Enact's $30M Pre-Seed Bets Robot Dexterity Starts in the Actuator

Manuel Yves Galliker spent the past few years teaching humanoids to move. At 1X he led the software team behind NEO, a machine built to work next to people. The demos looked smooth. The hands kept giving the game away.

A robot can find a cup, plan a path to it and close its fingers around the handle. Then the cup slips. Or the fingers crush it. Or the arm stalls halfway through a plug because the socket is two millimetres off. Vision tells the machine where the object is. It says almost nothing about how hard to hold on.

Galliker left 1X, helped build its AI team in Silicon Valley, then joined Physical Intelligence. In March 2026 he founded Enact Intelligence in Zurich. On 22 October the company came out of stealth with $30 million in pre-seed funding, one of the largest rounds of its kind in Europe. The pitch is narrow: give robots a nervous system that measures force inside their own actuators, then turn the resulting data into the moat.

The hand is where the demo dies

Most commercial robots perceive through cameras. That works for picking a box off a pallet in good light. It works less well for inserting a connector, fastening a bolt, or handling something soft enough to deform under its own weight.

Haptic add-ons exist. Fingertip pads and wrist-mounted force-torque sensors can measure contact, and research labs have used them for years. The trouble is industrial maths. Every extra sensor adds cost, cabling and a failure point, and the readings arrive at the fingertip rather than across the whole body. Enact argues the sensor belongs inside the actuator, the motor-and-gearbox unit that already moves the joint.

The analogy the company uses is muscle. People do not feel a heavy door only through their fingertips. They feel tension through the arm, the shoulder and the back, and the brain merges those signals into one sense of effort. Enact calls its version a proprioceptive nerve system plus a cortex that makes sense of it. The engineering claim is that force becomes a continuous signal from hardware the robot already carries.

Force as a data problem

“Our robots sense force and touch through their own actuators, so they feel what’s happening in real time, and learn more data-efficiently while simultaneously pushing reliability and performance,” Galliker said when the round was announced.

We’re rethinking the full stack, from model architectures enabling multi-frequency and multi-scale intelligence, to the data pipelines all the way down to the actuators and sensors that feed the model.— Manuel Yves Galliker, founder and CEO, Enact Intelligence

That full-stack language matters for how the company plans to compete. Enact is not selling a software layer that runs on anyone’s hardware. It is designing bespoke actuators, the control electronics around them, and the models that consume their signals. Job listings reviewed by SiliconANGLE point to custom pre-training work and FPGA-based control, which suggests the team intends to own the parts of the stack that determine signal quality.

As we wrote in October, Innatera’s 200-neuron finger takes a different route to the same problem: neuromorphic sensing at the point of contact. Enact is betting the joint, not the skin, is the better place to feel. Both approaches produce force data. The difference is where the hardware cost lands and how easily the data transfers between robot shapes.

That transfer question is the commercial heart of the story. A model trained on one humanoid’s force signature does not automatically work on a different arm with different gearing. Enact says its approach is form-factor agnostic, from a simple arm to a full humanoid. If true, the company can sell the same intelligence across many machine designs. If false, it has built an expensive component business with a software story attached.

Dr. Harald Nieder, general partner at redalpine, framed the investment as a data play rather than a hardware bet. “Dexterity is the next frontier in robotics, and Enact is tackling it with a new data modality: starting with high precision force data from bespoke actuators they are rethinking the entire stack from the hardware up,” he said. Kamil Mieczakowski of Notion Capital put it more bluntly: “Robots won’t generalise until they have a nervous system, one that lets them not only see the world but feel it.”

A $30 million bet on a twelve-person team

The round was led by redalpine, Notion Capital and firstminute capital. Angel investors include Yann LeCun, Mistral founder Arthur Mensch, Hugging Face co-founder Thomas Wolf and Andrew Lonsberry of Path Robotics. Enact’s legal name is Eta Robotics AG, a detail that rarely makes the headlines but matters for anyone tracking the cap table.

Twelve people built the first version. The team draws engineers from Google DeepMind, 1X, Amazon Robotics, Physical Intelligence and ETH Zurich. Philip Kneis, previously a deeptech investor at redalpine, joined to lead growth, a move that mirrors a wider pattern of European VCs crossing into operating roles.

The location choice is deliberate. Galliker could have stayed in the Bay Area, where the AI talent and the largest robotics cheques sit. He picked Zurich. “Silicon Valley has traditionally been very strong in software and AI, but it’s lacking in robotics and hardware engineering talent, and with the current excitement around robotics a lot of money is chasing a few people,” he said. “What we will do is bring the AI experience with us from Silicon Valley and combine it with Zürich’s local talent pool.”

That is a credible arbitrage for a company whose product depends on mechanical engineering as much as model quality. It also sets a ceiling. A twelve-person team can prototype fast. It cannot yet run a supply chain, qualify actuators for factory floors, or support customers across four industries at once. The fivefold hiring plan by the end of 2027 is the bridge between the two.

What the money has to prove

Enact says it is in discussions with major European industrial players in manufacturing, infrastructure, retail and food processing. The first deployments will be the real test. A pilot that runs for a month in a controlled cell proves the demo. A deployment that survives dust, vibration and shift work proves the business.

Three things have to go right. The custom actuators have to hit a cost and reliability point that industrial buyers accept. The force models have to transfer across tasks and robot designs without a fresh data collection project each time. And the company has to keep its lead while better-funded humanoid makers build similar sensing in-house.

The competitive field is crowded. Physical Intelligence, where Galliker worked, is building general-purpose robot foundation models. Figure and Tesla are vertically integrating humanoids. 1X, his former employer, is building a humanoid with tendon-driven hands that sense touch. Each of them can add force sensing. The question is whether they can match a company that designed its actuators around the signal from day one.

Enact’s answer is that vision and language are becoming commodities, while force is not. “Unlike the early days of LLMs, physical AI has the advantage of starting from strong reasoning, language, and vision capabilities already in place,” Galliker said. The company can build on existing video models and adapt them to output robot actions. What it adds is the part the video models never had: a measurement of resistance.

The bet, then, is not that robots will become dexterous. Almost every well-funded robotics lab is chasing that. The bet is that the durable advantage sits in the hardware layer that produces force data, and that whoever owns that layer can sell intelligence across every robot shape that follows. Thirty million dollars is enough to test that proposition. It is not enough to prove it.