AMD buys Taalas startup to develop new AI chips

AMD company / illustrative
Фото: AMD company / illustrative

American chipmaker Advanced Micro Devices has agreed to purchase Canadian startup Taalas, which develops specialized chips for artificial intelligence.

The acquisition will allow AMD to obtain technology that turns individual artificial intelligence models into specialized chips. Such processors are primarily intended for inference — running already trained neural networks and processing user requests.

Unlike general-purpose graphics processors capable of performing various tasks, Taalas proposes creating a separate hardware solution for a specific AI model. Its parameters are essentially embedded into the chip, which should reduce data processing time and energy consumption.

The startup calls its approach 'model is the computer' and claims that specialized solutions can be significantly more efficient than software-based deployment of neural networks on general-purpose hardware. There is no independent confirmation of the stated performance figures yet.

The first Taalas processor, HC1, was developed to work with the Llama 3.1 8B language model. The company positioned it as a demonstration of ultra-fast inference technology with lower energy consumption compared to large GPU systems.

Taalas was founded in Toronto by specialists who previously worked at AMD, Nvidia, and AI chip developer Tenstorrent. In February 2026, the startup raised $169 million, bringing its total funding to over $200 million.

For AMD, the purchase of Taalas will be a way to expand its portfolio of AI technologies beyond traditional Instinct graphics accelerators. The company is trying to strengthen its position in the data center equipment market, where Nvidia remains its main competitor.

The acquisition cost, deal closing timeline, and the future fate of the Taalas brand are not disclosed. At the time of the announcement, the companies also did not publish a separate official release with detailed terms of the agreement.

Based on sources: Bloomberg, Taalas, AMD

analytics