12 companies tracked in this category. Names and one-line descriptions are free to browse — pricing model, funding, founding year, headquarters, and cited sources are in the full dataset.
| Company | Description |
|---|---|
| Cerebras Systems | Designs the Wafer Scale Engine (WSE-3), the largest single AI chip made via wafer-scale integration, packaged into CS-3 supercomputer systems, plus a cloud-based AI training/inference API service. |
| d-Matrix | Designs the Corsair AI inference accelerator chip, using digital in-memory computing to accelerate large language model inference in data centers. |
| Etched | Designs Sohu, an ASIC hardcoded for the transformer architecture, to run transformer-based LLM inference at higher throughput and lower cost/power than general-purpose GPUs. |
| Graphcore | British semiconductor company that designs the Intelligence Processing Unit (IPU), a processor architecture built for machine learning training and inference. |
| Groq | Designs the LPU (Language Processing Unit), a deterministic ASIC accelerator purpose-built for AI inference, sold via GroqCloud, a developer cloud platform for token-based LLM inference. |
| Hailo | Designs AI inference accelerator chips and modules for edge devices, including the Hailo-8/Hailo-8L accelerators and the Hailo-10H generative-AI edge module. |
| MatX | Designs custom semiconductor chips (MatX One) purpose-built for large language model training, reinforcement learning, and inference, keeping model weights in SRAM and KV-cache in HBM. |
| Positron | Builds purpose-built AI inference hardware, currently the Atlas inference appliance/server (supports models up to 500B parameters), with next-gen chip Asimov and system Titan planned for 2027. |
| Rain AI | Develops neuromorphic, brain-inspired AI chips, notably the Memristive Nanowire Neural Network (MN3), using in-memory compute aimed at more energy-efficient AI processing. |
| SambaNova Systems | Designs the SN40L Reconfigurable Dataflow Unit (RDU) AI accelerator chip, sold both as on-premises DataScale systems and as pay-per-token access via SambaNova Cloud. |
| Tenstorrent | Designs AI accelerator chips (Grayskull, Wormhole, Blackhole) built around its Tensix core architecture, and separately designs and licenses RISC-V CPU IP; led by CEO Jim Keller. |
| Untether AI | Built at-memory-compute AI inference accelerator chips (runAI200, speedAI240) that place compute directly next to memory; the company is now defunct. |