AMD has acquired Taalas, a Canadian startup that embeds AI models directly into silicon. The acquisition follows Taalas' public debut in February 2026, when the company revealed its approach of embedding a model's architecture and trained parameters into the chip. This method enables extremely fast inference but restricts each chip to a single model. A demo chip from Taalas achieved over 16,000 tokens per second for Llama 3.1-8B, significantly outperforming competing hardware. The company's approach, which locks each chip to a specific model, has drawn attention from industry players like Google, which is reportedly developing a similar chip for Gemini.
AMD plans to integrate Taalas' technology into its accelerator roadmap, offering it alongside Instinct GPUs as a system-level solution. Vamsi Boppana, SVP of AMD's AI division, stated the deal strengthens the company's AI portfolio. Taalas co-founder Ljubisa Bajic noted that AMD provides the scale and reach the startup needs. The acquisition is subject to standard regulatory approvals.
Taalas was founded in Toronto in 2023 and remained in stealth mode until its February 2026 debut. The company's technology involves embedding a model's architecture and trained parameters directly into the chip, which enables high inference speeds but limits flexibility. The startup's demo chip achieved over 16,000 tokens per second for Llama 3.1-8B, many times faster than competing hardware. The company's approach has drawn attention from industry players like Google, which is reportedly working on a similar chip for Gemini.
Source: thedecoder
