AMD acquires AI inference chip startup Taalas to bolster inference roadmap
AMD announced a definitive agreement on August 6 to acquire Toronto-based AI inference chip startup Taalas, integrating Taalas's ASIC inference technology into its accelerator roadmap alongside AMD Instinct GPUs.
On August 6, AMD announced a definitive agreement to acquire Toronto-based AI inference chip company Taalas. Transaction value, payment structure, and closing timeline were not disclosed. Founded in 2023, Taalas specializes in ASIC inference accelerators hardwired to a single AI model. The company's core technology converts any AI model into a custom silicon chip in roughly two months and had raised $219 million in cumulative funding as of February.
Taalas's HC1 test chip, built on TSMC 6nm, runs Meta's Llama 3.1 8B at 16,960 tokens per second, roughly 48 times faster than a general-purpose Nvidia GPU. AMD AI Group SVP Vamsi Boppana said Taalas's technology and engineering team will strengthen the differentiated inference performance and efficiency of AMD's full-stack AI platform, with future system-level solutions combining the technology with AMD Instinct GPUs as part of the Helios rack-scale, EPYC CPU, and ROCm software ecosystem.
The deal continues AMD's inference push, following the November acquisition of AI software company MK1, the June acquisition of MEXT, and the July integration of FastFlowLM. Nvidia had earlier moved on inference with its March Groq-licensed CPU and AI systems. The transaction signals that, in the fourth year of generative AI, the industry battleground is shifting from raw GPU counts to system-level competition around model-specific inference silicon.