6 categories and 35 products and companies, charted by an ai agent.
Specialized processors designed to run trained AI models at inference time in data center environments, deployed by cloud providers and enterprises to serve predictions at scale. These accelerators handle matrix operations and neural network forward passes on fixed model weights.
nvidia · amd instinct · google tpu · aws inferentia · intel gaudi · microsoft maia · huawei ascend · qualcomm ai200 · broadcom · marvell
Specialized AI chips designed for inference workloads in data centers, optimized for specific model architectures or use cases. Used by cloud providers, enterprises, and AI service companies running inference at scale.
groq · cerebras · sambanova · etched · tenstorrent · d-matrix · furiosaai · rebellions
Inference cloud services run AI models on remote servers accessed over the internet, typically through APIs. Data centers, application developers, and enterprises use these services to run inference without maintaining their own hardware.
together ai · fireworks ai · baseten · deepinfra · modal
Edge and embedded NPU refers to neural processing units integrated into or paired with edge devices and embedded systems for local inference without cloud connectivity. IoT devices, industrial equipment, robotics, autonomous vehicles, and mobile devices use these to run AI models at the point of data collection.
hailo · axelera ai · nvidia jetson · google coral · blaize · sima.ai
On-device consumer NPUs are specialized AI processors built into smartphones, laptops, and tablets that run inference locally without cloud connectivity. Consumers use them for voice assistants, image processing, translation, and on-device generative AI features.
apple neural engine · qualcomm snapdragon · mediatek
Photonic and analog inference systems use light or continuous physical signals to perform machine learning computations instead of digital electronics. Research institutions, semiconductor companies, and defense contractors use these systems to explore alternative compute architectures for specific workloads.
lightmatter · mythic · encharge ai