The startup’s investors include Quiet Capital, Fidelity and Pierre Lamond, a former partner at Sequoia Capital, Khosla Ventures and Eclipse Ventures. In a blog post, Taalas CEO Ljubisa Bajic said the firm has 24 employees and has spent US$30 million of the over US$200 million it has raised so far to develop its first product. (Reuters, The Logic)

Talking point: Taalas is one of several startups trying to sell specialized chips to power the AI boom, but it’s taking an unconventional approach. Firms like Nvidia, the clear market leader, offer hardware that can be used for both training and running almost any AI model. Taalas is instead tailoring its chips to specific systems, carving the way they work into the design of the processor itself, which it claims makes them run faster and cheaper. Its first product is designed for a specific version of Meta’s Llama model, which was launched in July 2024 but which underpins a lot of AI tools because it’s open source. Taalas claims it can launch new chips in just two months, much faster than the industry standard. Bajic previously founded another Toronto AI chip startup, Tenstorrent, which later re-incorporated in the U.S.