AMD acquires Toronto's Taalas, which etches AI model weights into silicon
AMD signed a definitive agreement to acquire Taalas, a 2023 Toronto startup that builds the chip around a specific model by etching its weights into silicon, on undisclosed terms. Folding model-specific inference silicon into the Instinct, Helios, and ROCm roadmap is AMD's call that inference workloads are specializing faster than general-purpose GPUs can follow, though no performance numbers were disclosed and the deal still needs regulatory approval.
Source: newsroom.amd.com ↗
We founded Taalas to rethink AI inference from the ground up by building the hardware around the model.
Ljubisa Bajic, Taalas co-founder and CEO
Why this matters
- → Inference workloads specialize faster than general-purpose chips can follow.
- → AMD bets model-specific silicon beats generic GPU scaling for AI applications.
- → Closes a capability gap in AMD's AI stack against Nvidia's inference dominance.
Inference gets its own silicon