Untether AI

Untether Artificial Intelligence (AI) is a semiconductor company that develops specialized AI inference accelerators and supporting hardware platforms for data center and edge deployment.

• AI inference accelerator chips and cards for Neural Network (NN) workloads (AI infrastructure)
• Memory-centric architecture for high-throughput, low-latency inference at low power (AI infrastructure)
• Hardware platforms for deployment in servers, data centers, and embedded systems (AI infrastructure)
• Software tools and SDKs for compiling, optimizing, and deploying AI models onto Untether AI hardware (AI developer tools)
• Solutions for computer vision, recommendation, and other deep learning inference use cases (AI applications)

More About Untether AI

Untether AI focuses on hardware and software for inference acceleration, targeting enterprise and embedded environments that run deep learning workloads at scale. Its products are designed for deployment in data centers, cloud infrastructure, and edge systems where inference throughput, latency, and power efficiency are primary constraints. The company’s accelerators are positioned as alternatives or complements to general-purpose CPUs and GPUs in inference pipelines.

The core technology approach centers on a memory-centric architecture that places compute elements adjacent to on-chip memory, often described as near-memory or at-memory compute. By reducing data movement between memory and compute, Untether AI hardware aims to increase operations per second per watt for NN inference. This architecture targets workloads such as convolutional neural networks, transformers, and other feed-forward models used in vision, recommendation, and language inference.

In enterprise environments, Untether AI products fit into AI infrastructure categories such as PCI Express (PCIe) accelerator cards, server-integrated AI modules, and embedded inference boards. These are typically installed in standard x86 or ARM servers and accessed through PCIe interfaces, integrating with existing data center networks and storage systems. The company also addresses Original Equipment Manufacturer (OEM) and system integrator use cases, where its chips can be integrated into custom boards or appliances for domain-specific AI functions.

On the software side, Untether AI provides tools and SDKs that map trained NN models onto its hardware. These tools generally support common AI frameworks through export or conversion workflows, enabling deployment of models trained in environments such as PyTorch or TensorFlow. The software stack handles graph compilation, quantization, scheduling, and runtime management to utilize the available parallelism and memory hierarchy of the accelerators.

From a marketplace taxonomy perspective, Untether AI aligns with AI infrastructure (inference accelerators, data center AI hardware), embedded AI (edge inference modules), and AI developer tools (compilers and SDKs). Enterprise technical teams typically evaluate Untether AI alongside other AI accelerator vendors when optimizing cost, power, and latency for inference at scale, especially in scenarios where workload characteristics match the strengths of memory-centric compute architectures.