Mark RadarMARK RADAR
About
EN
Sign in
Event File AI AI Chips AMD

AMD Strikes Deal to Buy Taalas, Boost AI Inference Push

1 reports · First detected 2026-08-24 · Last active 2026-08-24

The artificial-intelligence market is shifting from training models toward running them at scale, increasing demand for lower latency, better energy efficiency and reduced computing costs. Toronto-based Taalas, founded in 2023, develops model-specific silicon that embeds AI weights and dataflows into purpose-built hardware, easing the compute and memory bottlenecks of general-purpose architectures. The technology gives Advanced Micro Devices a specialized inference component as it competes with Nvidia across the expanding AI accelerator market.

AMD said on Aug. 6, 2026, that it signed a definitive agreement to acquire Taalas for an undisclosed amount, subject to customary closing conditions and regulatory approvals. Taalas has raised about $219 million from investors. AMD plans to incorporate its technology into the company’s accelerator roadmap and pair it with Instinct GPUs, Helios rack-scale systems, EPYC CPUs and ROCm software, creating heterogeneous, system-level products for real-time and high-volume AI inference workloads.

All Coverage

1 original reports

The Backstory

The history behind this event
AMD to Acquire Taalas in AI Inference Push2026-08-07 · 2 reports · similarity 0.96

AI competition is shifting from training increasingly large models to running them at scale, making inference speed, power use and cost critical battlegrounds. Nvidia’s GPUs dominate much of that market, while Toronto-based Taalas, founded in 2023, takes a more specialized approach: hard-wiring model weights into silicon to reduce the data movement and memory bottlenecks associated with general-purpose processors.

AMD said on Aug. 6, 2026, that it had reached a definitive agreement to acquire Taalas, with financial terms undisclosed and the deal subject to regulatory approval. Taalas raised $169 million on Feb. 19 and has secured more than $200 million in total funding. Its HC1 chip, built for Meta’s Llama 3.1 8B, delivers a claimed 17,000 tokens per second per user. AMD plans to add the technology to its accelerator roadmap and develop system-level products combining it with Instinct GPUs.

Mark Radar|MARK RADAR

If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →

All times are in Taipei time (GMT+8)