AMD to acquire Taalas to enhance AI inference capabilities

AMD has entered into a definitive agreement to acquire Taalas, a Toronto-based startup specializing in AI inference silicon. The acquisition aims to strengthen AMD’s AI roadmap by integrating specialized hardware designed to run specific AI models more efficiently.

Taalas utilizes a structured ASIC approach that optimizes dataflow by hardwiring model weights and data movement directly onto the silicon. This method significantly reduces the compute and memory bottlenecks common in general-purpose architectures. While this approach offers high-speed inference—such as generating over 16,000 tokens per second for Meta’s Llama 3.1-8B model—it trades off programmability, as the hardware is largely optimized for a single model.

AMD plans to integrate Taalas’ technology into its existing accelerator portfolio, including its Instinct GPUs, to provide differentiated performance for high-volume, real-time AI applications. The Taalas team, led by co-founder and CEO Ljubisa Bajic, will join AMD’s AI organization.