Toronto-based AI chip startup Taalas emerged from underground with $50 million in funding and the lofty goal of disrupting the GPU-centric world dominated by Nvidia.
Founded by Ljubisa Bajic, Lejla Bajic, and Drago Ignjatovic, all formerly of Tenstorrent (the creator of Grayskull), Taalas is developing an automated pipeline to quickly convert any AI model (transformers, SSM, diffusers, MoE, etc.) into custom silicon . . The company claims that the resulting Hardcore models are 1,000 times more efficient than their software counterparts.
The startup also says that one of its chips can contain an entire large AI model without requiring external memory, and the efficiency of wired computing allows a single chip to outperform a small data center GPU.
Launch intelligence directly into silicon
“Artificial intelligence is like electrical power: an essential commodity that will need to be available to everyone. The commoditization of AI requires a 1,000-fold improvement in computational power and efficiency, a goal that is unattainable through current incremental approaches. “The way forward is that we should not simulate intelligence on general-purpose computers, but rather embed intelligence directly into silicon. Implementing deep learning models in silicon is the most direct path to sustainable AI,” said Ljubisa Bajic, CEO of Taalas.
“We believe Taalas' direct-to-silicon foundry unlocks three fundamental advances: dramatically reset the current AI cost structure, viably enable the next 10- to 100-fold growth in model size, and efficiently execute powerful models locally on any consumer device. This is perhaps the most important mission in computing today for the future scalability of AI. And we are proud to support this extraordinary n-of-1 team as they do it,” said Matt Humphrey , a partner at Quiet Capital who led both rounds. financing together with Pierre Lamond, advisor to Eclipse Ventures.
Taalas says it will launch its first large language model chip in the third quarter of 2024 and aims to have its chips available to the first customers in the first quarter of 2025.
