Amazon, Alphabet, Microsoft and Meta Platforms are on track to spend about $745 billion on capital expenditures this year. Much of that investment is directed toward data centers, although the total also includes spending unrelated to artificial intelligence. Nvidia, Broadcom and Arista Networks supply different parts of the infrastructure needed to turn those facilities into operating AI capacity.
Nvidia provides graphics processors and computing systems for AI training and inference.
Revenue from enterprise and other industrial opportunities rose 138% year over year to $40 billion in the latest quarter. Custom-designed chips could take a larger share of data-center demand, but Nvidia is responding by widening its customer base; management expects 70% revenue growth in fiscal 2028.
Broadcom is supplying custom AI accelerators and high-speed networking equipment. Its designs include Google’s Tensor Processing Units, OpenAI’s Jalapeño and Meta’s MTIA.
Arista Networks supplies high-performance Ethernet switches and other equipment that links the chips inside data centers. Its EOS operating system runs its switches and routers, creating switching costs for customers. Arista said its AI networking systems now serve more than 100 companies across hyperscalers, specialized cloud providers, AI labs and enterprises.
The companies face different execution risks as spending continues. Nvidia confronts competition from custom accelerators, Broadcom’s outlook depends on continued AI infrastructure demand and could be affected by data-center construction rules, and Arista has warned that complex deployments may require extended testing and customer acceptance.