LIQUID AI

Liquid AI is a spinout from MIT's Computer Science and Artificial Intelligence Laboratory, better known as CSAIL. The core science came from Ramin Hasani, who spent his PhD studying the nervous system of a tiny worm called C.

elegans. That worm has just 302 neurons and still manages to move, hunt, and react to its world.

Hasani and his collaborators asked a simple question. What if you built an AI that worked more like those neurons and less like a giant static network?

The answer was the liquid neural network, a model whose internal wiring keeps adjusting even after training is done. The team incorporated the company in 2023.

Daniela Rus, who runs CSAIL and is one of the most respected robotics researchers alive, came on as a co-founder. So did Mathias Lechner and Alexander Amini, both longtime research partners of Hasani.

Liquid AI builds foundation models, the big general-purpose brains that power chatbots and other AI tools. The difference is efficiency.

Their models, called Liquid Foundation Models, are designed to do more with far less memory and computing power. Companies pay to use them, either through a cloud API or by running the models on their own machines.

That second option matters. A bank or a hospital often cannot send private data to someone else's server.

Liquid AI lets them keep the model in-house. The pitch to a customer is basically this.

You get strong AI without renting a mountain of expensive chips.

The main products are the Liquid Foundation Models, a family that ships in several sizes so a customer can pick the smallest one that does the job. The company has leaned hard into on-device AI, releasing lighter models built to run directly on phones and embedded hardware rather than in the cloud.

The focus is always the same. Match the quality of a much larger model while using a fraction of the memory and power.

The growth angle is size, or the lack of it. Most frontier models are so heavy they only run in data centers.

Liquid AI aims for models light enough to run on a phone, a car, or a factory robot. That opens doors the giants cannot easily reach.

The AMD partnership pushes the same idea. AMD makes chips, and it wants efficient models that show off its hardware.

So AMD led the Series A and became both an investor and a distribution partner. Liquid AI also releases open models to researchers, which builds a following before the sales team ever makes a call.

The honest problem is Goliath. OpenAI, Google, Anthropic, and Meta have raised tens of billions of dollars between them.

Liquid AI has raised $250 million. The transformer architecture those giants use is battle-tested and improving fast.

Liquid AI has to prove its different approach not only works but keeps working as models get bigger. Betting against the most successful design in AI history is bold.

It is also the kind of bet that ends in either a great acquisition or a quiet shutdown, with not much room in between.

AMD Ventures led the $250 million Series A in December 2024, which is unusual because AMD is a chipmaker, not a typical venture fund. That backing gave Liquid AI both money and a hardware ally.

Earlier supporters included OSS Capital and Automattic, the company behind WordPress. The AMD name is the headline.

When a major chip company writes the lead check on an efficiency-focused model maker, it signals a bet that the future of AI is not just bigger, it is leaner.