Broadcom is heading into its June 3 earnings report with investors focused on a simple question: how much more fuel can its custom-chip business get from the AI buildout. The company is the leading designer of ASICs, and Goldman Sachs expects demand for those chips to outpace GPU demand in the coming years.
That is why avgo stock is drawing attention now. Broadcom has said its AI chips alone could generate $100 billion in fiscal 2027 sales, a target that would put it among the central suppliers of the AI infrastructure race. Its reach extends through custom accelerator work with Alphabet and Meta Platforms, while Amazon has its own semiconductor arm, Annapurna Labs, building chips of a similar type for cloud use.
The appeal is easy to see. ASICs, or application-specific integrated circuits, are highly cost-effective when deployed at scale because they are built for one narrow job. Alphabet's Tensor Processing Units, Meta's Meta Training and Inference Accelerator, and Amazon's Trainium chips show how the biggest cloud companies are already using custom silicon to lower costs and improve efficiency. Google uses TPUs to run its Gemini large language model and other AI-powered services such as Search, Maps and Photos.
Broadcom's own role is bigger than a single customer order. Meta and Google work with its custom accelerator platform to design chips, tying the company to the spending plans of two of the largest hyperscalers in the market. That matters because the first wave of AI data-center investment centered on training hardware, but the next wave is increasingly about inference, the everyday work of running models for users.
There is a catch, and it is a big one. Rising ASIC demand does not mean GPUs are going away. ASICs are rigid because they are hardwired for specific tasks, while Nvidia's GPUs can be reprogrammed more easily as workloads change. That flexibility still gives GPUs a strong place in data centers, even if custom chips win more business where scale and efficiency matter most.
The backdrop helps explain the shift. Data centers are getting larger, energy use is becoming a bottleneck and cost pressure is rising as AI deployments spread. Broadcom's June 3 report will give the market a fresh read on whether its AI-chip growth is still accelerating toward that fiscal 2027 target, but it will not settle the larger fight over whether custom silicon or general-purpose GPUs will define the next stage of AI infrastructure.
