Taalas raises $50M to develop chips optimized for specific AI models

Taalas Inc., a startup that plans to sell chips tailor-made for specific artificial intelligence models, has raised $50 million to support its commercialization effort.

The corporate announced the raise on Tuesday. The funding was provided by Quiet Capital and Pierre Lamond, a outstanding enterprise capitalist who backed several of the semiconductor industry’s early players and whose chip bona fides goes back to Fairchild Semiconductor and National Semiconductor within the Nineteen Sixties.

Taalas is led by Chief Executive Officer Ljubisa Bajic, who previously launched AI chip startup Tenstorrent Inc. in 2016. The latter company received a $1 billion valuation after its most up-to-date $200 million funding round. Bajic founded Taalas last August with Drago Ignjatovic and Lejla Bajic, who held engineering leadership roles at Tenstorrent.

Practically all AI chips include optimizations designed to hurry up matrix multiplications, the mathematical operations that neural networks use to process data. Some processors feature additional optimizations geared towards specific AI use cases. Nvidia Corp.’s latest H200 graphics card, for instance, ships with a considerable amount of high-speed memory to speed up language models.

Taalas plans to take the concept a step further. Based on The Information, the corporate is working on processors that won’t simply be optimized for AI, but will likely be built with the necessities of a particular neural network in mind. The corporate hopes that this approach will make its chips significantly faster than today’s graphics cards.

“Commoditizing AI requires a 1000x improvement in computational power and efficiency, a goal that’s unattainable via the present incremental approaches,” Bajic said. “The trail forward is to comprehend that we should always not be simulating intelligence on general purpose computers, but casting intelligence directly into silicon.”

Developing a custom processor can take years and a whole bunch of thousands and thousands of dollars in some cases. In consequence, Taalas’ plan to create multiple chips that can each be optimized for a unique AI algorithm is more likely to involve significant technical challenges. To handle those challenges, the corporate is developing an automatic engineering workflow that it says will speed up its semiconductor design efforts.

Taalas says certainly one of its processors will contain enough memory to carry an “entire large AI model.” Running a model entirely on-chip removes the necessity for external RAM, which suggests data doesn’t need to travel between the RAM module and the chip. Reducing data travel, in turn, hurries up processing.

Taalas plans to deliver the tape-out of its first product, a processor optimized for giant language models, within the third quarter. A tape-out is the version of a chip design that is shipped to the fab for manufacturing. Taalas will start shipping the processor to customers in early 2025. 

Photo: Unsplash