Divisions / Layer 02
Semiconductor
Silicon designed for what comes next.
CPUs, GPUs, AI accelerators, networking chips, and memory technology, designed as one architecture and licensed as semiconductor IP.
What we do
Capabilities
CPUs
Energy-efficient server and client cores built to feed accelerators, not bottleneck them.
GPUs
Programmable parallel processors for graphics, simulation, and general-purpose compute.
AI accelerators
Tensor engines tuned for low-precision training and high-throughput inference.
Networking chips
Switch silicon, SmartNICs, and optical interconnect for clusters of 100,000+ accelerators.
Memory technology
High-bandwidth memory interfaces, near-memory compute, and CXL memory pooling.
Semiconductor IP
Cores, interconnect, and chip architecture licensed to partners building their own silicon.
Portfolio
Featured products
Vega GPU
Our flagship data-center GPU for training and inference.
Orion CPU
A high-core-count server CPU paired to our accelerators over a coherent link.
Photon Switch
A 102.4 Tb/s switch chip with co-packaged optics.