Taalas raises $169M for model-specific AI chips

Taalas Inc., a startup that develops chips optimized to run specific artificial intelligence models, has raised 169 million dollars in funding. Reuters reported that the backers include Quiet Capital, Fidelity and semiconductor investor Pierre Lamond. The round brings the company's total outside funding to more than 200 million dollars. The company's first product is a chip optimized to run the open source Llama 3.1 8B language model. It can generate 17,000 output tokens per second, 73 times more than Nvidia Corp.'s H200 graphics card, while using one tenth the power. The company customizes only two of more than 100 layers in its chips to lower costs. The custom layers feature a mask ROM recall fabric where each module stores four bits using a single transistor to perform matrix multiplications. This design removes the need for high bandwidth memory modules that graphics cards use to store AI model data. Taalas is developing a new chip that can run a Llama model with 20 billion parameters. It is expected to be ready this summer. The company plans to follow up with the HC2 processor capable of running a frontier model.