Liquid AI raises $250M led by AMD to build a new type of generative AI model

Liquid AI Inc., an artificial intelligence startup and MIT spinoff developing generative AI models based on a fundamentally different architecture than transformer-based AI, today announced it has raised $250 million in an early-stage funding round led by Advanced Micro Devices Inc.

OSS Capital, Duke Capital Partners and PagsGroup also participated in the Series A round. The company previously raised $46.6 million in seed funding in 2023.

The Boston-based startup’s new models are called “Liquid Foundation Models,” or LFMs, which the company says can deliver performance comparable to, or superior to, traditional large language models on the market today.

The company’s models are based on a concept of “liquid neural networks,” a classification of AI networking that differs from Generative Pre-Trained Transformer-based models, or GPTs, that underpin today’s popular chatbots such as OpenAI’s ChatGPT and Google LLC’s Gemini.

“At Liquid, our mission is to build the most capable and efficient AI systems at every scale,” said Chief Executive Ramin Hasani. “Since our inception, we have demonstrated the scalability of our technology by releasing text-based models, unveiling multimodal LFMs, and collaborating with key partners to showcase real-world impact.”

Liquid AI offers three model sizes: LFM-1B, a dense 1.3 billion-parameter model for on-device, such as mobile phones; LFM-3B, a 3.1 billion=parameter model optimized for edge deployments; and LFM-40B Mixture of Experts, designed to tackle complex tasks.

The company said LFMs have a reduced memory footprint compared with GPT architectures that is even more substantial for lengthy inputs – for example, when processing extremely long documents or video. Liquid said the LFM-3.1B model used significantly less memory than other small models when deployed than models such as Google LLC’s Gemma 2 2B-it or Meta Platform Inc.’s Llama 3.2 3B.

Along with the funding, the company has joined in a strategic partnership with AMD, strengthening Liquid’s ability to optimize its foundation models with the chipmaker’s graphics processing unitscentral processing units and neural processing units.

Liquid AI added that it will use the additional funding to scale infrastructure to support the development of future LFMs for diverse model sizes and data types. It will also develop of tailored LFM research to optimize the models for specific industries such as consumer electronics, biotechnology, telecommunications, financial services and e-commerce.

Image: SiliconANGLE/Microsoft Designer