Enact Leaves Stealth With $30 Million and a Bet That Robots Need to Feel, Not Just See

A twelve-person robotics startup in Zurich came out of stealth on Thursday with $30 million in pre-seed funding, one of the largest rounds of its kind in Europe, and an argument that most of the industry is building robots that can see but cannot feel. Enact Intelligence, founded in March 2026 by Manuel Yves Galliker, is developing hardware and models that read force and touch from the actuators inside a robot rather than from cameras or dedicated fingertip sensors.

Institutional investors redalpine, Notion Capital and firstminute capital led the round, which Enact valued at roughly 26.68 million euros. The angel list is more interesting than the institutional one: Yann LeCun, founder of AMI Labs and former chief AI scientist at Meta; Arthur Mensch, founder of Mistral; Andrew Lonsberry of Path Robotics; and Thomas Wolf of Hugging Face, alongside Northvolt co-founder Paolo Cerruti and ETH Zurich professors Marc Pollefeys and Thomas Hofmann. The company operates publicly as Enact, while corporate records list the registered entity as Eta Robotics AG in the canton of Zurich.

The technical thesis is narrow and specific. Vision-first robots handle pick-and-place well and fail at the last millimeters of assembly, where a machine has to feel whether a connector is seated, a clip has snapped or a screw has caught its thread. Adding haptic sensing usually means bolting on force sensors that are either too expensive or too fragile for a factory floor. Enact instead captures force data inside custom actuators, an approach its founders compare to how humans sense tension through muscle rather than only through fingertips, and pairs it with model architectures built for multi-frequency and multi-scale signals. Galliker describes the goal as a proprioceptive nervous system plus the cortex that makes sense of it.

The founder path explains the ambition. Galliker led the software team for the NEO humanoid at 1X, helped build that company's AI team in Silicon Valley, and later worked at Physical Intelligence. The current team draws engineers from Google DeepMind, 1X, Amazon Robotics, Helsing, Physical Intelligence and ETH Zurich, and Philip Kneis, formerly a deeptech investor at redalpine, joined to run growth. Galliker argues that physical AI starts from an unusual position of strength because reasoning, language and vision already exist, so the work is adapting models such as video models to output robot actions rather than teaching a machine common sense from scratch.

Enact says the money will fund hardware and model development, initial deployments with industrial partners in manufacturing, infrastructure, retail and food processing, and a fivefold expansion of headcount by the end of 2027. The commercial pitch is aimed at work where variability rather than volume determines value: short production runs, frequently changing products, and tasks where conventional automation is too expensive to reprogram. Galliker suggests that demonstrating a task could eventually replace some of that programming, a claim that will need evidence before anyone rewrites a line.

That evidence does not exist yet. Enact has disclosed no valuation for the round, no named customers, and no independently measured manipulation performance, and its first industrial deployments are planned rather than completed. Investors are pricing a thesis about a data modality, not a shipping product. Harald Nieder of redalpine described the bet as leveraging orthogonal data to improve existing models and build a data moat, while Kamil Mieczakowski of Notion Capital put it more simply: robots will not generalise until they have a nervous system that lets them not only see the world but feel it.

The round also says something about where physical AI capital is going. A pre-seed of this size for a company with a dozen employees and no product is only rational if force data is genuinely defensible, since cameras and vision-language models are commoditising quickly while contact-rich manipulation data remains scarce and expensive to collect. If Enact is right, the bottleneck in robotics is not intelligence but proprioception, and the moat is a dataset nobody else can gather without building their own actuators.