Saturday, August 8, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

AMD acquired Taalas, a startup that hard-codes AI model weights directly into inference chips for extreme speed.

AMD enters custom inference silicon competition with a lock-in play; validates the market viability of specialized AI chips outside training accelerators.
Trade pressSlicast · August 8, 2026 · Global · Source: The Decoder
importance 78

AMD is acquiring Taalas, a Canadian AI startup that designs specialized inference chips. Founded in Toronto in 2023, Taalas emerged from stealth in February with a distinctive approach: embedding a model's architecture and trained parameters directly into silicon. This design enables exceptionally fast inference but commits each chip to a single model. The company's demonstration chip achieved over 16,000 tokens per second per user while running Llama 3.1-8B—substantially faster than competing hardware. Google is reportedly pursuing a similar architecture for Gemini.

AMD intends to integrate the technology into its accelerator product roadmap, positioning it as a complementary system-level solution alongside its Instinct GPUs. Vamsi Boppana, SVP of AMD's AI division, stated the acquisition strengthens the company's AI portfolio. Taalas co-founder Ljubisa Bajic noted that AMD provides the scale and market reach essential to the startup's growth. The acquisition remains subject to standard regulatory approvals.

Read the original
AMD acquired Taalas, a startup that hard-codes… · Slicast