Taalas Unveils 'Hardwired' AI Chips for Inference
Startup Taalas has launched a new type of "hardwired" AI chip designed specifically to accelerate inference tasks. The company claims its HC1 chip can process 17,000 tokens per second, offering a more energy-efficient alternative to programmable GPUs for scalable, low-latency AI applications at the edge.
- Taalas was co-founded by Ljubisa Bajic, Lejla Bajic, and Drago Ignjatovic, all of whom were early engineering leaders at the AI chip startup Tenstorrent. Ljubisa Bajic founded Tenstorrent in 2016 and served as its CEO before leaving in March 2023. The founding team has extensive experience working on AI processors, GPUs, and CPUs at major companies like AMD and NVIDIA. - The Toronto-based startup has raised over $200 million in total funding since it was founded in 2023, including a recent $169 million round. Key investors include Quiet Capital and Pierre Lamond, a well-known figure in the semiconductor industry. - For comparison, the Llama 3.1 8B model runs at approximately 180 tokens per second on other hardware. Groq, another company focused on high-speed inference, has benchmarked its systems running a Llama 3 8B model at 877 tokens per second. Taalas claims its specialized chip is nearly 10 times faster than current state-of-the-art solutions. - Enterprise AI procurement cycles for Fortune 500 companies can span 8-12 weeks for a single project and often involve rigorous security audits, proof-of-concept trials with real data, and validation of performance claims. A significant challenge for AI vendors is that buyers may not be aware that a solution to their problem is even possible, requiring an educational sales approach that focuses on aligning with business needs before discussing ROI. - In enterprise sales, Chief Revenue Officers (CROs) are leveraging AI for tasks like sales forecasting, lead scoring, and identifying inefficiencies in the sales process. AI-powered revenue intelligence tools can provide real-time coaching to sales reps and automate data capture, which has been shown to increase meetings booked and generate millions in additional sales. - For founders building agentic AI, multi-agent orchestration is a key architectural pattern where a complex task is broken down and assigned to multiple specialized AI agents. This approach can improve scalability and reliability but introduces challenges in coordinating the agents, managing data flow, and controlling costs. - The fundraising environment for AI startups in the Bay Area has seen a surge in early 2026, with investors prioritizing companies with clear enterprise applications and scalable infrastructure. While mega-rounds for established players continue, the bar for Series A funding has risen, with investors now expecting efficient growth (burn multiple under 2.0) and strong net revenue retention (above 120%). - Founders can utilize productivity frameworks like the Eisenhower Matrix to differentiate between urgent and important tasks, helping them focus on high-leverage activities. Strategic calendar planning, where time is blocked for specific "zones of genius" like fundraising or business development, is another effective technique for managing limited time.
