AMD acquires Canadian AI chip startup Taalas

AMD said it will acquire chip startup Taalas for an undisclosed price to strengthen its AI inference technology. Taalas, based in Toronto, makes specialized silicon aimed at reducing computing and memory bottlenecks for AI inference. AMD plans to integrate the tech into its accelerator roadmap and build system solutions with AMD Instinct GPUs. Taalas raised $169 million in February, totaling about $219 million.

AMD reported Q2 results that beat on revenue and earnings and raised guidance, while projecting capital expenditures to more than double to $808 million from $389 million to expand capacity for EPYC processors and Helios rack-scale AI systems. The higher spending may pressure margins and free cash flow near term, contributing to a negative stock reaction, but management expects strong data center growth.

Intel, AMD and Micron shares hit 52-week highs as investors bought into AI-infrastructure related chip demand. Intel gained after reports Apple may use Intel foundries for future iPhone and Mac chips. AMD rose on Tuesday’s Q1 results, with $10.3B revenue (+38% YoY) and $5.8B data center revenue (+57% YoY), plus raised price targets. Micron also advanced amid tighter AI memory supply.

AMD said it plans to acquire AI chip startup Taalas to strengthen its position in artificial intelligence hardware and compete more directly with NVIDIA. The deal would give AMD AI-optimized chip technology, engineering talent, and IP, supporting its data center and AI accelerator roadmap as demand for AI computing rises.

AMD agreed to acquire Taalas, a Toronto startup, to hard-wire AI model weights into silicon to reduce DRAM memory reads that limit GPU LLM inference speed. Deal terms were not disclosed and is expected to close in Q4 2026, subject to approvals. AMD shares rose about 1.5% on the announcement. Taalas’ HC1 targets Llama 3.1 8B throughput and power claims; financial terms and independent verification were not provided.

AMD said Aug. 6 it agreed to acquire Taalas, a Toronto chip startup, to etch AI model weights directly into silicon to speed and reduce the cost of inference. Deal terms were not disclosed. Taalas claims its HC1 can run Llama 3.1 8B at 16,960 tokens per second, and AMD expects closing in Q4 2026.