AMD's Taalas Acquisition: 48x Faster Inference, 20x Cheaper, Reshaping AI Chip Landscape
adrianscottcom · x · 2026-08-08
AMD has agreed to acquire Toronto-based AI startup Taalas, potentially disrupting the economics of generative AI. Taalas' HC1 chip etches models like Llama 3.1 directly into silicon, achieving 17,000 tokens per second (48x faster than legacy GPUs), 20x lower build costs (no HBM or liquid cooling), and 10x power efficiency. The 6nm chip has 53 billion transistors, dubbed a 'Hardcore Model.'
The acquisition signals a shift from general-purpose GPUs to specialized ASICs for inference. AMD projects 80% annual growth in AI inference and plans to integrate Taalas tech into its Instinct roadmap, pairing with EPYC CPUs to challenge Nvidia. Industry impacts: inference costs could drop by an order of magnitude, hardware becomes model-specific (e.g., 'Llama chips'), and Google is reportedly exploring similar etched chips. Founder Ljubisa Bajic, former AMD lead architect and Tenstorrent co-founder, returns as a 'homecoming' of elite talent.
More from Infra
- Google Raises 2026 Capital Spending to Up to $205 Billion — emmanuelvivier · 2026-10-16
- Google open-sources AX, a Kubernetes-style orchestrator for billions of AI agent tasks — thisdudelikesAI · 2026-10-16
- Valtz v0.1.0.11 adds FLUX.3 support, local image gen up to 4K on Mac — TgoAI · 2026-10-16
- Baidu open-sources FlashMask: up to 1.55x faster than FlashAttention-4 for long-context training — BaiduResearch · 2026-10-16
- GYRO quant of Qwen3.8-Flash-Next tested: near Q3 quality at much smaller size — PlasticRevenue4601 · 2026-10-16
- RL train-sample systems are frontier AI's hardest infra problem — and a trade secret — brianryhuang · 2026-10-16