Taalas

AI Chips · Semiconductors · AI Inference

Acquisition

AMD to buy Taalas, startup that hardwires AI models into silicon

The move follows Nvidia's $20 billion Groq deal, signaling custom inference silicon is now a competitive battleground for chipmakers.

• AMD signed a definitive agreement to acquire Toronto-based Taalas, a startup that etches AI model weights directly into silicon chips; terms were not disclosed.
• Founded in 2023, Taalas builds what it calls model-specific integrated circuits (MSICs) that bake a single AI model's weights permanently into transistors instead of loading them from memory.
• Its first test chip, the HC1, was fabbed on TSMC's 6nm process and served Meta's Llama 3.1 8B model at roughly 16,960 tokens per second, about 48x faster than Nvidia GPUs in Taalas' own benchmarks.
• Taalas had raised $219 million in venture funding since its 2023 founding, according to CNBC.
• The deal lands about seven months after Nvidia's $20 billion asset purchase from Groq and just weeks after AMD's own partnership with Cerebras, both aimed at boosting inference speed for AI agents.
• As inference workloads for AI agents and code assistants grow, chipmakers are racing to pair flexible GPUs with fixed-function silicon that trades adaptability for raw speed and lower cost per token.