Shawn Shen believes that AI will need to remember what it sees in order to succeed in the physical world. His company Memories.ai is using Nvidia AI tools to build the infrastructure for wearables and robotics to be able to remember and recall visual memories.
The company announced a collaboration with Nvidia at its GTC conference on Monday, using Nvidia's Cosmos-Reason 2, a reasoning vision language model, and Nvidia Metropolis, an application for video search and summarization, to continue developing its visual memory technology.
The announcement reflects a significant gap in how AI systems currently process the world. Shen argues that while "AI is already doing really well in the digital world", the question becomes: "What about the physical world?" He contends that "AI wearables, robotics need memories as well" and that "ultimately, you need AI to have visual memories."
While AI memory itself is relatively new, with OpenAI updating ChatGPT to remember past chats in 2024 and others launching memory tools over the past two years, these advancements have largely focused on text-based memory. Text-based memory is much more structured and easier to index but isn't as helpful for physical AI applications that largely interact with the world through sight and visuals.
Shen and co-founder Ben Zhou, both former Meta Reality Labs researchers, got the idea for the company while building the AI system behind Meta's Ray-Ban glasses. Building the AI glasses got them thinking about how people would actually use the tech in real life if users couldn't recall the video data they were recording.
The infrastructure play reflects Memories.ai's deliberate positioning. In terms of commercialisation, Shen explained that the company is "more focused on the model and the infrastructure, because ultimately we think the wearables and robotics market will come, but it's probably just not now." The company is already working with some of the large wearable companies, though Shen declined to disclose which ones.
Memories.ai was launched in 2024 and has raised $16 million thus far, through an $8 million seed round in July 2025 and an $8 million extension. The company has also partnered with Qualcomm to bring its visual memory model on-device, with capabilities expected to become available in consumer devices starting in 2026, allowing consumers and businesses with Qualcomm-powered phones, cameras, and wearables to turn raw video into searchable and structured memories.
The company's approach addresses a long-standing problem for video analysis. Unlike text, which large language models have made easy to search and understand, video remains dense and unstructured. Memories.ai's technology compresses video into indexed memories that AI can query, opening applications across smart glasses and wearables with enhanced AI recall, security agents that help cameras understand and respond in real time, and robotics with better context and understanding.
Whether these applications prove economically viable at scale remains uncertain. The wearable AI market is expected to grow significantly, but yesterday's failed experiments like the Humane AI Pin and Rabbit R1 demonstrate that raw capability does not guarantee commercial success. Memories.ai is betting that solving the memory problem is the missing ingredient to make future wearables actually useful in everyday life.