Over the last year, AI has undergone a large scale productization while deep learning models have become the world's most demanding computational workload. For many companies, however, AI is proving to be unsustainably capital intensive, power hungry, and GPU constrained.
"Artificial intelligence is like electrical power – an essential good that will need to be made available to all. Commoditising AI requires a 1000x improvement in computational power and efficiency, a goal that is unattainable via the current incremental approaches,” said Ljubisa Bajic, Taalas' CEO.
“The path forward is to realise that we should not be simulating intelligence on general purpose computers, but casting intelligence directly into silicon. Implementing deep learning models in silicon is the straightest path to sustainable AI."
Taalas is developing an automated flow for rapidly implementing all types of deep learning models (Transformers, SSMs, Diffusers, MoEs, etc.) in silicon. Proprietary innovations enable one of its chips to hold an entire large AI model without requiring external memory.
According to Bajic, the efficiency of hard-wired computation enables a single chip to outperform a small GPU data centre, opening the way to a 1000x improvement in the cost of AI.
"We believe the Taalas 'direct-to-silicon' foundry unlocks three fundamental breakthroughs: dramatically resetting the cost structure of AI today, viably enabling the next 10-100x growth in model size, and efficiently running powerful models locally on any consumer device. This is perhaps the most important mission in computing today for the future scalability of AI. And we are proud to support this remarkable n-of-1 team as they do it," said Matt Humphrey, Partner at Quiet Capital, a leading investor in Taalas.
The company is currently taping out its first large language model chip – due to appear in the third quarter of 2024 - and is planning to make it available to early customers in the first quarter of 2025.
Taalas was founded by Ljubisa Bajic, Drago Ignjatovic, and Lejla Bajic. Prior to co-founding Taalas, Ljubisa founded Tenstorrent in 2016. Drago and Lejla joined Tenstorrent soon after as early engineering leaders. The team has spent decades collectively working together on a long list of AI processors, GPUs, and CPUs across Tenstorrent, AMD, and Nvidia.
