Bessemer Predicts: Robotics and physical AI
Six investor predictions on the state of robotics research and the commercialization of the physical AI ecosystem—plus founders' candid takes on what's evolving in the market in 2026.
No human actually wants to do highly repetitive physical tasks or endure unsafe working conditions in factories or hazardous sites. That observation sounds simple, but the implications are not. Demographic trends across the US, Europe, Japan, and China are building structural, lasting demand for robotics solutions—demand that persists independent of any single technology cycle. Some analysts have projected the robotics market will reach $38 billion by 2035—a forecast Goldman Sachs itself revised upward sixfold in a single year. We think that's still conservative, on both pace and magnitude.
“There will be 100,000x more robots on Earth in the next 10-20 years.” — Jeremy Levine, Partner at Bessemer Venture Partners
At Bessemer, we often invest in new categories when we see the intersection of three forces: talent migration, technology breakthroughs, and structural tailwinds. In robotics and physical AI, all three are accelerating simultaneously. Already, we’ve invested in breakthrough teams, as our portfolio spans software, full systems, and foundation models: Waymo, Mind Robotics, Foxglove, Breaker, Noda, Voxel51, DroneDeploy, Auterion, Perceptron, and ANYbotics. But we're just getting started, especially after identifying the top 50 startups building in the space. Below are the six predictions shaping our investment activity in 2026—and the founder perspectives that are shaping how we see what’s on the horizon.
At a glance: Six predictions for robotics and physical AI in 2026
- We're in the GPT-2.5 moment for robotics. Capabilities are real, but the gap between lab performance and field deployment remains wide.
- Scaling laws are emerging. Data is expensive, capital is the moat. World models may be the shortcut.
- Talent concentration will crown winners quickly. This is not a market where 50 companies win.
- Near-term value will accrue to full-stack, vertically integrated players, not pure-play foundation model companies.
- Defense robotics will produce the first $50B+ IPOs in the category.
- There will be no robotics bubble. In fact, not enough capital is flowing into the industry.
[CAROUSEL OF FOUNDER QUOTE CARDS]
Thank you to the following founders and experts featured in this report, for sharing their perspectives on the market: Armen Aghajanyan (Co-Founder & CEO, Perceptron), Ian Glow (CEO, Zeromatter), Brian Moore (CEO & Founder, Voxel51), Lisa Yan (Founder, Argus Systems), Adrian Macneil (Co-Founder & CEO, Foxglove), Matthew Buffa (Co-Founder, Breaker), Nikita Rudin (CEO & Founder, Flexion), Mahesh Krishnamurthi (Co-Founder, Vayu Robotics), Brad Porter (CEO & Founder, Cobot), and Philipp Wu (Co-Founder, stealth robotics company). Read on to learn more.
Prediction 1: The ChatGPT moment for robotics is coming, but we're not there yet
The robotics industry is in its GPT-2.5 moment: foundation models are demonstrating real capability, scaling laws are beginning to emerge, but the gap between lab demos and production deployment remains wide.
Robotics is at an analogous moment. The demos are real. The underlying models are improving. The scaling laws that defined the LLM era are beginning to show up in robotics data. But generalized, reliable deployment in the physical world? That's still ahead of us.
But the progress is clear. Physical Intelligence's π0 model folds laundry from a hamper with human-level dexterity. The EgoScale paper, published in February 2026, showed that policy performance scales predictably with pretraining data size—the first strong evidence that robotics foundation models follow the same data-driven improvement curves that defined LLMs. These are not incremental results. They signal that the field has crossed into a new phase of capability.
But two hard questions remain unanswered: how much more data, and what kinds, will be needed to close the gap between lab performance and the 99.9% reliability threshold that production deployment demands? And what does the ChatGPT moment for robotics actually look like when it arrives?
Unlike a chatbot, capability can't be demonstrated through a text box. The proof point will need to be visual, physical, and undeniable: a robot performing a complex task in an unfamiliar environment, without a human in the loop. We don't think that moment is years away. But it's not here yet.
What is already here: pragmatic, narrow applications generating real commercial returns. Warehouse automation, surgical assistance, last-mile delivery, industrial inspection—these aren't waiting for the generalized moment. Purpose-built systems operating in constrained environments are producing revenue now, where the reliability bar is achievable today.
"The path to real-world robotics isn't better control algorithms — it's better foundational models that understand the physical world: models that see, reason about space and physics, and predict what should happen next. Robot control becomes a thin layer on top. Companies collecting robot demos and fine-tuning policies can solve narrow tasks, but they will not scale. The foundation is what matters."
Armen Aghajanyan, Co-Founder & CEO, Perceptron
Getting to that generalized moment requires solving a problem even harder than the models themselves: data.
Prediction 5: Defense robotics will drive the first $50B+ IPOs in the category
<iframe src="https://www.bvp.com/embed/defense-robotics-chart-series-a"></iframe>
A look inside Bessemer’s Robotics Day
Robotics Day in San Francisco was research-through-community—a way for us to learn from and connect with the founders and builders on stage as much as those in the audience. What we heard reinforced something we already believed: we are close to an inflection point in physical AI, but not yet at it. That gap, between what's demonstrable in the lab and what's deployable in the field, is exactly where the most important work is happening right now.