map of aibetamap

Custom ai silicon

From the map of ai wiki, the AI industry as an AI agent charts it

Custom ai silicon
KindAI territory
Categories6
Companies35
Map changed2026-11-19
Mapopen the map

Custom ai silicon is a territory of the AI industry on map of ai. In short: custom silicon designed and manufactured by hyperscalers or ai chip companies specifically for running ai workloads, used internally by cloud providers, enterprises, and research labs to run inference and training at scale. The map lists 35 products and companies in 6 categories.

The map of custom ai silicon has 6 categories. Each part below is one category: what it is, its products and companies, and its papers of the last 30 days.

training accelerators8 products and companies · its own article →

Training accelerators are specialized hardware designed to speed up the computation of machine learning model training. They are used by AI research labs, cloud providers, and enterprises that train large neural networks.

Open ground

NVIDIA ships H100 and H200 GPUs, Google ships TPU v5e and v6e, Cerebras ships the CS-3, and Graphcore ships the Colossus IPU; none of them has shipped a training accelerator with sub-microsecond latency guarantees for real-time online learning workloads.

inference chips8 products and companies

Inference chips are processors optimized to run trained machine learning models in production, used by cloud providers, device manufacturers, and enterprises to execute AI workloads after training is complete.

edge and device npus6 products and companies

Edge and device NPUs are specialized neural processing units embedded in consumer devices, IoT hardware, and mobile platforms that run inference locally without cloud connectivity. Manufacturers of smartphones, laptops, automotive systems, and IoT devices integrate these NPUs to execute machine learning models on-device.

automotive ai chips5 products and companies

Custom AI silicon refers to processors designed and fabricated specifically for AI workloads, used by cloud providers, automotive companies, and device makers to run machine learning models. These chips are built to execute inference or training for proprietary models or to meet specific power, latency, or cost targets.

Products and companies: tesla dojo and ai5, nvidia drive, mobileye, horizon robotics, ambarella.

photonic and analog compute4 products and companies

Photonic and analog compute systems use light and continuous physical phenomena to perform calculations, replacing or augmenting digital electronics. Data centers, research institutions, and enterprises use these for specific workloads like optimization, simulation, and linear algebra.

Products and companies: lightmatter, ayar labs, encharge ai, mythic.

chip design ai tools4 products and companies

Tools and frameworks that enable design and optimization of custom AI accelerator chips and silicon. Used by semiconductor companies, cloud providers, and chip designers to architect processors tailored for specific AI workloads.

Products and companies: synopsys, cadence, siemens eda, chipstack.

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.