The Robotics Accelerator

Physical AI has entered that era.

In the past decade, the field steadily evolved from an emerging entrepreneurial hot spot to one of the greatest opportunities of our modern era. Now, Physical AI is advancing at an exponential rate, driven by the convergence of powerful (and accessible) AI models, lower compute and battery costs, and the maturation of world-class talent from the previous wave of autonomy pioneers like Waymo, Tesla, Kiva, and 6 River Systems.

But behind the scenes, there’s another element supercharging the Physical AI movement: The emergence of widely available developer tools and infrastructure that enable engineers to rapidly design, build, test, and iterate robotics and autonomy. Foxglove is leading this category today, providing engineers with the data infrastructure to develop and operate Physical AI for every critical industry, from construction and healthcare to transportation and defense. 

Working with more than 200 customers — up from 10 just two years ago — Foxglove has become an indispensable partner to the boldest innovators in the field of robotics. Having achieved this traction with its observability products, the company is now preparing to expand to serve the complete Physical AI infrastructure stack. Eclipse is thrilled to be supporting them in that effort as investors by participating in Foxglove’s $40 million Series B. We’re excited to deepen our partnership with the Foxglove team as they build the comprehensive platform to power the full lifecycle of robotics, solidifying their platform as one of the fundamental layers of the entire ecosystem. 

I’ve been working with Foxglove since Eclipse led their Series A in 2022, and have been partnering with robotics startups my entire investing career. I’ve been lucky to witness firsthand some of the most ground-breaking milestones in this field. But nothing can compare to the rate and volume of progress we’ve seen across the category in the past two years. Foxglove’s growth in the past 18 months perfectly encapsulates this moment.

Interestingly enough, the company’s vision wasn’t obvious to everyone at first. Founded and led by cloud and infrastructure engineering leaders from Cruise, Coinbase, and Stripe, Foxglove set out to tackle a problem everybody in the field had but nobody wanted to solve at scale.

Since the initial rise of autonomous driving and other applied robotics in the mid 2010s, developing and testing products required each company to build their own infrastructure in-house, from scratch. This only became more of a bottleneck to progress as the field evolved over the next decade, which led Foxglove co-founders Adrian Macneil and Roman Shtylman to start imagining what modern developer infra for robotics should look like. Having experienced firsthand what the advent of off-the shelf, best-in-class infrastructure had done for Web 2.0 by lowering the barrier to entry and enabling rapid scale, they drew parallels to the AV industry. It was still in its comparative pre-SaaS moment, which had to build everything from the ground up. 

I met Foxglove when their product was in its early stages of development, and Adrian and Roman were pitching customers on their observability platform. This is a universally useful application — collecting and gathering multimodal sensor data is paramount to develop, test, debug, or operate Physical AI systems. But initial customers couldn’t seem to grasp how an outside product could actually work for their unique applied robotics.

Until they tried it.

“Everybody was just ready to move a lot faster,” says Macneil. “We’ve moved on from the previous generation, when all autonomous companies were forced to build their own infrastructure in-house, to one where all the undifferentiated layers can be commoditized. Now companies can focus on finding their core value proposition, building things that are more differentiated for their specific application of robotics, instead of building plumbing layers.”

Using Foxglove, companies ranging from Bedrock, which is automating heavy machinery, to autonomous driving startup Wayve have been able to move through product iterations significantly faster. 

“Having a toolset like this is huge,” says Bedrock Co-Founder and CEO Boris Sofman, whose company launched in July 2025 in partnership with Eclipse Venture Equity. “When you can actually lean into the existing ecosystem on things that are not core to the IP but are still critical, like visualization or infrastructure for ML training, it’s really enabling.”

Tools like Foxglove are part of the reason Bedrock was able to get up and running so quickly, Sofman says. Officially founded in 2024, the company had a working prototype within six months, and today has dozens of machines outfitted with the Bedrock system on construction sites around the country. That’s night and day compared to his previous experience (both as a lead engineer at Waymo as well as his prior role as the CEO of consumer robotics startup Anki), says Sofman.

Along with saving time and money on the initial build of products, Foxglove allows users to learn more about the actual problems they are trying to solve. This is especially helpful for getting multiple team members aligned on the development of a product, including those without a formal robotics background, says Mike Brevoort, the Principal Engineer and AI lead at warehouse automation startup Mytra.

“It drives you down a certain path for how to solve problems but it also gives someone from the outside the opportunity to look at those from a first principle's perspective and question, ‘Why is it that way? And ‘Should I take that same road and follow that?’” says Brevoort. “Or maybe you can then start to apply some learnings from adjacent industries, domains to say, ‘Well, why should it be that way? Can it be different?’ A lot of data emerges that is really helpful.”

Getting team members from various disciplines aligned on how to develop tools that will actually work for the problem they are trying to solve is critical for any robotics company, which will inevitably come into contact with human beings. This is heightened for those working in complex industries such as defense or healthcare.

For Shield AI, which provides Hivemind, an autonomy stack that accelerates the development of autonomous systems across multiple domains, including aircraft operating in contested environments, Foxglove helps the company and its customers develop, test, and visualize autonomy faster. The Foxglove platform enables Shield AI to focus on advancing Hivemind’s capabilities while supporting customers as they build and field AI-enabled autonomy that can continually evolve to meet mission needs. It also helps streamline collaboration with partners who may not have deep technical backgrounds in robotics or autonomy, says Tom Schaefer, Shield AI’s VP of Hivemind.

“Foxglove and Hivemind together lower the barrier to entry,” says Schaefer. “Our partners have a wide range of expertise and experience levels, and this tooling enables them to meaningfully interact with our systems and achieve their mission objectives in support of national security.”

We don’t need robots to become commonplace just for the sake of having cool technology on our streets, workplaces, homes, and anywhere else we go. We need robots in order to move forward as a society — from an economic, national security, and everyday health and safety perspective. From labor shortages to the need for advanced manufacturing and efficiency, demand for automation is here. Industrial applications of embodied AI represent the greatest opportunities to contribute to this effort.

Intelligent robotics may not have had a “ChatGPT moment” of its own yet — because it probably won’t. Physical AI just isn’t the same as digital AI, so instead of a singular, definitive product to essentially rule them all, we expect the true Turing Test moment of embodied AI will sneak up around us. Embodied AI applications are already making real traction in the real world, and one day they’ll fade into the background.

Foxglove aims to be the fuel to help us get there.

“People need to be able to focus on their actual business domains and solve their robotics and autonomy problems,” says Shtylman, who serves as CTO. “We need to focus on building even better tools that help them build faster.”