AMD has acquired Canadian startup Taalas, a company specializing in embedding AI model weights directly into inference chips, according to The Decoder. This approach aims to significantly enhance the speed and efficiency of AI model execution on hardware.
A demonstration of Taalas’ chip reportedly achieved processing speeds of over 16,000 tokens per second per user while running the Llama 3.1-8B model, showcasing the potential for high-throughput AI applications. The Decoder also noted that Google is pursuing a similar hardware strategy for its Gemini AI models.
This acquisition could position AMD strongly in the competitive AI chip market, an important development for Japanese investors and tech firms focusing on AI-driven innovation and hardware acceleration in the FX, crypto, and equities sectors.