For years, graphics processing units (GPUs) have been the dominant processor for artificial intelligence (AI). That's because GPUs are great at training AI and helping it perform a wide array of tasks.
But lately, there's been a shift toward custom processors, called application-specific integrated circuits (ASICs). These custom processors can be designed for specific AI models or specific agentic tasks, making them more cost-efficient and faster than using GPUs for all AI tasks.
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. And analysts at Omdia predict custom ASICs will exceed GPU volumes by 2028.
One of the best ways to tap into this rapidly expanding market is by owning shares of Broadcom (AVGO +0.19%). Here's how Broadcom is benefiting from this AI supercycle.
Image source: The Motley Fool.
AI leaders are lining up to buy Broadcom's custom chips
Custom AI silicon isn't exactly a new idea. Alphabet's Google has had its own Tensor Processing Units (TPUs) for more than a decade now. But even Google is ramping up its use of the processors, and other tech giants are jumping on board, too.
Alphabet recently signed a multi-year deal for Broadcom to design TPUs that will run and train its Google Gemini frontier AI model. Alphabet is Broadcom's largest customer, and the tech giant's purchase of custom processors accounted for an estimated $13 billion of Broadcom's 2025 sales.
Meta Platforms is another important customer, with Broadcom designing the company's Meta Training and Inference Accelerator (MTIA) chips. AI has become increasingly important for Meta, as the company just launched its popular Muse agent app, which can autonomously shop online for users, book trips, and manage daily tasks. Muse shot to the top of Apple's U.S. App Store last month, indicating that consumers are eager for personal AI agents.
OpenAI jumped on the custom AI processor bandwagon this year by hiring Broadcom to design its Jalapeno processor, which it describes as an "LLM-optimized inference chip" that makes AI faster and more reliable. The first of the new processors will be deployed in data centers by the end of this year, with a ramp-up in 2027 and 2028.
And Anthropic has been a Broadcom customer for at least a couple of years, purchasing custom processors from the company valued at tens of billions of dollars. The two companies recently expanded their relationship with Broadcom, agreeing to lend Anthropic up to $42 billion to expand its artificial intelligence data center infrastructure investments.
NASDAQ: AVGO
Key Data Points
Broadcom's growth is impressive by any metric
It's one thing to see a company sign a slew of new deals, but it's even more impressive to see its sales and earnings accelerate as its customers keep coming back.
Broadcom's AI semiconductor revenue rose 221% to $16.7 billion in the most recent quarter, and managers are guiding for AI sales to double in fiscal 2027 to $115 billion. Even more shocking is that CEO Hock Tan has already said that AI revenue will double again in 2028, reaching $230 billion.
Tan also said that his company is on target to exceed $30 in earnings per share in fiscal 2018 -- a sixfold increase from fiscal 2025. With demand for custom processors already ramping up and the broader chip market expanding over the next few years, there's little doubt that Broadcom will continue benefiting from this trend.
The company will hold an estimated 60% of the custom AI processor market in 2027, and it's proven to be a reliable partner for the world's leading tech companies. If Broadcom continues on this path, and there's no reason to believe it won't, owning Broadcom shares is one of the best ways to tap into this accelerating AI supercycle.
