Scaling Intelligent Robotics: How Intel Is Helping Bring Physical AI to the Enterprise

Enterprise robotics already runs in production across manufacturing, warehousing, and logistics. The next hurdle is moving a pilot into full deployment, and John Healy, Vice President and General Manager of Intel's Industrial Robotics and Automation group, points to two issues behind that gap: adapting a model to real-world conditions, and sustaining the total cost of ownership at scale. At the Six Five Summit: AI Unleashed 2026, David Nicholson and Ryan Shrout spoke with Healy for a Connected Intelligent Edge, Networks and Agents Spotlight interview.

Healy separates today's enterprise robotics market from the humanoid demonstrations drawing attention on trade show floors. Most of today's deployed value already runs in production: fixed-function arms, rail-mounted warehouse robots, and small delivery AMRs. Physical AI is changing inside those systems, giving robots a sense-reason-act cycle. Healy's example: a welding robot that spots a defective weld can now adjust its own parameters for the next one, instead of only flagging an alarm.

Two Intel investments target the deployment gap directly. The Physical AI Studio lets teams fine-tune models through imitation learning without writing code. The OpenVINO Physical AI Framework standardizes how a trained model integrates with a robot's other subsystems. Both tie back into cost: openness across hardware and software layers gives enterprises vendor choice, and that choice brings total cost of ownership down enough to justify scaling out of the lab.

Key Insights:

🔹 Physical AI turns existing robots into adaptive systems, adding a sense-reason-act loop that lets a welding arm detect and correct its own output instead of simply triggering an alarm.

🔹 The near-term enterprise opportunity is not humanoids, but the robots already at work: fixed-function arms, warehouse systems, and autonomous mobile robots operating in production today.
🔹 Where robot intelligence runs depends on the application, with latency, connectivity, and data-security requirements determining whether processing stays on the robot, at the edge, or in the data center.

🔹 Intel’s Physical AI Studio lowers the barrier to robotics development, enabling teams to fine-tune models through no-code imitation learning without requiring specialized AI expertise.

🔹 The OpenVINO Physical AI Framework addresses deployment complexity, while an open hardware and software ecosystem helps reduce integration costs and vendor lock-in.

🔹 Intel’s roadmap is built around scalable economics, combining CPUs, integrated GPUs, and NPUs for real-time control with an open ecosystem designed to keep total cost of ownership manageable as deployments grow.

Watch the full session at sixfivemedia.com/summit, and explore the rest of our Six Five Summit: AI Unleashed 2026 coverage.

Disclaimer: Six Five Media is for information and entertainment purposes only. Over the course of this video, we may discuss companies that are publicly traded, and we may reference their equity share prices. Nothing discussed during this webcast should be considered investment advice or a recommendation to buy or sell any security. We are not investment advisors, and you should not rely on this content as financial advice. Six Five Media collaborates with technology companies and industry leaders to produce research-driven interviews and multimedia programming for enterprise technology audiences.

‍

John Healy:
The difference is that the robot takes action based on what it learned. That's the adaptability. And so the change in the production flow or in the development flow is ensuring that that optimization sequence or capability is built as part of the solution.

David Nicholson: 

Hi everyone and welcome to the Six Five Summit AI Unleashed 2026. For this Connected Intelligent Edge Networks and Agents Spotlight, we'll be taking a closer look at the next phase of enterprise robotics. While breakthroughs in AI continue to capture headlines, organizations are increasingly focused on a different question, how to deploy intelligent AI-powered systems at scale in real-world operational environments. I'm here with my colleague, Ryan Shrout. And joining us is John Healy, Vice President and General Manager, Industrial Robotics and Automation at Intel. John, welcome to The Six Five. Thanks David, I'm delighted to be here. So John, tell us more about your role at Intel.

John Healy: 

Yeah, so David, I have the privilege of working with the industrial industry, manufacturers and utilities in enabling them to really transform the way they do what it is they do. A lot of it software centric and compute centric as we enable them to bring new possibilities to the way they implement their production lines, the way they manage their utilities grid and really enable more intelligence in their production flows. And a big part of that is enabling them through that softwarization of their processes to apply more AI at the edge, at the point of application. And as we deploy more and more AI, we're enabling increased levels of automation, a lot of which leads to robotics. So a great synergy between the different domains. And it's a super exciting part of the industry to be working in right now.

Ryan Shrout: 

We're here to talk about kind of enterprise robotics specifically. Which I'm very curious in your mind what falls under that definition of quote unquote enterprise robotics. I think a lot of the attention today in the robotics industry itself is focused on you know, humanoids, things that kind of look like us and move like us and that are going to, you know, help me do my laundry and walk up and down the stairs. But it does feel like we're overlooking to some degree what enterprise robotics is actually already doing today, what value it creates in the market today. Do you think that's actually the case?

John Healy: 

Yeah, I think it's a true statement, Ryan. It's a great question, by the way, because enterprise robotics have existed for, or robotics in enterprises, have existed for years now. We've worked with the industries in deployment of robot functions for decades, actually, with a lot of the technologies we provide. And there are things people might be familiar with. An enterprise robot in a manufacturing environment is like the fixed function arm you'll see sitting on doing pick and place or doing assembly on a production line. We've often seen pictures of vehicles being manufactured in an automotive vendor and you'll see the welding occurring and those are robot arms moving very large components of the vehicle together and supporting the assembly. Or there are robots that are moving pallets around in a warehouse, in a logistics environment, where there is a flat robot running along on rails, sometimes on wheels, moving goods around inside in the warehouse. And some of us have seen the robots in the street doing deliveries like DoorDash, bringing things to your home or bringing things to your place of business. And it's a small AMR, an autonomous mobile robot doing that work. All of those are classes of enterprise robots. But you mentioned then the advent of the humanoid. And I think what's exciting about humanoids and what grabs the attention is that we tend to relate very quickly to that view of robot. A lot of us can remember, I certainly can see 3PO in Star Wars, classic humanoid, gold plated. And we said, wow, he's sick of it like us. And so the humanoid, as we see them today, are very much representative of what robots can become when we want to deploy more forms of robotics in applications and in environments that have been actually designed around the human and replace that or support that and augment that with robotics. But it's a very wide spectrum. And most of the actual value being derived in the enterprises is the work we do with them in all of the existing forms of enterprise robots that I mentioned, the fixed, the more fixed, functional kind of constrained environment robots that add a tremendous amount of value in supporting production lines and logistics support and warehousing, as I mentioned already.

David Nicholson: 

Well, as these robots get more and more intelligent, how does that change the way that technology is deployed? What's your view on where that intelligence needs to live?

John Healy: 

Yeah, robots have developed, they've really sort of, they're on a continuum, David. You know, we've seen robots, it's initially designed to support primarily a specific task. I talked about the assembly robot. And then with the advent of computer vision, robots started to be capable of seeing what they were working on. And so in supporting customers to add computer vision, we began the journey to forms of AI. And now, as you mentioned, as it becomes increasingly intelligent, what we're really doing is enabling robots to become capable of not only understanding their environment, we call it perception. So the video and maybe the sensors that exist in a robot. to understand their world, but then to reason on what that world is telling them. What does that insight provide in terms of what's occurring in their application, and then take action based on that reasoning. So a sense, reason, and act flow is what we typically call physical AI. That's increased intelligence within the robot. So we spend a lot of time ensuring that the right technologies are available with the processor, you know, subsystems, our core ultra products are designed for that. And then that we're enabling the right kind of software to enable that insight to be derived right on the robot. But implicit in your question is an interesting thought. The industry is now starting to understand and maybe figure out architecturally, where should we place all of that functioning? So if you think about a robot needing to be increasingly intelligent, there is work we want to do at the point of action on the robot. And increasingly, physical AI capable robots will do that. But there's also support for the robot within the premise or within the factory floor or within the enterprise that can be served from a server local to the robot. We might call that an edge server. So you have a portioning of the work between the robot and the edge server. And then there will be some examples or some environments where some of the work is appropriate to run from the data center, which could be local or maybe remote from a cloud perspective. But what's really important as we think about where the work needs to be done is what are the criteria by which that work will be measured. Latency will be important. So ensuring that we're minimizing delay means you need to run the work closest to the point of action. a security. You want to keep data secure and close to where the work is happening as well. And so that will dictate what exists local to the robot or even local to the enterprise within the premise. And so it's more a function of how does the robot fit within the workflow and the existing activities that the robot is supporting. That's really what tends to underpin the decision criteria. What we believe, at least at Intel, is that we need to enable that choice so that the decision about where work occurs and how it's best optimized is facilitated across all of those domains, whether it's data center, on-premises edge, or on the robot itself, and ensure that the work we're doing in the hardware and in the enabling software enables that flexibility and choice to occur. That's super important.

Ryan Shrout: 

I love that explanation of physical AI and robotics. It's something that I think maybe doesn't get enough attention, like kind of the differentiation between the two and how they merge together. One of the things, you know, if you've traveled to any of these shows recently, you've seen robotics demonstrations left and right, right? I don't think I've been doing anything in the last, yeah, like two years, you know, they're just up walking around. And there's always two or three of them on every show floor I seem to go to now. But we all know that moving from a pilot experience or a pilot program out to an actual enterprise deployment of anything, whether that be compute infrastructure or robotics, introduces a lot of hurdles. I'm curious, that gap between pilot and deployment, what does that encompass from a robotics standpoint?

John Healy: 

Yeah, it's a good question. By the way, you're right. There's almost no show I go to now that doesn't have robotics as part of it because it's captured the imagination. But it's also, I think, pointing to the path forward as to how much value and opportunity robotics creates, almost irrespective of the industry type. But the point you're making is really significant. And we describe that as the deployment gap, you know, moving from the robot implementation that occurred inside in the laboratory or the lab or the constrained environment of a trial or a POC into actual real world. And it's hard. That's a difficult transition. And so one of the things we've found as we work through with many, many partners is it sort of comprises to two components. There's a technology challenge, there's friction in the technology because it's complex, and there's a cost challenge. We'll talk about both maybe a little bit. But on the technology side, really one of the big gaps in deployment is understanding how to continually refine and optimize and modify the deployment when it starts to interact with the real world. So what you may have conceived in a lab with an AI model and the data set that you based that model on, was constrained because it was based on the information you had at hand at the time. Then as you go through simulating of what that might look like in the real world, you make decisions about what is likely to occur. Then you learn what actually occurs when you move into deployment. That is complex for developers, software developers particularly, the implementation of a robot It's one of the hardest things they can go and deploy, because they have to think about the computing system, which is more classic PC server-like, and all of the other subsystems that make that robot real. Think of the actuators that are managing the arm, the manipulation of grip and complexity of the hand in the case of a humanoid. or navigating an unstructured environment for a mobile robot, like I talked about in logistics, that may encounter barriers or things it didn't expect as part of its deployment. You have to consider all of these, and how am I going to make sure the right subsystems work really well together to ensure that unknown can be adapted to and learned. And this is exactly what physical AI is intending to do. As it finds differences in the expectation, how does it adapt and modify for those? So that complexity is what we have to lower in terms of the barrier. And that's where we focused a lot of our software development efforts in Intel to enable our partners to have the right tools, the frameworks, and the software modules to build solutions that allow them to abstract that complexity and focus on the application category that they're trying to develop. So we've invested in things like the Physical AI Studio that allows a non AI specialists to fine tune the model based on the work they're actually trying to get done. That's super important if you're using, if you want to train a robot to emulate the work you do, we call it imitation learning, without knowing how do I go in and write all the code to make the model work? That's a lowering of the complexity. We've invested in an effort that we describe as the OpenVINO Physical AI Framework. That's to ensure that we're really making it easy to bring a trained model and deploy it with all of the other subsystems that make up the robot. So just two examples of lowering that barrier. But the second one is the complexity or the challenge of cost, the complexity of the cost. And a lot of times moving from initial proof of concept to scale deployment is a function of the cost equation. How do you make sure that the TCO, as we describe it, the total cost of ownership in deployment is one that you can sustain? For that, you need, at least we believe, a degree of openness by design and by definition in how you procure the different components and software elements and bring those together to create a robot solution. So if you're locked in to a fixed sort of vertical implementation, you're kind of at the mercy of it. If you have openness at each layer of your solution and the ability to work with different vendors in market and know that there's an interoperability because of the compliance with a known good framework, for the implementation of a robot solution. Now you've you've choice and choice brings down cost. And so we think about it on two vectors. And that, I think, helps to address the challenges, because these challenges are significant, but they're resolvable. And then you open up the advent of robots deploying in multiple different application categories across multiple different markets. And it really only becomes limited by the imagination as to what we can actually have them do.

David Nicholson: 

Yeah, I want to hear a little more about what development looks like in this environment. You know, we talk about legacy computing or traditional computing. We think about programming. Well, implicit in programming is the idea is there's a program, there's a descriptive instruction set. And now we're talking about physical AI, robotics, with greater and greater levels of intelligence on kind of a sliding scale, that's fine. But how does autonomy change the development process? And what do you mean by autonomy? when we're developing for these environments that include robotics?

John Healy: 

How does that change the game? It changes the game in how the solution is composed, David. So you're right. In a more classical development cycle, we have a known specification or functionalities we're trying to solve for. We define a set of parameters by which we would measure compliance. And then we define the architecture that meets that need. And that's the flow. You have a hardware need and a software need. And then you just do the development and test that you achieved what you believed you'd achieve. And then you deploy. The difference with physical AI is the autonomy. It's exactly what you described. It's that you do define a known initial expectation from an application perspective, but you build into that the reasoning engine to allow the robot or in this case, the robot, to make changes and adapt to the environment in deployment. So the classic flow, if I kind of high up level it a little bit, is that you start with a foundational model, which is not tuned for any one application, but is supportive of multiple different robot categories. And then you pre-tune that model with the data that's available to focus it a bit more on the application category that you're intending to deliver, but it's not fully optimized until you start to implement. And then you work through a cycle of fine-tuning, of optimization before you deploy, and then you deploy. So we think of it as training, simulate and fine-tune, then deploy. In deployment, the robot then needs to be capable of continually iterating based on what it's learning that changes and adapts its ability to respond to the environment. So instead of a robot maybe identifying a defect, we often see this in manufacturing, the robot's set up to do welding, but it has a vision component that's watching for the quality of the weld. In the more structured environment, it would say, I see the weld is damaged, flag an alarm, we need to make a change. In the adaptive world, it sees the weld and says, I've seen a problem, I need to make some change so that the next weld is better. And that might trigger a preventive maintenance cycle on its welding head, or it might trigger its need to change some of the parameters about how that welding head is performing so that the welds increasingly improve. The difference is that the robot takes action based on what it learned. That's the adaptability. And so the change in the production flow or in the development flow is ensuring that that optimization sequence or capability is built as part of the solution. So the way we think about it actually in Intel, is you need to be capable of training anywhere, simulating anywhere, but then deploying in the more optimal way on a flexible platform. And we believe that's the Intel platforms. But that flow is fundamentally different because it builds into it the ability for the solution, when it's deployed, to be adaptive after the fact. And that's what physical AI actually promises. That's what it provides.

Ryan Shrout: 

One of the things I always like to kind of get into, especially here at the end of an interview like this is something a little bit more forward looking. I think this conversation has been amazing. I think a lot of the challenges and complexity of enterprise robotics deployments are something that a lot of our viewers and listeners maybe didn't really wrap their head around, but From a enterprise intelligent robotics standpoint, these enterprises are going to continue to move from these pilots to deployments. But what should they be expecting and looking for on Intel's roadmap about where you're focused next and how that's going to help them with these deployments and improvements going forward?

John Healy: 

Yeah. So as I mentioned, and this is what you should expect from us, to continue to improve and continue to advance on the vectors that enable technical complexity to be reduced and cost optimization from a TCO perspective to be achieved in how robots will evolve. And so we focus on how do we ensure we're building the right, you know, hardware architectures to allow for the optimal placement of the different components of that application category. I talked about real time and deterministic control as being foundational to robots, both the traditional robots, but also the future robots. We think we support that best in our CPU and processing complex. If you think about the model distribution, we have integrated GPUs, the graphics processors, and integrated NPUs to ensure that we have the optimal acceleration in the hardware for the differing elements of the software application or the models that deploy. So not to get too technically deep, but that's really foundational. And we continue from a roadmap perspective to build on those capabilities so they continually improve and continually provide for the optimal performance, power, and price equation, typically how we think about it. And then from a software perspective, actually really key is what I mentioned in terms of the openness of the solution availability across the ecosystem. We spend a significant amount of time with a very wide number of partners or ecosystem players as we describe them on the hardware and software and application categories across the industry. to ensure that the right technology is in the hands of the players who build the solutions to support a diverse set of market categories. So, what do you expect from Intel? We continue to provide that underpinnings, the foundational software and hardware assets that will allow scale across multiple different market categories and unleash innovation in the hands of our partners. So then the consumers, the enterprises in this case, can benefit from what robotics actually really provides, which is an increasing path to automation that allows them to do their jobs more effectively and grows their own footprint in the industries that they service. We think we have a fundamental role to play to continue to accelerate that across the entire industry.

David Nicholson: 

Well, this has been a great conversation about deployment of robotics, physical AI within the enterprise. But I have a quick bonus question for each of you. And the possible answers are a given calendar year or never. So it's a fairly binary choice. And the question is, starting with you, Don, when will you have a humanoid robot living with you in your home?

John Healy: 

I think there's real potential, David, that maybe a decade from now, something of that order.

David Nicholson: 

So give me a year, John, give me a year.

John Healy: 

2035, 2036, I can see that being a real possibility.

David Nicholson: 

And Ryan, Ryan in your house. Wow.

Ryan Shrout: 

Well, I was, I was gonna, I was, I was, I was thinking like 20, 20, 30, I was earlier, 20, 31, 20, 32, you know, that early adopter tax that you have to pay. Right. Uh, although, although maybe it's a, it's a riskier endeavor than most early adoption, uh, uh, technologies.

David Nicholson: 

Both of you have shocked me. My answer is 2028, probably because I was deprived of having enough toys as a child. But all over the map, we have this recorded, so we'll see what happens. And if I do get mine first, you guys are welcome to come over and play with it. within reason. John, I want to thank you for joining us for this Connected Intelligent Edge Spotlight. To our viewers, don't forget to hit subscribe, follow us on social media, and check out all of our Six Five Summit content at sixfivemedia.com slash summit. For Ryan Shrout, I'm Dave Nicholson. See you next time.

‍

‍