A fireside Chat with Ramin Hasani, Co-Founder & CEO of Liquid AI ft. Robert Cahill (Enhance Venture)

Company Profile & Geography

• Liquid AI is a two-year-old startup and MIT spin-off founded by Ramin Hassani.
• Headquarters are located in Boston, with existing offices in San Francisco and Tokyo, and a future European location planned.
• The company positions itself as the only foundation model firm 100% focused on deploying generative AI directly on the edge (devices).

Technological Origin & Architecture

• Core technology, "Liquid Neural Networks" (LNNs), originated from Hassani's PhD research studying the C. elegans worm.
• C. elegans was selected as a model organism because it has 302 neurons (vs. humans' 86-100 billion) and 78% genome similarity to humans, allowing for the extraction of cell dynamics and intelligence priors.
• The architecture replaces standard matrix multiplications with physical models of nervous systems to introduce biases that improve performance in small model regimes.
• Unlike Transformer models which have quadratic computational costs relative to sequence length, Liquid AI's foundation models scale linearly.
• This linear scaling significantly reduces energy consumption and compute requirements, enabling deployment on resource-constrained hardware like Raspberry Pis.

Product Strategy & Offerings

• Liquid AI sells a full-stack software suite comprising three distinct components:
  • Pre-training stack: For building foundation models in-house with optimized unit economics.
  • Customization stack: For efficiently adapting models into specialized AI systems.
  • Inference/Serving stack: For deploying models directly to devices without API limits or data center dependency.
• The company aims to balance model quality against hardware memory limitations, determining the maximum intelligence that can be packed into a device's memory footprint.

Market Position & Forecasts

• Target industries include consumer electronics (mobile, laptop, wearables), automotive (car intelligence), aerospace (satellites), and home appliances.
• The projected market size for this "on-device AI" sector is estimated at $1 trillion by 2035.
• The company asserts that the current landscape is "deserted," with few players focusing exclusively on foundation models for edge devices.
• Forward-looking vision anticipates a shift where devices replace data centers, allowing AI to function offline and in real-time for form factors ranging from phones to smart glasses.

Future AI Paradigm: Oracles vs. Edge Gatekeepers

• Hassani distinguishes between "Oracle builders" (e.g., OpenAI, Anthropic) creating massive, general-purpose superintelligent models, and Liquid AI's approach.
• The predicted future involves a coexistence of these massive "Oracles" and specialized local edge systems.
• Liquid AI envisions its technology acting as a "gatekeeper" or local intelligence that protects user privacy and safety when interacting with external Oracles.
• This local layer will handle day-to-day tasks and filter interactions, ensuring safe and direct access to global AI oracles.