Bay Area AI Startup Taalas Raises $169M for Custom Chips
Taalas, a Bay Area AI startup, has closed $169 million in funding to develop chips that are optimized for specific artificial intelligence models. The investment highlights continued venture capital focus on specialized hardware designed to improve the efficiency and scalability of next-generation AI systems.
- The company was founded by Ljubisa Bajic, who previously founded Tenstorrent, a competitor in the AI chip space. Co-founders Drago Ignjatovic and Lejla Bajic were also early engineers at Tenstorrent. - Taalas's core technology involves "hardwiring" AI model components directly into the silicon, creating a "model-specific processor." This approach aims to dramatically reduce latency and computational costs by integrating high-speed SRAM memory and eliminating the need for external memory access during inference. - The startup claims its first chip, the HC1, can achieve over 16,000 tokens per second for the Llama3.1-8B model, a speed multiple times faster than competitors like Nvidia, Cerebras, and Groq. However, the trade-off for this performance is that the chip is purpose-built and can only run that specific AI model. - In partnership with Taiwan Semiconductor Manufacturing Co. (TSMC), Taalas has a "foundry-optimized workflow" that allows it to produce custom chips in approximately two months. This rapid turnaround is achieved by customizing the final two metal layers of a nearly complete chip. - This latest funding round brings the Toronto-based startup's total capital raised to over $200 million since its founding in 2023. Key investors include Quiet Capital and Fidelity. - The company's first product, a hard-wired Llama 3.1 8B model, is already available as a chatbot demo and an inference API service. Taalas plans to release two more models in 2026. - Competitors in the specialized AI inference hardware space include Cerebras and Groq. The funding announcement follows an intensified investor interest in the sector, highlighted by Nvidia's recent $20 billion agreement to acquire intellectual property from Groq. - The engineering team consists of 25 people with prior experience at major tech and chip companies including AMD, Apple, Google, and Nvidia. Recently, Paresh Kharya joined as VP of Products after serving as a director of AI infrastructure product management at Google Cloud.
