Avathon CEO Pervinder Johar discusses AI-driven autonomous operations and $100B market potential
Summary
Avathon CEO Pervinder Johar describes the company as a physical AI platform extending infrastructure life in asset-heavy industries like energy and manufacturing. He argues that industrial AI is shifting from deterministic robotics to cognitive algorithms that reason, citing a $100 trillion aging infrastructure market. Johar claims customers see 10x to 100x ROI through predictive maintenance and agentic planning. He highlights partnerships with Saudi Aramco and addresses technician shortages in aerospace and defense. The company, which raised over $400 million, aims for a $10B-$100B market cap by focusing on proprietary industrial models rather than generalist LLMs.
Key takeaways
Episode sections
- 0:19 – 4:01
Introduction and Company Overview
Woo Jin Ho introduces Avathon as an AI platform for autonomous operations in asset-heavy industries. Pervinder Johar defines it as a physical AI platform focused on extending infrastructure life and achieving autonomy.
- 4:01 – 7:08
Shift from Deterministic to Cognitive Operations
Johar explains the shift from deterministic robotics to cognitive algorithms that reason, predicting 'lights out' factories managed by a central operating system.
- 7:08 – 9:15
Knowledge Platform vs. Data Problem
Johar argues companies face a 'knowledge problem' rather than a data problem. Avathon’s knowledge layer converts data into context for cognitive agents.
- 9:15 – 14:08
Mining Customer Case Study
A mining customer uses Avathon for exploration, safety, and predictive maintenance, predicting failures and prescribing actions to increase uptime.
- 14:08 – 16:36
ROI and Platform Modularity
Customers report 10x to 100x ROI. The platform is modular, including a knowledge layer, cognitive platform with 150+ patents, and specific applications.
- 16:36 – 20:52
Agentic AI and Continuous Planning
Agentic AI enables continuous planning with thousands of active scenarios, allowing instantaneous switching based on real-time conditions.
- 20:52 – 23:44
Safety, Guardrails, and Security
Avathon uses a knowledge base as a boundary for agent behavior and a control system layer to enforce guardrails in hazardous environments.
- 23:44 – 34:47
Target Markets and Competitive Moat
Key markets include energy (Aramco partnership), aerospace/defense, and semiconductors. Johar cites 13 years of experience and proprietary models as a moat.
- 34:47 – 35:13
Company Scale and Financials
Avathon has over 200 employees and raised over $400 million. Johar states employee count is no longer the primary metric of scale due to AI usage.
- 35:13 – 36:49
Revenue and Market Cap Potential
Johar suggests the addressable market could reach $1 billion in revenue, with Avathon aiming for a $10B-$100B market cap to become a 'decacon'.
- 36:49 – 37:56
Strategic Partnerships and Industry Focus
Johar highlights partnerships in energy, aerospace, defense, and mining, predicting major announcements from large global companies soon.
- 37:56 – 39:56
Physical AI and Human Analogy
Johar compares physical AI to human capabilities like sensory perception and continuous learning, suggesting humans are essentially physical AI systems.
Claims worth verifying
There is about $100 trillion worth of aging infrastructure in the world.
Pervinder Johar ·2:55
How to verify: Check World Bank or McKinsey Global Institute reports on global infrastructure stock and valuation. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon has more than 150 patents and 200 plus models in its cognitive platform.
Pervinder Johar ·15:52
How to verify: Search USPTO or EPO databases for assignee 'Avathon' or 'Avathon AI'; check company technical whitepapers. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Customers see between 10x to 100x ROI on investment.
Pervinder Johar ·14:56
How to verify: Request case studies or ROI calculators from Avathon; verify against customer earnings calls if public. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Primary sourceAvathon announced a partnership with Aramco Digital.
Unidentified speaker (UNKNOWN) ·23:56
How to verify: Check press releases from Aramco Digital or Avathon on their official websites or news wires. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
There is a shortage of about 750,000 aircraft technicians in the US.
Pervinder Johar ·24:35
How to verify: Check Bureau of Labor Statistics (BLS) data on aircraft mechanics or industry reports from Boeing/Airbus on workforce shortages. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
The US spends about a trillion dollar a year on defense.
Pervinder Johar ·25:15
How to verify: Check US Department of Defense budget requests or Congressional Budget Office (CBO) reports. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
More than $100 billion is spent on logistics and more than $100 billion on maintenance in defense.
Pervinder Johar ·25:15
How to verify: Check DoD budget breakdowns for Logistics Civil Augmentation Program (LOGCAP) and Maintenance, Repair, and Overhaul (MRO) contracts. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Fleet readiness hangs between 55% to 65%.
Unidentified speaker (UNKNOWN) ·25:33
How to verify: Check GAO reports on military readiness or DoD annual readiness reports. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon has been around since 2013.
Pervinder Johar ·33:20
How to verify: Check company registration documents, Crunchbase, or LinkedIn company page for founding date. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Industrial problems rarely require large language models with 10 billion parameters.
Pervinder Johar ·30:44
How to verify: Technical analysis of industrial AI model sizes vs. general LLMs; verify via technical papers on edge AI/industrial AI model efficiency. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon has over 200 employees.
Pervinder Johar ·34:47
How to verify: Check Avathon's official website 'About' page, LinkedIn company profile, or recent press releases for employee count. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon became a unicorn a couple of years back.
Pervinder Johar ·35:08
How to verify: Search Crunchbase or PitchBook for Avathon's valuation history and funding rounds to identify when it crossed $1B valuation. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon has raised over $400 million in investment.
Pervinder Johar ·35:13
How to verify: Review Avathon's press releases for funding announcements or check SEC Form D filings if US-based, or Crunchbase/PitchBook for total capital raised. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Primary sourceAvathon's potential revenue could reach $1 billion or more.
Pervinder Johar ·35:33
How to verify: This is a forward-looking claim; verify by tracking Avathon's actual revenue growth in future earnings reports or press releases against this target. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
The industrial AI space could be a trillion-dollar space over 10-15 years.
Unidentified speaker (UNKNOWN) ·35:43
How to verify: Compare against market research reports from firms like Gartner, IDC, or McKinsey on the projected size of the industrial AI or automation market. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon aims for a $10 billion to $100 billion market cap.
Pervinder Johar ·36:03
How to verify: This is a forward-looking claim; verify by monitoring Avathon's valuation in subsequent funding rounds if it remains private, or market cap if it goes public. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon has partners in energy, aerospace, defense, and mining.
Unidentified speaker (UNKNOWN) ·37:13
How to verify: Check Avathon's press releases for partnership announcements or customer case studies in these sectors. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Saudi Aramco is a partner of Avathon.
Unidentified speaker (UNKNOWN) ·37:13
How to verify: Search for joint press releases or news articles confirming a partnership between Avathon and Saudi Aramco. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Major partnership announcements are expected in the next several weeks.
Pervinder Johar ·37:33
How to verify: Monitor Avathon's official newsroom or press releases for the specific timeframe mentioned (weeks following Nov 3, 2026). For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Avathon is a private company.
Pervinder Johar ·35:08
How to verify: Check if Avathon is listed on any stock exchange; if not, it is private. Confirm via SEC EDGAR (if US) or company website status. For recent events, verify 2026-08-03 through 2026-11-03. Use another period only when the speaker explicitly names it.
Primary sourceDo your own research
- Verify Avathon's valuation history and funding rounds via Crunchbase or PitchBook to confirm the 'unicorn' status and $400M raised.
- Check Saudi Aramco Digital's public partnership announcements or press releases to validate the stated collaboration.
- Research the US Bureau of Labor Statistics data on aircraft technician shortages to verify the 750,000 figure.
- Review Avathon's patent filings (USPTO) to confirm the claim of over 150 patents in industrial cognitive models.
- Analyze independent case studies or white papers from mining customers to validate the 10x-100x ROI claims.
- Compare Avathon's proprietary models against generalist LLM providers (e.g., NVIDIA, Microsoft) for specific industrial use cases.
- Investigate the definition and current market size of 'physical AI' to assess the $1 trillion market forecast.
- Check defense budget reports (e.g., DoD annual reports) to verify the $100 billion logistics and maintenance spending figure.
Full transcriptexpand
- 0:19
- Woo Jin Ho: Hello and welcome to the Tech Disruptus Disruptions Podcast hosted by Bloomberg Intelligence. In this podcast series, we speak with sea C-level level company executives. and management teams about their views on disruption and how it's driving their decision-making strategy. Now, uh, my name is Wujin Ho. Uh, I'm with Bloomberg Intelligence, and it's uh which is Bloomberg's Research Arm. And our group uh
- 0:39
- Woo Jin Ho: covers um two thousand companies globally across multiple asset classes backed by Bloomberg and third-party data, supported by 500 research uh professionals. And uh as we continue this. uh podcast series, you know, I I like to find new companies uh that are trying to do things that are disruptive, especially with AI. And today we're happy to have
- 0:59
- Woo Jin Ho: uh Avaton CEO, Pravinda Johar. Pravinda, welcome to the show.
- 1:03
- Pervinder Johar: Thank you. Thank you for having me on the show.
- 1:05
- Woo Jin Ho: Perfect. Now, uh before we jump into our discussion, uh, I want to give a little background on Avaton and why it matters. Now, in my younger days, uh, I actually covered the uh the supply chain and logistics companies. And one of the things I learned quickly is how complex these operations can be. Industrial companies manage vast networks of equipment
- 1:25
- Woo Jin Ho: facilities and supply chains, often built on aging infrastructure and fragmented systems. Now, that makes it harder uh to spot problems early, avoid downtime, and improve efficiency. Now, Avaton is using AI to help companies better understand how those operations are performing, identify issues
- 1:45
- Woo Jin Ho: sooner, and make better operating decisions across industries such as energy, manufacturing, logistics, and aerospace. I think that should be a decent high-level You know, um, overview of the company, but I'm sure you can could tell the story better. So, you know, Provender, Proven, for for someone who who's hearing the company for the first time,
- 2:05
- Woo Jin Ho: how do you describe the the company and what it does and what problems you're trying to solve?
- 2:09
- Pervinder Johar: Think of Avaton. We are a physical AI platform,
- 2:15
- Pervinder Johar: which is focused on how do you get to autonomous operations. And I think if you talked about supply chain, logistics, and others. When we talk about kind of operating operations, we actually focus on three three parts of the uptop operations. One being how do you run
- 2:35
- Pervinder Johar: a large, complex, asset-heavy kind of site, whether it's a mining site, whether it's an oil rig, whether it is a uh for renewable energy, if you look at battery storage, so all of them is that how do you if eventually get to autonomy because what you're dealing with is a lot of physical
- 2:55
- Pervinder Johar: asset, which is what the physical AI is uh is related to in operating this uh this infrastructure. Just to give you a context on that side, um uh there is about hundred trillion dollar worth of infrastructure in the world which is aging.
- 3:12
- Unidentified speaker (UNKNOWN): And so
- 3:13
- Pervinder Johar: think of kind of our ships and planes. and power grids and all other things that each and every one of us relies on. So it's it's aging. So we are applying kind of uh AI to both extend the life of that infrastructure. And then the other two sides kind of that we focus on, as you mentioned, which is supply
- 3:33
- Pervinder Johar: chain and manufacturing, we are focused on how do we make the new infrastructure. And so
- 3:39
- Unidentified speaker (UNKNOWN): if you
- 3:39
- Pervinder Johar: think of it kind of like manufacturing, uh we want manufacturing to be autonomous, uh too. Uh so that's autonomy for for manufacturing and same thing as autonomy for supply chain. So think of us as kind of a critical infrastructure AI company. So we've been calling it physical AI, No, but it is all around that kind of the the
- 3:59
- Pervinder Johar: infrastructure that runs the world.
- 4:01
- Woo Jin Ho: Got it. So all right. No, when I will when I used to cover supply chain logistics, right? Um
- 4:07
- Unidentified speaker (UNKNOWN): no,
- 4:08
- Woo Jin Ho: and also I used to cover some of the semiconductor companies as well. Uh there is some sort of communications and robotics. So if I think about uh industrial companies, right, they've actually used automation, machine learning tools and predictive analytics uh for years.
- 4:25
- Unidentified speaker (UNKNOWN): Um
- 4:25
- Woo Jin Ho: I'm I'm curious what has changed uh that makes industrial AI uh one an an opportunity for you and and it seems like uh you know uh an opportunity that uh your customers would be interested in. What's different uh between what happened over the last t decade to what you foresee as an opportunity for the
- 4:45
- Woo Jin Ho: next decade?
- 4:46
- Pervinder Johar: Absolutely. I think uh the the one I uh think of it as a continuum. wordsage uh uh words a abrupt change when it comes to robotics and control.
- 4:57
- Pervinder Johar: Like but see, robotics and control
- 5:01
- Pervinder Johar: were designed to be very deterministic. Like kind of it was back to that kind of everything we have done for the last two, three decades kind of has been very deterministic, control, controlled, deterministic algorithms uh too. And now we're dealing with cognitive algorithms. Like so the difference between a robot and a
- 5:21
- Pervinder Johar: humanoid actually is just this that you're going from kind of a robotic uh by by definition, kind of when you say the word robotic, kind of what goes through our head is that it's a robot. It does a certain Little set of things kind of in there. When we talk about humanoid, kind of now you're talking about that it actually have cognitive abilities to
- 5:41
- Pervinder Johar: think in there. Now, more than that, but what Avaton is focused on, that think of a factory of the future which is lights out. And future might not be that far away. It might be a year or two away where you have only humanoids, robots, and equipment running a manufacturing site.
- 6:00
- Pervinder Johar: But you need a control system or an operating system which can go across all of these kind of intelligent beings, right? So it's no longer kind of that uh they're not robots, they're kind of cognitive machines which are working on doing doing production in this particular case. That what's the control system? What is the
- 6:20
- Pervinder Johar: operating platform, which is the collective intelligence and giving directions to number of these these humanoids? So it's a very kind of uh in the simplistic term, you're going from kind of uh it being deterministic to it being cognitive kind of in there. But but uh holistically actually it's a major, major disruption
- 6:40
- Pervinder Johar: on how we think of running kind of either our our operations or All manufacturing new uh new assets.
- 6:48
- Woo Jin Ho: Got it. So so i i if I think about the uh industrial robotics space today, uh and and and you talk about uh cognitive versus deterministic. I mean, you have a human intervention to help the industrial uh robots, right? But then does that mean that some of these um with
- 7:08
- Woo Jin Ho: your operating system and your layer, will that allow the cognitive nature of it? So it is self-learning. If it is, there are some changes into the uh let's just say the conveyor belt, it will learn and help make uh deliver more efficiencies?
- 7:23
- Pervinder Johar: No, absolutely. And I think but if I just take a step back, right when we talk about um cognitive abilities, there are two things that go together, one being knowledge and then our ability to reason over that knowledge.
- 7:39
- Pervinder Johar: So very, very first thing we focus on that kind of for the company,
- 7:43
- Pervinder Johar: we are the knowledge platform first. We have our own proprietary technology, which is that how do you go capture knowledge? So if you're a human and you start working for a company, you kind of know what the company does, right? So all that knowledge goes into person's head, but someone is teaching you on an onboarding, you're reading it now. think of it that
- 8:03
- Pervinder Johar: you are an agent or a humanoids and you start working for a company. How do you get the knowledge or the context of what that company is? So what we are focused on first is that kind of if you want to go do reasoning, you have to define kind of what is the knowledge as a base to begin with. So what we do is it's kind of and we we say it kind
- 8:23
- Pervinder Johar: of um fairly often that companies don't have a data problem. Data is abundance, they have a knowledge problem, right? And then how do I go convert data into knowledge? Once that knowledge is there now, now you have kind of whether it is um uh uh cognitive agents which run kind
- 8:43
- Pervinder Johar: of only in um
- 8:45
- Unidentified speaker (UNKNOWN): uh
- 8:45
- Pervinder Johar: only in the compute space or humanoids uh which can be doing kind of both the physical. work as well as doing doing the reasoning, both of them now get kind of this ability to reason over kind of this uh this knowledge base.
- 9:00
- Unidentified speaker (UNKNOWN): I
- 9:00
- Woo Jin Ho: got it. Okay. So help us frame this with with with one of your customer examples, right? I I know you have a a a fairly extensive customer base and some big name ones as well. So so if you could walk us through a customer example, what problem are they trying
- 9:14
- Unidentified speaker (UNKNOWN): to do?
- 9:15
- Woo Jin Ho: solve and what has Avaton deployed and the type of impact it's had?
- 9:20
- Pervinder Johar: Absolutely. So I will give you an example in um mining. So if you think of mining as one of those industries which is um has been relevant in our history of humankind kind of for forever, but it's kind of even more kind of relevant now given the shortage of critical raw. earth materials. So
- 9:40
- Pervinder Johar: if you think of a mining site, typically has many different functions.
- 9:45
- Pervinder Johar: So you when when you're starting off, you start off with something called an exploration, right? Where you're looking at kind of the seismic data. So one of our customers, we worked on that. How do you look at kind of the data, which might be just the drilling data or the sound data to apply AI techniques, generating generative AI techniques to now tell
- 10:05
- Pervinder Johar: you whether to explore for a particular kind of material. And same thing we do on energy. It can be for natural gas, it can be for any other material that you're extracting. There's a lot of AI applications on just doing that exploration work. But more critically, kind of we have a customer in North America which is now um
- 10:25
- Pervinder Johar: working on how do you apply it to three different areas, one being something as simple as health and safety. And so there are cameras, like there are cameras now uh all all around. We connect to all of these cameras to start predicting uh kind of will there be a safety incident which is likely going to go
- 10:45
- Pervinder Johar: happen, right? So you're preventing kind of these safety incidents before they happen. So again it brings it bringing kind of that physical
- 10:53
- Pervinder Johar: physical vision side, which is coming through cameras with the cognitive abilities. Second area has been that mining uses a lot of assets. So there are autonomous trucks, there are kind of like if you think of it as heavy equipment, you also deal with these roads, which are just roads made for the mine. right? So So they're not kind of your high
- 11:13
- Pervinder Johar: wage that you're dealing with. right? So you the people don't think of a road as an asset, but it's again an another asset kind of on that side that you need to do the do the mining. So this customer we are applying what we call a cycle of predict, diagnose, and prescribe, meaning that we are trying to predict failures
- 11:33
- Pervinder Johar: before they happen.
- 11:34
- Unidentified speaker (UNKNOWN): And
- 11:34
- Pervinder Johar: so so so we sit on top of a lot of IoT data, sensor data and you're trying to predict a failure before it happens. Because of course, if you can predict it, now I can maintain it better. We're doing diagnostics. And so failures, we may or may not be able to always predict it because predictions are never 100% correct in there.
- 11:55
- Pervinder Johar: But failures will happen. When failures happen, how do I go manage the knowledge of that? How will I go repair that of prediction? particular particular equipment? And think of it like going to a mechanic, right? So if mechanic has never really seen a particular type of a fault, it takes them a lot of trial and error. What we are doing is that we're doing the knowledge management
- 12:15
- Pervinder Johar: to say that we have seen this before, this is how you should solve it, or applying cognitive ability that how do I go solve it? Down to prescription, all that results in that my uptime goes up. Like so if I'm a mine, if I can kind of operate it 5% more, 10% more, kind of there's a significant business ROI
- 12:35
- Pervinder Johar: related to kind of that equipment uptime. But it doesn't stop there. You also now have a supply chain that if a part fails and I do not have it on hand, like it is going to take me longer. Then because I will have to order it. So you start. getting into inventory management now. So how do I go know what
- 12:55
- Pervinder Johar: inventory do I need to look at? And if you think from that perspective, it's interesting that uh that um when we look at a failure for a particular equipment, it's very focused on that equipment. But when you look at say something Is as a mine site or all mine sites combined, you're looking at failure across all
- 13:15
- Pervinder Johar: the equipment. So you don't know kind of which equipment will fail. You know that one of them is going to fail, is how you're doing supply chain planning. So it's a different application of AI. And then going all the way down to actually automating the procurement process, the sourcing process. So that's one example just on mining, that you're going from
- 13:34
- Pervinder Johar: exploration, health and safety, how do I keep the mine up and running? How do I run the supply chain for it? And they all come together into a single platform through Avaton.
- 13:47
- Woo Jin Ho: Okay. Got it. All right. So so there's a lot there, right? A lot that uh unpacked. Are are these all separate modules or are these one uh one big module that uh one what big application package uh for the customer uh to to accept? and if we think about each of these different um buckets
- 14:08
- Woo Jin Ho: whether it's uh it sounds like there's some computer vision element to it but from from uh from a health and safety aspect of it there seems to be a a maintenance aspect and then a supply chain aspect of it right um how do we think about well the ROI for for computer vision and health and safety is you know unmet unmeasurable, of
- 14:27
- Woo Jin Ho: course, right? But if we think about the other aspect, you know, how are you thinking about the magnitude of the ROI or how the customers think about the magnitude of the ROI there?
- 14:36
- Pervinder Johar: I think the ROI itself tends to be between somewhere around 10x to 100x investment, uh uh a hundred X of what they invest in. Like because you're not talking dealing with kind of incremental improvements. You're you're dealing with um you're dealing with kind of transformations now, uh kind of
- 14:56
- Pervinder Johar: Bujin. Kind of we are back to that think of it mine of the future, a factory of the future. Kind of every area we work on, it's almost that kind of it's a journey that how do you get to a full autonomy? And once you get to full autonomy, there is a lot of kind of um uh uh lot of potential uh on that side. But typically
- 15:16
- Pervinder Johar: our customers see between 10x to 100x ROI on investment. Now, to your answer your question, is it a one module, multiple module? Think of it as three different layers for the platform, one being the uh the knowledge platform.
- 15:32
- Unidentified speaker (UNKNOWN): See
- 15:32
- Pervinder Johar: the knowledge platform you can apply it to the knowledge for an entire mine, because a mine might have 10. 15, 20 hundred sites, right? As a mining company. Like so you can be you can apply it to only one site kind of in there. So you have modularity from the perspective of how much knowledge you want to store. Then middle is what we call a
- 15:52
- Pervinder Johar: cognitive platform. So company today has more than 150 patents, 200 plus kind of our models, which are all part of this cognitive platform that. they're using. And then there are applications. So the things you describe, like health and safety, supply chain, maintenance, there are applications. And you can
- 16:12
- Pervinder Johar: uh buy or implement only one of those applications, all of those applications, so it's modular in nature in terms of how you go about the journey.
- 16:21
- Woo Jin Ho: So from what I'm hearing, right, you know, AI is is helping to drive well. beyond predictive maintenance, right? Uh operational streamlining, supply chain streamlining, help with the decision making.
- 16:36
- Unidentified speaker (UNKNOWN): Um uh,
- 16:37
- Woo Jin Ho: you know, for for some of your customers, has it also found ways to um I I I I guess uh improve operations and change operations in a way, right? Because that that change that that changes the scope of what you guys are actually offering, right? Because they're probably buying it for one thing and they're
- 16:57
- Woo Jin Ho: discovering something new.
- 16:58
- Pervinder Johar: Yeah. Yep. Absolutely. Absolutely. Right. So I will give you kind of an example on maybe the manufacturing side that how your entire operations can change when when you combine the physical AI along with kind of some of the cognitive capabilities on
- 17:17
- Pervinder Johar: agentic AI too. So if I'm running a factory, say it's for automotive, it can be for the dimensional semiconductor, it can be any of those. When you're running the factory, your job is to maximize the utilization for the factory, right? Which means that you tend to do planning to say that this is what I'm going to
- 17:36
- Pervinder Johar: manufacture today, tomorrow, day after.
- 17:39
- Pervinder Johar: And typically, kind of people running operations, they tend to pick a plan. As humans, we love picking a plan. We're back to that. Okay, what's the plan? What's my weekdaily plan? What's my
- 17:49
- Unidentified speaker (UNKNOWN): weekly plan?
- 17:50
- Pervinder Johar: What's my yearly plan? Or so. And the reality is that all plans are kind of only one possible plan, because as soon as you put it in place, something will happen,
- 18:01
- Unidentified speaker (UNKNOWN): which
- 18:01
- Pervinder Johar: will cause you to change your your plans. So most companies run today on that they will have one plan and then on a weekly or a monthly basis, kind of they will adjust that particular plan. What we ended up discovering is that kind of when you start doing planning with agente AI now, versus kind of on verse age
- 18:21
- Pervinder Johar: humans, there is no limitation on how many plants I can have active. So instead of one plan, plant, I can have a thousand plants and I can switch from one plant to another plan instantaneously.
- 18:34
- Pervinder Johar: Typically, what it takes kind of for people to do planning for manufacturing is that there will be a meeting for kind of sales and meeting between supply chain and manufacturing to go say what we should do. So you're doing doing all of these decisions are human decisions which tend to kind of take time. not because of our ability to make decisions, but the
- 18:54
- Pervinder Johar: number of humans that are involved in making that decision actually kind of causes causes causes delays.
- 19:01
- Unidentified speaker (UNKNOWN): So
- 19:01
- Pervinder Johar: now all of a sudden it is that do I really need to run on one plan, or can I have a continuous planning? But in reality, it's not continuous planning. You have already done many plans, thinking of all possible kind of outcomes. Now, of course, you don't want to do infinite number of plans, but that changes your operations completely. Kind
- 19:21
- Pervinder Johar: of something that looks uh chaotic, uh kind of in um uh in a manufacturing environment, right? All of a sudden it looks like it's normal now for me to switch from one plant to the other. And then I think everyone's everyone life has gotten so much better when they've moved to this kind of uh continuous. planning.
- 19:39
- Unidentified speaker (UNKNOWN): Yeah,
- 19:40
- Woo Jin Ho: it's it's pretty interesting. Like, like in the world of agentic, it means, you know, to me, it just means persistent data flow and and and persistent discovery and and and planning. How do how do you narrow it down so that there isn't data and information overload on on the human aspect of it?
- 19:57
- Pervinder Johar: I think uh and that's can been part of it, is that um uh the this uh notion of human in the loop or human being aware of what what you're doing on that side, it all comes down to making it kind of uh making it simple to ex understand the reasoning behind it. So so on the
- 20:17
- Pervinder Johar: data overload, that's why we talk about it kind of a data versus knowledge. The good part about knowledge is knowledge can be stored in our brain kind of so much. more effectively than data. So we might not remember every number, but the knowledge part is that we know kind of how to do something, right? So we try to distill it
- 20:37
- Pervinder Johar: down to kind of that knowledge. And explaining it back to the human in the loop in the context of the knowledge of what is going on is kind of what the trick is, so that they don't get overloaded by what's going on on the agency. agent TKI.
- 20:52
- Woo Jin Ho: Right. Okay. Now now similar to that, you know, any anytime we think about and and I'm gonna joke here.
- 20:59
- Pervinder Johar: Yeah. Uh
- 21:00
- Woo Jin Ho: robots having cognitive knowledge, right? We we we think about Terminator.
- 21:09
- Woo Jin Ho: Right. So so
- 21:11
- Unidentified speaker (UNKNOWN): uh
- 21:11
- Woo Jin Ho: how do you get cust customers? Well number one our are customers currently comfortable with giving uh the robots more uh control uh with with AI systems and and you know, um especially when it comes to like safety, reliability and uptime uh when it's critical and and and uh security control uh
- 21:31
- Woo Jin Ho: concern security concerns, that's been in the news lately. Just curious how how your approach to it and how customers are thinking about it.
- 21:38
- Pervinder Johar: I think so. I think that's where when you are in industrial setting, you know, I will go back um
- 21:46
- Pervinder Johar: to my very first job, which was building distributed control systems for nuclear power plants, kind of long
- 21:53
- Unidentified speaker (UNKNOWN): time
- 21:53
- Pervinder Johar: back. So if you talk about the level of risk, kind of that was there. It it almost was that my first manager told me that if you have one bug or an issue with your in software. you can cause so much pain that you have to be very careful about what you do. And industrial settings are no different today. So you're dealing with kind of hazardous
- 22:12
- Pervinder Johar: environments in there to begin with. Go drill at the wrong place, kind of in a in a uh LNG site or somewhere else, it can have kind of a lot of lot of issues to begin with. Which is where can kind if of we build the built plan. platform the way we did. Right. When I talked about that knowledge platform, that knowledge becomes
- 22:32
- Pervinder Johar: the context in which we allow humanoids and robots to operate in. So there are guardrails to begin with, that then you're not going to be able to go do kind of anything which is kind of outside Side of the knowledge base. And then the second part that I talked about that which is the control. system or the operating system
- 22:52
- Pervinder Johar: which sits on top, which is constantly monitoring what is being done by either kind of the uh agents
- 23:01
- Pervinder Johar: or by uh by the equipment or by the robots. Like so, so you have two layers of protection. Almost the first one is that kind of we are describing that this is the this is the ontology, and this is the knowledge that you want. to operate in. And sitting on top is a control system which is monitoring kind of and giving guardrails of
- 23:21
- Pervinder Johar: what any individual component is doing.
- 23:24
- Woo Jin Ho: Okay. I I want to talk about you you gave me a mining example. Uh you know, look, supply chain is universal supply chain and and industrial is is universal. It's not mining only. Just just curious about the the industrial area. AI opportunity overall. Uh mining is one example. Where are you seeing the strongest
- 23:44
- Woo Jin Ho: demand? Where do you think you can take this uh going forward?
- 23:47
- Pervinder Johar: Yeah, absolutely. So I think as we mentioned early in the early on kind of in there our focus areas are uh energy.
- 23:56
- Unidentified speaker (UNKNOWN): So
- 23:56
- Pervinder Johar: there's a tremendous opportunity in energy and we have a number of customers in um in energy. I think uh recently we announced our uh partnership with um uh Aramco Digital. So so so if you think of it as one of the largest kind of energy or oil and gas company in the world. So so uh tremendous
- 24:15
- Pervinder Johar: opportunity kind of there and think of it as kind of a uh something which will be in um tens of billions of dollars of ROI that is possible kind of in energy sector. if not getting to a kind of a trillion dollar impact across all industrials. We are focused on aerospace and defense. So if you think of
- 24:35
- Pervinder Johar: aerospace and defense, uh there's a shortage of about 750,000 aircraft technicians today. And so if you think of all our planes, and I think if you think uh on that one, the reason we go it together, aerospace and defense. because defense has a large number of fighter jets, right? Which is
- 24:55
- Pervinder Johar: again kind of we talk about fleet readiness. So we see a huge impact on fleet readiness. Just taking kind of uh US as an example, kind of so we're spending about a trillion dollar a year on defense, so for Department of War, and going up on that side. If you peel it down, there is kind
- 25:15
- Pervinder Johar: of significant spend on logistics. So think of like more than 100 billion on logistics, more than 100 billion on maintenance. And our fleet readiness, depending on which fleet we talk about, is hangs between 55% to 60, 65%, right? So which is meaning
- 25:33
- Unidentified speaker (UNKNOWN): that
- 25:33
- Pervinder Johar: even though we have this large fleet of equipment, whether it's spider jets, whether it's tankers and others, they need to be maintained and they need to be ready. So huge ROI we see kind of on
- 25:46
- Pervinder Johar: that side to be able to go do it well. The third segment we are focused on is kind of high tech. And high tech again has been the focused on manufacturing and supply chain operations. right? And I think all of us are kind of very familiar with the uh with the how dependent we have become on semiconductors, right?
- 26:06
- Pervinder Johar: And if you think of that semiconductor lifecycle, it starts off with the critical raw material, then to a kind of an equipment manufacturers for the semiconductor fab and the fabs themselves. So there's a tremendous amount of ROI on that side, and then there is automotive. And so I think if you think of these segments,
- 26:26
- Pervinder Johar: if you combine them all, kind of the market is such that this overall spend on physical or industrial AI can eventually reach to kind of hundreds of billions a year, if not a trillion dollar a year.
- 26:40
- Woo Jin Ho: Got it. So so let's let's talk about deployments, right? Because you you're you're coming across multiple types of of industries, right? So when when when you first meet with one of your customers, uh, what is usually the bigger problem that you're trying to solve? Is it the technology itself, the available
- 27:00
- Woo Jin Ho: data, uh or or the way the or uh the operation is organized and structured? Because it seems like you know, your customers have very extensive and complex supply chains, right? So you're you're trying
- 27:11
- Unidentified speaker (UNKNOWN): to
- 27:12
- Woo Jin Ho: organize and optimize multiple types of buckets so your system, uh your cognitive AI operating system can help um solve that uh and organize it in a in a way that uh a customer can digest it easily.
- 27:25
- Unidentified speaker (UNKNOWN): I
- 27:26
- Pervinder Johar: think if you think of the customers, there are two initial set of what I call education that we have to go through. When we engage kind of any of the customers, some of the customers are very sophisticated but typically you run into explaining the difference between data and knowledge, and
- 27:45
- Pervinder Johar: why do they need a knowledge platform? Because a lot of the people want to jump into the use cases on AI right away. But reality is that you have to also think of the context that these agents will have and the knowledge platform. So explaining the difference between data. and knowledge usually is half half of our
- 28:05
- Pervinder Johar: job. And the second one is that we work in the industries which are traditionally been left behind kind of from technology perspective. So they these are not the fastest moving industries outside of high-tech Tech that I use as an example, in terms of the technology
- 28:25
- Pervinder Johar: itself. And then the best way for us to get through it is just by kind of showing versus continuing to talk, talk about it, right? So typical engagement would be that we call it kind of more of a conference room pilot type of a situation where we will go and simulate kind of what the world would look like if you
- 28:45
- Pervinder Johar: if you are on this uh this particular platform. But this whole notion of people understanding. what is knowledge. I think some other companies call it context now that we have for for AI, we have to give context to the agents too. Uh usually it's kind of the first part of the journey. Um, because people are used to data. They're
- 29:05
- Pervinder Johar: used to how do I store data, how do I look at data, but converting it into knowledge is it just something something new.
- 29:12
- Woo Jin Ho: Okay. So uh we laid the framework there. What are the first initial responses by your customers once they start implementing the system? And how do they react in six to twelve months' time? You know, does it get better as they start using the the the uh avathon uh more extensively?
- 29:32
- Pervinder Johar: And I think our typical implementations are that we are live with the first proof point in between three to six months, right? So by the time you get to a year, you have already seen the benefits kind of for uh uh within the first year. And uh and our normal reaction has been uh twofold. One, kind
- 29:52
- Pervinder Johar: of the system does get better as you start using it more because it's learning and it's learning on an ongoing basis.
- 30:00
- Unidentified speaker (UNKNOWN): Uh
- 30:00
- Pervinder Johar: but this problem that kind of uh I shouldn't say problem, but um uh what what happens then that kind of people start thinking of so many different use cases that a year later the challenge is that how do you narrow it down to go focus on a fewer things because there's a there's so much excitement about what is possible and what what
- 30:20
- Pervinder Johar: what our customers have already seen.
- 30:22
- Woo Jin Ho: Got it. All right, so so um
- 30:25
- Unidentified speaker (UNKNOWN): do
- 30:25
- Woo Jin Ho: you use your own LLM? or or do you reach out to do you use uh a third-party foundation? Foundation
- 30:31
- Unidentified speaker (UNKNOWN): models
- 30:31
- Woo Jin Ho: is already out there.
- 30:33
- Pervinder Johar: So for us, kind of, I think we have a lot of small language models which are all of our own one. Because I think if you think of industrial problems, right
- 30:43
- Unidentified speaker (UNKNOWN): uh
- 30:44
- Pervinder Johar: the size is such that it's very hard for any of the industrial problems to get to even 10 billion parameters. Like so, you're not dealing with that, it needs large language models. Uh, it does need small language models, which are our own. The second part of it to keep in mind that these languages are
- 31:03
- Pervinder Johar: different, uh, Bujin. They say language of geophysics, it's a language of physics, it's a language of time series data. So, so when we talk about the language models, they're not necessarily English.
- 31:15
- Pervinder Johar: or any of the kind of the spoken languages. You really have to be geophysics available and physics away for our small language models
- 31:26
- Pervinder Johar: to begin with. So in majority of the cases, kind of you're going to deal with that, there are own proprietary models. On LLMs, we rely on the open weight models. right? So we have taken kind of multiple. LLMs and we have customized them kind of for industrial use, right? So it's not kind of a generic
- 31:45
- Pervinder Johar: version of a large language model. It's actually trained on industrial data
- 31:53
- Pervinder Johar: to begin with. But that's kind of our approach. That kind of a lot of small language models, uh but but uh as I mentioned, keep in mind it's the language of physics and geo. geophysics and uh sensor data not necessarily kind of a spoken
- 32:08
- Woo Jin Ho: got it and and and then and then and then from a um competitive standpoint right and and we we get this all the time i'm i'm just curious in terms of as these models as these foundation models get better right and as uh and more open weight models start coming out
- 32:26
- Unidentified speaker (UNKNOWN): um you
- 32:27
- Woo Jin Ho: know does it
- 32:29
- Unidentified speaker (UNKNOWN): you
- 32:29
- Woo Jin Ho: know
- 32:29
- Unidentified speaker (UNKNOWN): do
- 32:30
- Woo Jin Ho: does your competitive mode shrink or is it your expertise that helps keep that uh competitive moat an arm's length uh from uh uh competition going forward?
- 32:40
- Pervinder Johar: I mean I think it's um that we don't really compete with kind of your traditional LLMs kind of today, right? Kind of and I think you uh our competition is more with companies which focus in this kind of space. Um, I think uh uh when it comes to defense, kind of it's all companies like Palantir kind of that we are
- 33:00
- Pervinder Johar: competing with today, uh, too, uh, which are all kind of uh whether it's for them or us, we are relying on kind of our expertise and models which are specific uh to the industries that kind of we um uh we we operate in. If anything, I see kind of our competitive mode
- 33:20
- Pervinder Johar: increasing because kind of as you do more and more of these deployments, um company has been around since 2013. So it's been
- 33:28
- Unidentified speaker (UNKNOWN): about
- 33:28
- Pervinder Johar: 13 years of us kind of working on it, right? So it's not very easy uh for um uh for for a new entrant that can to have the same level of experience or expertise that we have gotten uh kind of uh too uh but um uh but the part which is kind of critical about
- 33:48
- Pervinder Johar: that 13 years is that we have actually had enough failures to know what not to do. What's what
- 33:56
- Unidentified speaker (UNKNOWN): to do.
- 33:57
- Pervinder Johar: Right. And I think that's kind of the usually people forget that, but but we we've have had enough failures in the physical space to know what doesn't work.
- 34:05
- Unidentified speaker (UNKNOWN): Not
- 34:05
- Pervinder Johar: only kind of what does work. Uh and that gives us a competitive mood, which which is what it really means in the end for a customer that we are a safer bet and our time to value is much more quicker than someone who is starting now or trying to do this work now.
- 34:23
- Woo Jin Ho: Got it. So so uh I'm about to wrap. Uh I'm um I'm this is a financial podcast, and I'm just curious, and I'm sure my audience would be curious to know um how should we think about uh Avaton scale today, you know, from from a a customer's employees or revenue basis to the degree that you're comfortable, or any
- 34:44
- Woo Jin Ho: other metric that you want to share uh to the broader audience?
- 34:47
- Pervinder Johar: So I think for us kind of the company edge has uh over 200 employees, but we're not going to be a company which will ever become hundreds of thousands of employees. As you know, kind of internally we use AI to do as much of the work as possible. Uh so these days it's not about the employee count.
- 35:08
- Pervinder Johar: Uh we had become a unicorn kind of a couple of years back. Uh
- 35:13
- Pervinder Johar: we have had uh uh I think over four hundred million dollars or so invested in the company uh so far. Uh too. We think kind of the potential for this space is that um uh that um as I mentioned before, uh kind of that there is no reason why it can't get to kind of billion of revenue or more
- 35:33
- Pervinder Johar: kind of over uh over time. Uh and I think that might be underestimating the number kind of on the revenue because it's a it's going to be a trillion dollar space over
- 35:43
- Unidentified speaker (UNKNOWN): the
- 35:43
- Pervinder Johar: next kind of 10 years 15 years and there are not that many people kind of in the space right so even if you think of kind of the normal AI there is a lot of companies which are competing in it when you think of physical and industrial AI kind of is the number of companies are or handful of companies going after this particular particular
- 36:03
- Pervinder Johar: space. But it is potentially kind of a from from investor perspective, it's potentially a company which can have 10 billion, 100 billion or more of a market cap over uh over time, which is what what what what I'm focused on. Because the billion was nice to be in there, but kind of focus is now that how do you get to
- 36:23
- Pervinder Johar: get to be a decacon and kind of hopefully hopefully connect. continue to go from from from there.
- 36:29
- Unidentified speaker (UNKNOWN): All
- 36:29
- Woo Jin Ho: right. So so this actually goes leads into the visioning, right? Because you're talking about like a trillion dollar space. You know, the the sector has to evolve. And how do you think about uh how autonomous uh industrial operations uh will be over the next five years uh for the market opportunity to grow?
- 36:49
- Woo Jin Ho: And um, I mean, you mentioned robotics.
- 36:51
- Woo Jin Ho: right? Becoming the um uh the operating system possibly for humanoids as well in the interaction. So how how does the space evolve uh as my final question uh over the next uh uh five years for you?
- 37:03
- Pervinder Johar: I think if you look at it for us, I've been focused on evolving with our partners. Like so
- 37:10
- Unidentified speaker (UNKNOWN): kind of
- 37:10
- Pervinder Johar: I mentioned Saudi Aramco as an example, but it's
- 37:13
- Unidentified speaker (UNKNOWN): kind
- 37:13
- Pervinder Johar: of one in energy space. But we have partners in aerospace and defense. We have partners in mining. We have part which are all large names, kind of as as you can imagine. And kind of I think uh people will see kind of announcements coming in with their names over the next several weeks or so that who these partners
- 37:33
- Pervinder Johar: are. And because it's going to be led by kind of some of the largest companies in the world, kind of pushing on the innovation in this space, then than small to medium businesses, kind of uh just given the uh given the size of the space and what what it really means.
- 37:49
- Woo Jin Ho: Big big visions. Um so I'm gonna wrap up here. I'm just curious uh if you have any uh parting words to the audience.
- 37:56
- Pervinder Johar: I think uh my my only parting words are going to be that kind of uh there are a lot of buzzwords budge words around. which which people have to kind of sift through, right? Even in our conversation today, you have talked about physical AI, industrial AI, agentic AI, knowledge, and others. Uh to some extent
- 38:16
- Pervinder Johar: it can be overwhelming sometimes, but at the same time, when you push step back and think of what we are talking about is nothing different than how we as humans work. We as humans, we have our five senses. We can see, we can hear, we can smell, we can touch. When we are talking about physical AI, you're trying to do the same
- 38:36
- Pervinder Johar: thing. As humans, we always have knowledge, right? So we know kind of something at a given point in time, but we are constantly learning. And we are we're learning not only by reading, but also by seeing and hearing and kind of everything. everything else. So, as complex as all these things sound, kind of we
- 38:56
- Pervinder Johar: as humans have been doing it for a very, very long time. Like we are all physical AI systems, which have been kind of performing these things for a long time. So I just wanted to leave it with that, that it's no different than kind of what we have done since our how uh
- 39:10
- Woo Jin Ho: that that I will tell you that that is true, but but you are solving a very very complex problem by acting adding that cognitive nature uh
- 39:20
- Unidentified speaker (UNKNOWN): to it. So that
- 39:21
- Woo Jin Ho: that you're making it a lot sound a lot simpler than than than it actually is, uh Provinda. All right. Uh Pravinder, uh thank you for joining us and sharing your thoughts uh on you know uh in industrial AI and and and cognitive uh uh uh robotics. And thank you every everyone uh for joining us. We have a great lineup
- 39:41
- Woo Jin Ho: of future disruptors like Avaton uh and provender. So hit the subscribe button to keep us up to date, to keep up to date with the Tech Disruptors podcast and not miss an episode. I'm Wu Jin Ho. Uh thank you for listening to Tech Disruptors and we'll see you next time.
More from Tech Disruptors
Honeywell CTO on AI and Physical-World Autonomy
Episode discussion: Honeywell CTO outlines AI-driven autonomy strategy, citing edge computing and deterministic guardrails for industrial applications
Speakers’ claims: Mustafa Okur interviews Suresh Venkatarayalu, CTO and President of Forge at Honeywell, regarding the company's pivot from automation to autonomy. Venkatarayalu explains that Honeywell has simplified its portfolio to focus on commercial buildings and industrial process automation, leveraging its Forge platform to integrate AI with deterministic control systems. He describes a three-phase AI strategy involving internal knowledge models, partnerships with Nvidia for open-source models in secure hardware, and hybrid architectures for data sovereignty. The discussion highlights use cases in reducing false alarms in ethylene plants and building management systems, as well as collaborations with Google, LS Electric, and Dangote to expand ontologies and meet cybersecurity mandates.
Figma on Why Design Gains Value in the AI Era
Episode discussion: Kris Rasmussen on Figma's AI Strategy: Design Context as a Moat and Hybrid Pricing Models
Speakers’ claims: Kris Rasmussen, Chief Architect at Figma, discusses how AI agents serve as interface glue rather than replacements for traditional UIs, emphasizing that design systems provide critical context for better AI outputs. He argues that as code becomes commoditized, value shifts to design, with Figma leveraging proprietary user data to fine-tune open-weight models for specific workflows. Rasmussen highlights bottlenecks in AI adoption due to security and backend expertise gaps, noting that human curation remains essential. He also outlines Figma's infrastructure strategy, which avoids building foundation models from scratch, and predicts a shift toward hybrid seat-based and consumption-based pricing models.