Lab robots are theater. Factory robots are a business. The robotics industry has produced stunning demos for decades. Robots folding t-shirts. Making coffee. Performing flawlessly in controlled environments. Then failing the moment they hit a real factory floor.
That gap between a perfect lab demo and a messy, high-variability production environment is the chasm the industry hasn’t crossed.
Rhoda AI is crossing it. Today, they are coming out of stealth and announcing their $450M Series A to prove it.
I wrote about why Physical AI is the next era of Labor as a Service in Part 1 of this series. Now I want to explain why we backed Rhoda specifically, and what their systems-thinking approach reveals about how Physical AI companies actually win.
The problem no one talks about
AI has transformed how we process information. Words, images, video – symbolic AI is genuinely intelligent. But the physical world is different. The moment you ask an AI to move a pallet, navigate a crowded warehouse aisle, or handle an object it’s never seen before, today’s models break down.
The reason is something called a distribution shift. The data a robot trains on in a lab is narrow and controlled. The real world is messy, unpredictable, and full of variation. Most AI robot models – what the industry calls Vision Language Action models, or VLAs – simply don’t have the generality to handle that gap.
Rhoda does. Their proprietary architecture is purpose-built to generalize across real production environments, not just replicate controlled demos. Rather than relying on teleoperated robot trajectories, Rhoda pre-trains its models on internet-scale video – hundreds of millions of videos – to build a deep prior on motion, physics, and how the physical world behaves. They then post-train on robot-specific data to learn embodiment behaviors. The result: new tasks can be learned with as little as 10 hours of teleoperation data, compared with the thousands of hours most approaches require.
The system operates in a closed loop – continuously observing, predicting, acting, and re-observing every few hundred milliseconds. It doesn’t execute a plan and hope for the best. It adapts in real time.
Rhoda has already demonstrated autonomous operation in live manufacturing environments: a component-processing workflow completed in under two minutes per cycle, with no human intervention, exceeding customer KPIs. That’s not a lab result. That’s production.
Labor as a Service: The $30 trillion market for work
Rhoda isn’t selling a robot. They’re selling work.
Think about what constrains GDP growth. Capital is essentially unlimited — you can always deploy more dollars. But labor is finite. There aren’t enough workers to get the tasks done, especially in manufacturing and logistics. The jobs go unfilled. Output gets capped.
Rhoda is building what I’m calling “Labor as a Service” (LaaS). They’re not targeting jobs people want to do. They’re targeting the dull, dirty, and dangerous tasks that humans shouldn’t have to do in the first place. Think repetitive pallet movement – the kind of work that sits unfilled, constrains productivity, and keeps operations leaders up at night. By automating it, Rhoda frees humans to focus on higher-value exception handling.
The global market for manual labor is roughly $30 trillion annually. Even fractional penetration of that market is a business of staggering scale. At 1,000 deployed units, Rhoda generates ~$100M in recurring annual revenue. At 10,000 units, it’s a billion-dollar ARR business. The unit economics are as compelling for the customer as they are for Rhoda – because every robot deployed unlocks more output from the humans alongside it.
Bottom line: This isn’t a robotics company chasing a robotics market. It’s a robotics company chasing a labor market. That reframing changes everything about the size of the opportunity.
The wedge strategy
Jagdeep Singh has built seven companies. He’s not trying to boil the ocean on day one with general-purpose humanoids.
Rhoda is starting with manufacturing and logistics – environments that are structured enough to scale, but complex enough to demand their proprietary approach. The strategy is deliberate: pick a problem that’s real, narrow enough to win, and hard enough to defend. Solve it with high confidence and work with design partners with real demand. Then let the data flywheel take over.
Every deployment generates real-world data. That data makes the models more general. More generality unlocks the next category of tasks. And so the wedge widens.
This is the opposite of the viral humanoid demos flooding our social media feeds. Those are lab experiments. Rhoda is building for mission-critical workflows that enterprises actually need solved today.
Why systems thinking wins
At Mayfield, we have a simple belief: founders create categories, not companies.
What distinguishes the category-defining founders from everyone else is systems thinking. Jagdeep and his team don’t just build software and integrate third-party hardware. If a sensor or actuator doesn’t exist that solves their customer’s specific problem, they build it. Every piece of the system is designed to work together.
That’s harder. It takes longer. But it produces larger, more defensible, and more durable businesses. Systems companies don’t get commoditized the way point solutions do.
We’ve seen this pattern before and across Mayfield’s history of backing infrastructure builders. The founders who solve the whole problem – not just the easiest slice of it – are the ones who build the defining companies of each era.
The contrarian bet
When Jagdeep and I sat down, he said something that stuck with me: all value is created when you’re contrarian. In robotics, the consensus bet is on general-purpose humanoids and impressive demos. The contrarian bet is on unglamorous, mission-critical work that real enterprises desperately need solved – and that requires a fundamentally different technical approach to get there. That’s the bet Rhoda is making. Heads down on the hard problem. After 18 months in stealth, they are unveiling FutureVision, a new approach to robotic intelligence based on video-predictive control and designed to operate beyond controlled laboratory demonstrations and into real-world environments. And they’re ready to accelerate development and industrial deployment.
That’s exactly the kind of bet we want to make at Mayfield.
What’s next
We’re proud to partner with Jagdeep and the Rhoda AI team as they build what we believe could be the defining company in physical AI – and perhaps the largest market opportunity in technology.
The Industrial Revolution increased GDP growth from fractions of a percent to several percent annually. If Rhoda’s vision plays out, we may be in the early innings of a similar unlock – one powered not by steam, but by robots that can actually work in the real world.
