Taalas is an ambitious AI semiconductor startup pursuing one of the most radical approaches in the industry: turning an entire AI model directly into custom silicon. Rather than running a neural network on general-purpose GPUs or programmable accelerators, Taalas etches the model's structure and weights into dedicated transistors, creating what it calls 'Hardcore Models' — chips that are, in effect, a single AI model rendered in hardware. The promise is enormous gains in inference speed and energy efficiency, with the company claiming improvements on the order of 1,000x versus conventional GPU deployment.
The company was founded in 2023 by Ljubisa Bajic, the founder of AI chip company Tenstorrent, alongside co-founders Drago Ignjatovic and Lejla Bajic, both early Tenstorrent engineers. That pedigree lends Taalas unusual credibility in a field where silicon expertise is scarce and hard-won. The core insight behind Taalas is that as AI models stabilize into widely-deployed workhorses, the flexibility of general-purpose hardware becomes less valuable than the raw efficiency of dedicated silicon — and the cost of inference, which now dominates AI economics, can be slashed by hardwiring the model itself.
Taalas is building a platform to rapidly turn any AI model into custom silicon, automating the traditionally slow and expensive process of designing application-specific chips. If successful, this would let model providers and AI companies spin up bespoke inference chips for their most important models, achieving order-of-magnitude reductions in cost and power that programmable accelerators cannot match. Early demonstrations suggested a single Taalas chip could outperform a small GPU data center for the model it embodies.
The company has raised more than $200 million across its funding history. It emerged from stealth in March 2024 with $50 million across two rounds, backed by investors including Quiet Capital, Eclipse Ventures and veteran semiconductor investor Pierre Lamond. In February 2026 Taalas announced a further $169 million round to accelerate development and bring its hardwired-silicon approach to market.
Taalas represents a high-risk, high-reward bet on the future of AI inference: that the most efficient way to run a model is to become the model in silicon. With a world-class founding team and substantial capital, it is one of the most distinctive and closely watched challengers in AI hardware.