AMD Acquires Taalas to Etch AI Models Directly Into Silicon
AMD announced on August 6, 2026 that it is acquiring Toronto-based Taalas, whose chips etch model weights directly into silicon, making them model-specific integrated circuits built to run inference far faster.
What happened
AMD announced on Thursday, August 6, 2026, that it is acquiring Taalas, a Toronto-based AI chip startup founded in 2023. The deal was announced at market close and financial terms were not disclosed. It is AMD's latest bid to challenge Nvidia's dominance in AI hardware.
Weights etched into silicon
Taalas takes an unusual approach to inference: instead of storing model weights in HBM memory, its chips etch them directly into the silicon, making them model-specific integrated circuits (MSICs). The startup's first test chip, HC1, was fabbed on TSMC's 6nm process and served Meta's Llama 3.1 8B at 16,960 tokens per second — roughly 48x faster than Nvidia GPUs and 8.5x faster than Cerebras accelerators at the time.
Why it matters
The second-generation HC2, due out this summer, targets 20 billion parameters per chip. At that scale, just 50 accelerators could support a trillion-parameter model — and AMD already has rack-scale Instinct Helios systems and an in-house design team to hold them. AMD is expected to pair GPUs for prompt processing with Taalas-based accelerators for token generation, much like Nvidia's December licensing deal with Groq.