The robotics industry has spent billions of dollars over the years trying to solve a problem that seems deceptively simple enough to not require artificial intelligence: pick an object off a table and place it somewhere else. The challenge, as researchers know, isn't knowing where the object is — computer vision has largely solved that — but how to handle it once your gripper makes contact. That final millimeter, where precision meets physics, is where Eka Robotics is betting its future.
The Cambridge, Massachusetts startup, founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, announced this week the closing of a $150 million Series A round at a post-money valuation of $1.45 billion. The round was co-led by Andreessen Horowitz (a16z) and F-Prime, with continued backing from seed leads Obvious Ventures and E14 Fund. Other investors include Samsung Next, MFV Partners, Bison Ventures, Glasswing Ventures and undisclosed Fortune 500 strategic participants.
What distinguishes Eka isn't the round size alone — several other players in the physical AI space have raised comparable amounts — but the technical approach. While most competitors use vision-language-action (VLA) models, Eka has developed a vision-force-action (VFA) architecture that treats physical forces as first-class information rather than secondary sensor data.
"Vision is not the primary way in which you close this last millimeter problem," Agrawal told TechCrunch. "Vision tells you roughly where the object is, what the pose is, but the actual grabbing, the actual searching is happening through forces."
The insight sounds straightforward when explained, but the engineering behind it is complex. To implement the VFA model, Eka had to redesign its robot gripper from the ground up — adding force sensors that react to contact, enabling the robot to adjust its grip in real time. The result is a gripper that doesn't crush fragile objects, as demonstrated in videos where the robot picks up a raspberry without crushing it — something even well-trained humans routinely fail at.
According to the company, the system handles approximately 1,500 different objects (up to 1 kg each) per hour with a success rate exceeding 95%, compared to roughly 600 objects per hour that a human operator can manipulate. Eka is currently conducting confidential pilots with industrial customers focused on pick-and-place and wire manipulation tasks.
Eka's central bet rests on reinforcement learning in simulation. Rather than relying on human demonstrations or video datasets (as rivals like Generalist and Physical Intelligence do), Eka models the physical forces encountered during manipulation and generates thousands of hours of robotic experience in minutes inside a physics simulator. The model learns through computational trial and error, not imitation.
"Rather than being limited to the data you're collecting from, say, human demonstrations, Eka is pushing through reinforcement learning this approach where the robot can surpass any kind of demonstrations that are given to it," Hauser told TechCrunch.
The robotics community has responded to Eka's work in ways that few expected. Rodney Brooks, former director of MIT's Robotics Laboratory and a fierce critic of purely vision-based models, has argued for years that robotics without force and tactile representation is doomed to fail — and Eka's results appear to validate that position. Eka's own approach suggests that the current industry consensus may be heading toward a dead end.
But perhaps the most interesting aspect of the announcement is the valuation context. Eka is being valued at $1.45 billion — which sounds high but is actually a fraction of what its top competitors are worth. Physical Intelligence (founded by the creators of Google DeepMind's pi-zero model) is in discussions for an $11 billion valuation. Generalist, from ex-DeepMind researchers, reached a $3 billion valuation in August. Skild AI (ex-Nvidia) was valued at over $14 billion.
This relative discount in Eka's valuation can be read two ways. An optimistic reading is that the market is still discovering the space and that Eka has a technical differentiator that will be rewarded in the future. The cynical reading is that investors are beginning to question whether all physical AI models actually work at industrial scale — and that $1.45 billion reflects that doubt.
What makes Eka particularly interesting is that it's closer to a real product than most competitors. The gripper costs less than $1,000, is rated for 5 million duty cycles (with a goal of reaching 20 million), and is already being tested in industrial environments. The competitors valued at $10 billion-plus remain in lab demonstration phases.
Crunchbase tallied $47.4 billion in venture capital investment in physical AI during the first half of 2026 — nearly four times the amount in the second half of 2025. The sector is growing at a pace no one predicted. But growing doesn't mean every player will survive. AI history shows that each investment wave creates winners and losers — and the winners were those who delivered practical value, not those who raised the most money.
The question is whether Eka will be able to prove that its model works on real factory floors, or whether it will become just another lab demonstration that impressed investors but couldn't survive contact with production. The next milestone will be publishing the results of its industrial pilots. If they succeed, the $1.45 billion price tag will look cheap. If they fail, it will be another case of physical AI that promised too much and delivered too little — and the $150 million will fuel the next round of simulation experiments.
Sources: TechCrunch, TechFundingNews, Humanoids Daily, TechCrunch
✓ Independent sources cross-checked and verified before publishing
