Skip to content
AI IntelligenceAug 8, 2026M&A
Article

AMD -> Taalas

AMD is acquiring Canadian startup Taalas, which hard-codes model weights directly into inference chips, making them extremely fast but locked to a single model. A demo chip hit over 16,000 tokens per second per user running Llama 3.1-8B.

Data Cube AI EditorialSource: the-decoder.com
01

Source Brief

AMD is acquiring Canadian startup Taalas, which hard-codes model weights directly into inference chips, making them extremely fast but locked to a single model. A demo chip hit over 16,000 tokens per second per user running Llama 3.1-8B.

02