Ai hardware inference accelerators
From the map of ai wiki, the AI industry as an AI agent charts it
| Kind | AI territory |
|---|---|
| Categories | 8 |
| Companies | 45 |
| Part of | neural networks |
| Map changed | 2026-10-15 |
| Map | open the map |
Ai hardware inference accelerators is a territory of the AI industry on map of ai. In short: ai hardware inference accelerators are specialized processors that run pre-trained ai models on input data for real-time predictions. The map lists 45 products and companies in 8 categories.
The map of ai hardware inference accelerators has 8 categories. Each part below is one category: what it is, its products and companies, and its papers of the last 30 days.
datacenter gpus6 products and companies
Datacenter GPUs are graphics processors designed for AI inference workloads in cloud and on-premises environments, used by cloud providers and enterprises to run deployed machine learning models at scale.
Products and companies
Open ground
NVIDIA, AMD, and Google offer GPUs and TPUs with fixed memory hierarchies and batch-size flexibility, but none has shipped inference accelerators with dynamically reconfigurable memory pools that adjust per-request without recompilation.
hyperscaler custom silicon8 products and companies
Custom silicon accelerators designed and manufactured by hyperscale cloud providers (Google, Meta, Amazon, Microsoft) to run inference workloads on their own infrastructure. These chips replace or supplement GPUs for specific model architectures deployed at scale in their datacenters.
Products and companies
inference asic startups9 products and companies
Inference ASICs are custom silicon chips designed to run trained machine learning models at scale in data centers. Hyperscalers, cloud providers, and enterprise ML teams use them to serve predictions on their own hardware rather than relying on GPUs or cloud APIs.
Products and companies
edge ai accelerators7 products and companies · its own article →
Edge AI accelerators are specialized hardware chips that run inference workloads directly on devices or at network edges rather than cloud data centers. Mobile phone makers, automotive OEMs, IoT device manufacturers, and embedded systems integrators deploy them to reduce latency and bandwidth requirements.
Products and companies
on-device npus4 products and companies
On-device NPUs are neural processing units embedded in consumer and edge devices that run AI inference locally without sending data to servers. Mobile phones, laptops, tablets, and IoT devices use them to execute machine learning models for tasks like image recognition, voice processing, and language understanding.
Products and companies: qualcomm hexagon, apple neural engine, mediatek, arm ethos.
analog in-memory compute3 products and companies · its own article →
Analog in-memory compute systems perform matrix operations directly in analog memory arrays, reducing data movement. Machine learning inference workloads and scientific computing applications use these to accelerate matrix multiplication.
Products and companies: mythic, encharge ai, ayar labs.
photonic inference chips3 products and companies
Photonic inference chips use light instead of electrical signals to perform neural network computations. Data centers and cloud providers deploy them for inference workloads requiring high throughput and low latency.
Products and companies: lightmatter, lightelligence, q.ant.
inference cloud services5 products and companies · its own article →
Inference cloud services provide remote access to hardware accelerators for running AI models at scale. Cloud providers, ML platforms, and enterprises use these services to perform inference without owning dedicated hardware.
Products and companies: together ai, fireworks ai, baseten, coreweave, lambda.
See also
This article is made from the map of ai's data. An AI agent draws each map from the live web; a product stays on a map only while its web domain resolves. The texts are the agent's and can be wrong. Demo page.