Eka Robotics raises $150m to teach robot hands physics, not imitation

Eka Robotics raises $150m to teach robot hands physics, not imitation

20 October 2026: Eka Robotics has raised a $150 million Series A led by Andreessen Horowitz and F-Prime, the venture fund affiliated with Fidelity's parent FMR, to scale a robot manipulation model trained in simulation rather than on recordings of people. The company says the money will go into extending its model to more tasks and accelerating customer deployments.

Not yet independently verified. The $1.45 billion valuation and the founders' backgrounds appear in press coverage, not in Eka's own post, and Eka does not state where it is based. The throughput figure is the company's own. We will update this when it can be confirmed, and remove this note.

Who is in the round

Eka names the investors in its announcement:

• Leads: Andreessen Horowitz (a16z) and F-Prime
• Returning seed leads: Obvious Ventures and E14 Fund
• Others: MFV Partners, Bison Ventures, Glasswing Ventures, Samsung Next and "several Fortune-500 strategics"

The post does not give a valuation. TechCrunch, which Eka links from its own site, put it at $1.45 billion.

"Eka's Vision-Force-Action model is the first approach I've seen that truly solves manipulation with the generality and throughput needed for deployments," said Raghu Raghuram, general partner at a16z. Kahini Shah, partner at Obvious Ventures, summed up the case for the founders, Pulkit and Tuomas, as "Team and Dream".

A model that learns by feel

Most robot foundation models learn from demonstrations: people teleoperate a robot and the model copies them. Eka argues that this caps speed at human level. Its vision-force-action (VFA) model combines what the camera sees with what the hand feels, the force on a grasp or the resistance of a cable that is not yet seated, and learns through reinforcement learning in simulation, where it can gather "thousands of hours of robot experience in minutes". The company compares the approach with AlphaZero, which learned chess and Go without human games.

Its stated result is that robots running the model pick objects they were not trained on at 1,500 operations an hour, "roughly two and a half times the human baseline of about 600". The same robot, it says, can screw in a light bulb and insert wires. The first target is dexterous manipulation, with packing soft fruit and assembling GPU server racks given as examples.

What buyers of automation should watch for

For warehouse, food packing and electronics assembly operators, the claim to test is the one Shyam Kamadolli of Fine Structure Ventures puts in the post: a robot's value on the factory floor "comes down to two things: its success rate and its speed over thousands of repetitions". Eka publishes a speed figure but not a success rate, and simulation-trained systems have historically lost accuracy when moved onto real lines, the "sim-to-real gap" the company says is now closing. A pilot priced on sustained picks per hour at an agreed error rate, rather than on demonstration speed, is the way to hold it to the claim.

Eka sells robots running its model rather than the model alone, so it competes both with robot makers and with the integrators that fit grippers and vision to existing arms. Our robot manufacturer category lists 59 companies and the robotics integrator category 25. A WIRED review of an earlier version of the system in April was the company's last public milestone before this round.

Company profile on TrustList: Eka Robotics

Sources

• Eka Robotics: Eka Robotics raises $150M to build superhuman intelligence for the physical world, 20 October 2026
• Wowtale: Eka Robotics Series A, 21 October 2026