AI Hardware & Accelerator Chips
15 companies
0 of 0 categories open
Relationship:
Loading the map…
On this page
Silicon designed for training or running models: GPUs, purpose-built inference accelerators, wafer-scale processors and edge chips. Established semiconductor companies share the category with startups building alternative architectures, plus the in-house designs of the large cloud providers.
What to compare
Peak throughput rarely decides it. Software maturity does, along with memory capacity and bandwidth, whether the part can be bought outright or only rented through one cloud, and how much work porting an existing stack would take. A faster chip that nobody's framework targets runs slower in practice.
15 companies in this category
AMDChipmaker competing in AI accelerators with its Instinct GPU line.
AWS (Trainium/Inferentia)Amazon's custom silicon for AI training and inference.
BlaizeEdge AI chipmaker for computer-vision and inference workloads.
EtchedStartup building an ASIC chip specialized solely for transformer inference.
Google (TPU)Google's custom Tensor Processing Units for AI workloads.
HailoIsraeli maker of edge AI inference chips.
Intel (Gaudi)Intel's Gaudi AI accelerator chip line.
MatXStartup designing chips purpose-built for large-model training.
MythicAI chipmaker using analog compute for efficient edge inference.
NVIDIADominant maker of GPUs powering most AI training and inference.
Positron AIStartup building custom inference accelerators.
Rain AIStartup developing brain-inspired, energy-efficient AI chips.
TenstorrentAI chip company led by Jim Keller, building RISC-V AI processors.- Untether AICanadian AI chipmaker focused on energy-efficient inference.
d-MatrixAI chip startup focused on efficient inference compute.