Taalas Raises $169M for AI Chips

- Taalas was co-founded by Ljubisa Bajic, the founder of another AI chip startup Tenstorrent, along with former Tenstorrent engineers Drago Ignjatovic and Lejla Bajic. The engineering team includes talent from AMD, Apple, Google, and Nvidia. - The recent $169 million funding round included investors like Quiet Capital, Fidelity, and semiconductor industry veteran Pierre Lamond. This brings the company's total funding to over $200 million since its founding in 2023. - The company's "direct-to-silicon" approach involves hardwiring a specific AI model's architecture and weights directly onto a chip. This eliminates the need for external memory and reduces energy consumption by about 90% compared to traditional chips that constantly move data. - This specialized design allows for a rapid manufacturing timeline of about two months in partnership with Taiwan Semiconductor Manufacturing Co. (TSMC), a significant reduction from the typical years-long process for custom chips. - Taalas's first chip, the HC1, is designed specifically for Meta's Llama 3.1 8B model and can achieve speeds of over 17,000 tokens per second. This is reportedly orders of magnitude faster than high-end GPUs like Nvidia's H200. - While the current chip is highly specialized, Taalas plans to release a second chip for a mid-sized reasoning AI model in the spring of 2026, with a more advanced "HC2" platform for frontier models planned for the winter. - The trade-off for this speed and efficiency is a lack of flexibility; a new chip must be created for each new AI model. However, the chips do allow for some fine-tuning using low-rank adapters (LoRAs). - By making AI inference significantly cheaper and faster, Taalas aims to enable powerful AI models to run locally on consumer devices rather than relying on large, power-intensive data centers.