AMD Challenges Nvidia with Acquisition of AI Chip Startup Taalas
In a daring attempt to disrupt Nvidia’s supremacy in the AI hardware sector, Advanced Micro Devices (AMD) made a public announcement on August 6, 2026. The company has decided to acquire Taalas, a Toronto-based startup that innovatively etches AI model weights directly into silicon, bypassing the need for conventional high-bandwidth memory. The financial details of the deal remain undisclosed, but it’s known that Taalas has raised a substantial $219 million in venture capital since its inception in 2023.
Taalas is renowned for its creation of model-specific integrated circuits — chips that are exclusively designed to run a single AI model at an impressive speed. The company’s debut chip, the HC1, was fabricated on TSMC’s 6-nanometer process and was used to serve Meta’s Llama 3.1 8B model at a staggering 17,000 tokens per second. This is approximately 73 times faster than Nvidia’s H200, and it uses only one-tenth of the power, as per Taalas. AMD has plans to incorporate this technology into its Helios rack-scale systems, alongside its Instinct GPUs and EPYC CPUs. The deal is expected to be finalized in Q4 2026.
This acquisition marks AMD’s fourth inference-focused deal since November. It follows just seven months after Nvidia’s hefty $20 billion acquisition of assets from Groq. AMD CEO Lisa Su has emphasized that “there’s no one-size-fits-all when it comes to chips,” and the company projects that the AI inference market could witness an annual growth of more than 80%. Following the announcement, AMD shares experienced a rise of approximately 1.5%.
Source: CNBC – AMD acquires chip startup that hardwires AI models into its silicon