Most valuable public AI chip companies

Nvidia is the company every other name on this list is measured against. Its data-center GPUs and the CUDA software moat still define AI compute, and its market value dwarfs the rest of the field combined. In this cycle it has moved from selling chips to shaping the whole market: licensing rival architectures, backing neoclouds, and reaching deeper into the software and networking layers around its silicon.

Broadcom is the quiet giant of AI silicon. It co-designs custom accelerators (XPUs) for the largest hyperscalers and supplies the switching and networking that stitches AI clusters together. That combination of bespoke compute plus the fabric around it has made AI one of its fastest-growing lines and pushed it into the trillion-dollar tier.

AMD is the most credible direct challenger to Nvidia in merchant AI GPUs. Its Instinct MI-series accelerators and open ROCm software stack give hyperscalers a second source, and each generation narrows the gap on both performance and the software ecosystem that has long been Nvidia's defense.

Qualcomm owns the on-device edge of AI. Its Snapdragon platforms put neural processing in phones, PCs, cars, and headsets, and it has been pushing that efficiency-first design philosophy toward data-center inference, where power per token increasingly decides the economics.

Arm sits one layer beneath almost everyone here. It does not sell chips; it licenses the CPU cores that increasingly sit alongside accelerators in AI servers, from custom hyperscaler silicon to the host processors in merchant systems. Its inclusion stretches the merchant-silicon definition to the IP layer, but its cores are now everywhere AI compute runs.

Intel is the turnaround story. Its Gaudi accelerators and Xeon CPUs give it a foot in AI compute, and partnerships such as the heterogeneous inference work with SambaNova aim to keep it in the conversation while its foundry ambitions play out. The market still prices it well below its historical weight.

Marvell is the other custom-silicon and interconnect specialist, competing with Broadcom to design bespoke accelerators for hyperscalers and to supply the optical and electrical plumbing between them. Its fortunes track the build-out of large AI data centers.

Cerebras is the newest public name here, having listed on the Nasdaq in May 2026. Its wafer-scale engine takes the opposite bet to everyone else, putting an entire cluster's worth of compute on a single giant chip, and it has leaned into fast inference as a service. As a fresh IPO its valuation is the most volatile on the list.