Key Points:
- Taalas raised $169 million to build faster, cheaper AI chips.
- The technology “hard-wires” AI models directly onto silicon.
- Manufacturing takes only two months compared to the usual six.
- The startup aims to run advanced models like GPT 5.2 this year.
Toronto-based chip startup Taalas has secured $169 million in new funding to change how computers handle artificial intelligence. On Thursday, the company announced it has developed a specialized chip that runs AI applications faster and cheaper than current standard methods. This news comes just weeks after industry giant Nvidia sparked a frenzy by paying $20 billion to license similar technology from Groq.
Taalas takes a unique approach to building hardware. Instead of making a general-purpose chip that runs software, they essentially “print” parts of the AI model directly onto the silicon. This creates a custom chip perfectly suited for specific AI models, such as smaller versions of Meta’s Llama. The design pairs this customized silicon with large amounts of SRAM, a type of memory that is expensive but incredibly fast.
CEO Ljubisa Bajic explains that this “hard wiring” is the secret to their speed. By physically baking the AI’s logic into the metal, the chip doesn’t have to work as hard as a traditional processor to process data.
The manufacturing process is also surprisingly quick. Bajic says the company assembles a nearly complete chip with about 100 layers. They only need to customize the final two metal layers to fit a specific AI model. Using manufacturing partner TSMC, Taalas can finish a custom chip in just two months. In comparison, making a standard high-end AI processor like Nvidia’s Blackwell often takes roughly six months.
Currently, Taalas can produce chips for less complex AI models. However, the company has ambitious plans. By the end of this year, they aim to build a processor powerful enough to deploy cutting-edge models, such as GPT 5.2.
The race for specialized AI hardware is heating up. Taalas joins a competitive field that includes Groq, Cerebras, and D-Matrix. These startups are all betting on new chip designs to solve the massive computing needs of modern artificial intelligence.
