Noosphere Labs raises $10.25M for human-centered physical intelligence

A former Meta AI research director has raised $10.25 million for a stealth startup called Noosphere Labs, which is working on what it calls "human-centered physical intelligence" -- AI designed to help people interact with the physical world beyond screens. The round was led by Trilogy Equity Partners with major participation from Madrona, signaling continued investor appetite for AI that grounds itself in real-world perception rather than just text and images.

What Noosphere Labs is building

Noosphere Labs is led by Kevin Carlberg, who spent more than five years at Meta leading research teams across Reality Labs and Fundamental AI Research, focused on physical AI and simulation for wearable and mixed-reality devices. Before Meta, he spent eight years at Sandia National Laboratories developing near-real-time physics simulations for national security. The company remains in stealth with a "Coming Soon" site that describes its mission as "human-centered physical intelligence. Built for the world we inhabit."

The startup's Form D filing lists Amy McCullough, a managing director at Trilogy, as a director, along with Tyler Simpson, a former Meta and Microsoft executive. Carlberg left Meta at the end of 2024 and incorporated Noosphere Labs early in 2026.

The technical challenge behind the vision

Although Carlberg hasn't publicly detailed the product, his recent research points directly at the problem he's trying to solve. In a NeurIPS paper last year, he and Meta colleagues built a benchmark for "assistive wearable agents" -- smart glasses and similar devices that infer what a person is trying to do from video, audio, and other signals without being explicitly asked. They assembled a dataset from 348 participants and found that the best AI models produced a relevant answer only 55% of the time. The researchers concluded that current models "remain far from practical usefulness" at the task.

That 55% relevance rate highlights the core difficulty: wearable AI needs to understand context, intent, and the physical environment in real time, and today's models are not reliable enough. Noosphere Labs appears to be targeting exactly this gap, though the specific approach remains undisclosed.

What this means for AI builders

For founders and developers building wearable, AR, or physical AI products, this funding round is a signal that top-tier VCs are willing to back hardware-enabled AI that operates in the physical world. The involvement of Trilogy and Madrona suggests that the market for real-world intelligence systems is seen as underbuilt and worth the long development cycle.

If Noosphere Labs succeeds, it could provide infrastructure or models that make it easier to build assistive agents that actually work outside the lab. For now, the startup's stealth status means builders should watch for technical publications or early partnerships that reveal the actual architecture. The NeurIPS benchmark also serves as a useful reference point: any product claiming to solve wearable AI will need to demonstrate far better than 55% relevance on similar tasks.

Caveats and unknowns

Noosphere Labs is still in stealth, and no product, API, or technical preview has been announced. The $10.25 million round is relatively modest for an AI hardware play, which may indicate an early-stage focus on research and prototyping rather than immediate commercialization. Carlberg's background in physics simulation and wearable AI is strong, but translating that into a shippable product remains unproven. Builders should treat this as a signal of direction, not a solution ready for integration.