invest, custom processors could account for more than one-third of AI compute by 2030, up from single-digit percentages today. Analysts at Omdia predict custom ASICs will exceed GPU volumes by 2028.

One of the most effective ways to capitalize on this rapidly expanding market is through shares of Broadcom (NASDAQ: AVGO). Here's how Broadcom is positioned to benefit from the AI supercycle.

AI leaders are lining up to purchase Broadcom's custom chips. Custom AI silicon is not entirely new—Alphabet's Google developed its Tensor Processing Units (TPUs) over a decade ago. However, even Google is expanding its use of these processors, and other tech giants are following suit.

Alphabet recently signed a multi-year agreement with Broadcom to design TPUs for its Google Gemini frontier AI model. As Broadcom's largest customer, Alphabet's custom processor purchases accounted for an estimated $13 billion of Broadcom's 2025 sales.

Meta Platforms represents another significant customer, with Broadcom designing its Meta Training and Inference Accelerator (MTIA) chips. AI has become increasingly central to Meta's strategy, particularly following the launch of its Muse agent app, which autonomously shops online for users, books trips, and manages daily tasks. Muse quickly reached the top of Apple's U.S. App Store last month, demonstrating strong consumer demand for personal AI agents.

OpenAI joined the custom AI processor market this year by engaging Broadcom to design its Jalapeno processor, an LLM-optimized inference chip that enhances AI performance and reliability. Initial processors are scheduled for data center deployment by year-end, with scaling in 2027 and 2028.

Anthropic has been a Broadcom customer for at least two years, procuring custom processors valued at tens of billions of dollars. The companies recently expanded their relationship, with Broadcom agreeing to provide Anthropic up to $42 billion in financing for artificial intelligence data center infrastructure expansion.