AMD acquires Taalas to enhance AI inference performance with silicon-etched models

Aug 6, 2026

AI Summary

AMD has acquired AI chip company Taalas, which specializes in etching model weights directly into silicon, potentially improving inference performance significantly. This acquisition aims to challenge Nvidia's market dominance in AI hardware and is expected to close in the fourth quarter, pending regulatory approval.

AMD has announced the acquisition of Taalas, an AI chip company based in Toronto, to improve AI inference performance.
Taalas' technology involves etching model weights directly into silicon, creating model-specific integrated circuits (MSICs) that promise faster processing speeds.
The company's first test chip, HC1, demonstrated the ability to process up to 16,960 tokens per second, significantly outperforming existing Nvidia GPUs and Cerebras accelerators.
Taalas plans to release a second-generation chip, HC2, with an increased parameter count of 20 billion, which will be compatible with AMD's existing compute platforms.
AMD intends to integrate Taalas' technology with its Instinct-based Helios racks, allowing for a hybrid architecture that optimizes performance for different AI workloads.
While the technology offers high speed, it requires careful model selection since changes to the model necessitate a re-spin of the chips, although this process is expected to be less costly than starting from scratch.
The acquisition is part of AMD's strategy to build a comprehensive AI platform and is anticipated to be beneficial for major AI model developers and infrastructure providers.

amdinferencesiliconacquisitionperformance

AMD acquires Taalas to enhance AI inference performance with silicon-etched models

Related Stories

OpenAI Releases ChatGPT Desktop Version for Linux Users

Automattic's Mesh CRM Launches on Android with Enhanced Features

Sandbar's Stream ring aims to revolutionize AI wearables with voice technology

DeepSeek Releases V4 Pro 0813 Model for Large-Scale Applications