When Rhoda AI stepped out of stealth mode on March 10, the Palo Alto startup didn't do things by halves. A $450 million Series A at a $1.7 billion valuation—among the largest opening institutional rounds ever raised by an AI robotics venture—buys a lot of credibility. It also buys time to prove whether training industrial robots on internet video, rather than the expensive teleoperation datasets competitors favor, is brilliant or just well-funded optimism.
Premji Invest led the financing, drawing in an eclectic group that included Khosla Ventures, Temasek, Mayfield, Capricorn Investment Group, Leitmotif, Matter Venture Partners, Prelude Ventures, Xora, and the legendary John Doerr writing a personal check. The round coincided with the public unveiling of FutureVision, Rhoda's robot intelligence platform, which had been gestating in stealth for a year and a half.
That valuation catapults Rhoda into rarefied air for an early-stage robotics company. Though the funding narrative gets a little murky when you dig. Forbes reported last October that Rhoda had already closed a $162.6 million Series A back in April 2025, valued at nearly $1 billion, citing PitchBook data. The company's most recent materials simply call the $450 million the Series A, suggesting that earlier report may have been a misunderstanding or reclassification in the context of this March announcement. Rhoda declined to elaborate.
A Founder With Hardware Pedigree
CEO Jagdeep Singh is no stranger to moonshot ventures. He previously co-founded Infinera, an optical networking company, and QuantumScape, the solid-state battery maker that went public via SPAC in 2020 and has faced its share of production delays. Singh served as QuantumScape's CEO until stepping down in February 2024 before turning his attention to robotics. His co-founder at Rhoda is Gordon Wetzstein, a Stanford professor specializing in computational imaging, who serves as scientific advisor rather than full-time executive.
The company, founded in 2024, lists between 51 and 200 employees on its LinkedIn profile as of March. It's still building out the team, in other words.
The Technical Gambit
What Rhoda is selling—philosophically and technically—is a departure from how most robotics companies think about training. FutureVision relies on what Rhoda calls Direct Video-Action (DVA) architecture. The system pre-trains on what the company describes as "over a million videos" scraped from the open internet, learning patterns of motion, physics, and causality before it ever touches a real robot.
Only after that foundation is laid does the platform fine-tune on robot-specific data—and here's where Rhoda makes its boldest claim. The company says it needs just 10 to 20 hours of demonstration footage to teach a robot a new task. That's orders of magnitude less than the teleoperation-heavy approaches pursued by competitors, which rely on humans piloting robots through tasks thousands of times to build training sets.
FutureVision reframes robotic control as a problem of causal video generation. The model predicts what should happen next in a visual sequence, then uses inverse dynamics to translate those predictions into motor commands. In demonstrations published alongside the funding announcement, Rhoda showed an automotive decanting task—moving parts from one container to another—running autonomously for 90 minutes after training on roughly 11 hours of data. The system also supports "one-shot" imitation learning, where a robot picks up a new behavior from watching a single human demonstration.
Whether this works outside the lab is the question that $450 million is meant to answer.
Industry Deployment, Hardware Ambitions
Rhoda says the fresh capital will fund rollouts across automotive, manufacturing, logistics, and ecommerce customers, though it hasn't named any publicly. The company has run successful pilots with at least one automotive manufacturer using off-the-shelf robotic hardware, positioning itself primarily as a software and intelligence layer that can be licensed to existing systems.
That said, Rhoda isn't ruling out building its own hardware. The company's website lists specifications for custom actuators rated to 25 kilograms with 40-kilogram peak loads, and it has hinted at developing humanoid-form-factor robots down the line. But the pitch, for now, is that Rhoda is a platform play, not a hardware manufacturer competing with the likes of Boston Dynamics or Agility Robotics.
The timing is interesting, if not surprising. Humanoid and industrial robotics have become one of the hottest sectors in venture capital, with Tesla pushing its Optimus bot, Figure AI raising at a $2.6 billion valuation earlier this year, and China's Unitree shipping low-cost humanoids in volume. Rhoda's internet-video training represents a distinct technical bet in a crowded field—one predicated on the idea that the internet contains enough implicit knowledge about how the physical world works to shortcut traditional data collection.
Investors, clearly, are intrigued. Whether customers will be requires proving the platform can handle the messy, unstructured chaos of real factory floors, warehouses, and assembly lines. Demonstrations are one thing. Production uptime is another entirely.
