Subscribe
Share
Share
Embed
Deep conversations with the founders, investors, and operators building real-world AI - robotics, automation, industrial systems & AI infrastructure. Past the headlines, into how these technologies are really built, deployed, and scaled. Hosted by Bogdan Cristei, venture partner and former systems engineer.
THE OPTIM UPDATE
Labor as a Cloud Service: One Operator, 100 Machines | Christoffer Jørgenvåg, Hive
Guest: Christoffer Jørgenvåg, Co-Founder & CEO, Hive
Host: Bogdan Cristei
---
Bogdan Cristei (00:00)
This is The OPTIM Update. I'm Bogdan Cristei, and today I'm talking with Christoffer Jørgenvåg, Co-Founder and CEO of Hive. Hive builds what they call a silicon brain for industrial machines - a sensor kit and an AI model installed on the wheel loaders, excavators, and forklifts a company already owns, so the operator moves from the cab to a control room, supervises several machines at once, and the machines take on more of the work themselves with every shift. Today Hive is running on forklifts moving pallets around the clock for a global logistics company, in avalanche zones on one of Norway's most exposed mountain passes, and in a tunnel with the operator sitting 90 kilometers away. They are expanding into the UK and the US. Without further ado, here's my conversation with Christoffer.
Bogdan Cristei (00:51)
Christoffer, so good to see you, man. It's been a while since we talked. So happy to do this today.
Christoffer Jørgenvåg (00:56)
Thank you, and likewise. So nice to see you again, Bogdan.
Bogdan Cristei (00:58)
Awesome. Let's jump right in, because we've been having these conversations for a while. But before anything else: why the name Hive, and what does "Built for the moon, proven on Earth" really mean?
Christoffer Jørgenvåg (01:11)
Those are two excellent questions. Let's start with the name. Hive came from the swarm: wanting to create something collective, a collective intelligence where many different agents could solve tasks together. Hence the name Hive.
"Built for the moon, proven on Earth" is more of a nudge toward our ambitions. As we'll probably talk about later, we are working on solving the world's dependence on physical labor - how expensive it is and how little it scales. The cost structure on Earth is one thing; maybe it costs 40 bucks an hour to get something done. If you go to space, that is a whole different league of how costly it is. So all the training, everything we do, makes a ton of sense on Earth, but it makes much more sense in orbit and on other planets. That's why it's part of how we present ourselves.
Bogdan Cristei (02:05)
Before we go into the vision, let's start with the past. You started Red Rock at 23, from an idea you had in a classroom. You grew it to over 80 people and sold it to Ocean Infinity in 2021. What did those twelve years of selling lifting systems to offshore and marine customers teach you about how industrial companies buy and use new technology?
Christoffer Jørgenvåg (02:29)
Actually, I never wanted to build machines, and I ended up building many hundreds of them. In the beginning I built AI systems. That's how I started Red Rock - building apps, payment systems, mobile ticketing systems, completely different things. I had taken an introductory course on neural nets at Michigan Tech and spent a considerable amount of time thinking about them, and the vision was: let's build a brain, a smart control system, and sell it to the machine manufacturers, the OEMs. And nobody wanted to buy that.
Bogdan Cristei (03:04)
It was before its time.
Christoffer Jørgenvåg (03:07)
That's right. So that meant I had to build the machines. Almost out of spite, I started hiring electrical engineers, mechanical engineers, hydraulics people, all of that, and started building machines - with a 180,000 US dollar bank loan and a lot of sleepless nights.
To answer your question: in that process I set up a factory in Norway, welded steel plate-box construction in Poland, multiple bases in Poland and Romania, built complete products in Brazil and India, and delivered to China, Japan, the US, Canada, all of Europe - basically all over the world. Simple things in the beginning, then more and more advanced, more and more robotic, more and more semi-autonomous as we went on.
It gave me an understanding of how important everything is, from how a cable is terminated to the quality of the steel you use and the paint systems - the whole business of building machines, and of understanding machine dynamics and how a machine works. It also gave me an appreciation of how extremely hard it is to build good machines, and how little innovation there actually is on the machine side of existing machines. They have been perfected over tens or hundreds of years. In my opinion, that is not where the real innovation lies. It's in different embodiments and at different levels of the system.
Bogdan Cristei (04:28)
You have to be very courageous to go into hardware in the first place.
Christoffer Jørgenvåg (04:33)
And it's super cool. Building the best crane in the world is, in my opinion, very, very cool. But it doesn't really change the world. If you want to make a meaningful impact on humankind, which is the purpose of Hive, then you need to be in a different place in that stack and think a bit differently.
Bogdan Cristei (04:55)
And then a year after the sale, you were starting from scratch again. What did you see in 2022 that told you autonomy for heavy machines was ready, or about to be ready? And why were you the one to even attempt to build it?
Christoffer Jørgenvåg (05:12)
There were a few different things. After selling Red Rock, I always thought the dream was to retire early and just do my own projects. I very quickly realized that was absolutely not the case. All I wanted to do was build meaningful things. At the same time, I'm realistic about how painful it is to be an entrepreneur. At Red Rock, I don't know how many nights I spent without sleep, either because of things that had to get done or because we had no cash at all and I was trying to figure out how to work with the cash flow. It's super rewarding, but it's also very painful. So for me, doing something new from scratch had to be super meaningful. It had to be world-changing - something where, when I look back in ten, fifteen, twenty years, I can say: okay, I did that, and it changed the world for the better. That was one thing.
The other, maybe more boring, part is what I saw at Ocean Infinity: how insanely cool that company was, what they did, and how impactful it was for very big companies - these big energy companies and big logistics companies really wanted to buy these kinds of services. So it was also a realization that the timing was right. The timing was not right when I tried the first time, but now it is.
And I never actually wanted to build machine automation. To me, that's not it. It is not an automation game, and it's not about heavy machinery. It is about building operators. I remember so well, when I started my first company, I started by building servers, which is also super cool, but it takes time. Then Amazon Web Services and Google and all of those guys came around and scaled it with something super simple. I think that is happening in this space. I don't think it's about selling automation projects, and I don't think it's about selling specialized machines. I think it is about selling scalable labor, packaged the way Amazon did it for computers. That, I think, is the game, and that's what I'm building.
Bogdan Cristei (07:07)
Let's talk a little bit about that. Maybe you can explain to everyone what a wheel loader is, and then take me through a full shift for a wheel loader operator at a given plant. What does that job take out of the person, and what does it cost the company if that person is not around?
Christoffer Jørgenvåg (07:26)
The right US term would probably be front-end loader. I struggle with that myself sometimes, because for some reason in Europe it's called a wheel loader, but it's the same machine. It's a heavy, wheeled machine with a bucket that carries a load - gravel, different kinds of rock, different kinds of industrial bulk material - and scoops it from one place and moves it to another. A typical operation is in a quarry: there's blasted rock, maybe crushed, and you use the front-end loader to lift that rock and load a truck, or move it to a different place on the plant, or re-crush it. Load, drive, dump it somewhere. On these job sites and in industrial applications, that operation is continuous and semi-repeatable.
The cost when people are not there - and this is one of the things I realized talking to countless operators - is a few things. One is the human cost of having somebody there: the wear and tear on their body. At almost all the sites where I get to spend time with operators, which is very important to me so that I understand what they think and how we can make jobs better for people, I see that everybody has some kind of wear-and-tear problem with their body. When it comes to the cost to the company, what we also see is that highly skilled operators are very difficult to find. Quite a lot of plants are running with one less shift than they wanted. Of course that is massively costly, because you have an asset that could be used more, and a full plant behind it. It's hard to put a general number on the actual cost, but it's for sure not low.
Bogdan Cristei (09:04)
Let's talk about the machines. What goes onto the machine? What sits in the control room? What happens in between while the machine is working? And I know you're installing these on 20-ton machines in a few days. What does the install involve, and how do you keep the safety systems intact when you install your own hardware?
Christoffer Jørgenvåg (09:25)
These machines can be small - the smallest are maybe a few tons - and the biggest are maybe a hundred tons, hopefully more in the future. So there can be some massively big machines. What we install on the machine is a set of sensors - radar, lidar, GPS, cameras - plus communication and compute. It's not that complicated, and a typical installation takes a day or two once we have pre-integrated with that machine.
What's cool about it is that what we sell is not the kit. We are selling the hours. We're selling labor. What the company buys from us is not a kit that we install for them; they buy an operator from us, a virtual operator. That also means there is communication on board, Wi-Fi or 5G, and we always have an operator at the back end. If something doesn't work - let's say our model can't perform - there is always an operator who can take over control and fix the problem. That's something the customer doesn't really see. We have operators to bridge the gap between current model capabilities and the real world, and of course that's also how we train our systems.
Bogdan Cristei (10:42)
So would it be correct to think that you're building a silicon brain that generalizes across loaders, forklifts, excavators, all these types of machines, and it works by itself most of the time, but once in a while, if it gets into an unknown state, it calls an operator, the operator puts it back into a known state, and it continues autonomous operation? Is that the right way to think about it? And how many machines per operator are you at today, and where do you want to be?
Christoffer Jørgenvåg (11:13)
That's the right way to think about it. My personal goal is to drive the cost of doing physical work down significantly and the flexibility up significantly. That means the ratio in 2029 should be one operator to 100 machines, minimum. If it's not one to 100, I cannot deliver labor cheap enough to make it super available and super abundant. The goal with Hive is to make sure you can solve any physical problem. If you want to move a mountain, the labor component of moving that mountain should not be what prevents you from doing it.
Right now, the most important thing for us is not the ratio of operators to machines. It is generating very, very unique training data, because these models need good training data, diverse training data, and a fairly high volume of it to generalize. The idea is to do all of that, build better and better models, and then hopefully be at one to ten next year and one to 100 in 2029. That's the goal.
Bogdan Cristei (12:21)
Got it. One thing I always think about, since I spend some time on these sites myself: the internet is always spotty and bandwidth is always an issue. What happens if the 5G link drops with the operator 90 kilometers away and nobody in the vehicle? Do you have a plan for that?
Christoffer Jørgenvåg (12:41)
All our systems are designed to return to a safe state. We do not allow any machine operation without human supervision. We think the future of this is a higher degree of autonomy, yes, but always with a human in the loop. There will always be a human who can take over control and who is the decision maker. I don't think we want machines doing everything by themselves; we still want human involvement. So let's say the 5G drops and you don't have any redundancy - no Starlink, no dedicated radio link. Then the machine will brake, stop, go to a safe state, and wait for the connection to come back.
Bogdan Cristei (13:20)
Interesting. And then somebody drives out there and fixes the problem.
Christoffer Jørgenvåg (13:25)
Hopefully it reconnects automatically and starts again. Or eventually a human can walk out and fix it.
Bogdan Cristei (13:32)
Let's not get into that. Anyway, there are three ways to get an autonomous excavator today. The OEM builds it, like Caterpillar. A startup builds something new from scratch, which I don't know is a good idea, but I've seen some of them do it. Or you retrofit what is already in the fleet. In your mind, why does the retrofit win? And are there places where it loses?
Christoffer Jørgenvåg (14:01)
There are always highly specialized scenarios where building a machine from scratch could be better. It's almost always cooler. It's always tempting to build cool machines. But it's almost never the right answer. These machines have been perfected over such a long time. The best modification would probably be to take away the driver comforts - the cabin, the seat, the joysticks - which would save quite a lot of cost on some of these machines. But I see no real reason to redesign them. They work well as they are.
For us, retrofitting is the winning strategy - and of course this might eventually come from the factory line, with our solution built in by some OEMs. Right now you can install it, you can use all the assets the companies we work with already have, you can start operating immediately, there is no massive engineering cost, you generate all the data you need, and you deliver value right now. To me it's a no-brainer. And again, this is not about making the best autonomous excavator. It's about making the absolute best generalized operator that can work on any machine. The machine is just the extension of that operator's body. That's what we're building. I do not think customers want one specialized excavator. They want a full fleet of operators running everything they have.
Bogdan Cristei (15:24)
You're already starting to answer this, but let's look at everybody else in the market. You have Bedrock, which raised hundreds of millions of dollars. TerraFirma raised, I think, a hundred million. Where are you in the competitive marketplace? What are you doing that's different?
Christoffer Jørgenvåg (15:42)
I think those are excellent companies. They're doing a lot of things right, and I admire a lot of what they do. I do think we're building slightly different things. Of course, I don't sit inside those companies, and they would need to answer that themselves. But looking from the outside in, and with the knowledge I have about how this should be built: we are, as much as possible, making generalizable physical laborers - a workforce that can sit in any machine, whether that's a forklift, a crane, an excavator, or whatever the customer has. I think that is the winning strategy, because you want extremely diverse data sets, and you want to be able to deploy to every asset the customer has, so that you are perceived as the provider of all the physical labor a customer might need across all their sites. You need to get there to win this game. This is a completely new category. And again, it's not machine automation. It is labor delivered as a cloud service.
Bogdan Cristei (16:39)
We keep going back to that. You describe what you sell as work hours rather than kits. With that model, how do customers react to paying per productive machine hour? And what has to be true to get from, let's say, 20 bucks an hour today to maybe five dollars in a year or two, to maybe one dollar in the long term?
Christoffer Jørgenvåg (17:02)
A few things need to be true to get there. There is some cost-down on the sensor package and the compute - probably some custom PCBs, which we're working on, and maybe some custom silicon as well, an inference chip, to get all the way down. But most of the gain will come from model improvements: one operator operating more and more machines, and the models becoming more and more advanced.
Bogdan Cristei (17:32)
That makes sense. Let's go back to the OEMs. Today, are they your channel, your competitor, or both? And where do you think they will see you in a few years?
Christoffer Jørgenvåg (17:44)
We are working with a lot of OEMs, at various stages, but it's a bit difficult to disclose too much about what's going on. To answer the question: I think some of them will see us as competition. Some will see us as a service they don't provide and can't provide, which they can tap into, and which really benefits their customers and themselves. And over time, I think most of them will see us as a partner. That's my gut feeling.
Bogdan Cristei (18:09)
You started in Norway, then you moved the headquarters to London, and now you're working on a US expansion. Why London first?
Christoffer Jørgenvåg (18:19)
London first was for a few different reasons. The most important thing for us, next to solving global physical labor and making it extremely abundant, is attracting very, very good talent - because that's what motivates me, working with the best people, the people who do amazing things. Coming into the office and saying, "Oh, what is happening?" - that's the feeling I always chase, and it's so rewarding. To some extent, you need to be where that talent is. In Europe, for AI talent - and we're building our own models and our own policies from scratch - London is a very, very good hub. We have been able to attract some amazing talent there. That's partially why we're going to the US as well, to support customers of course, but also to make the talent pool as big as it can possibly be, so we can access the top minds, the people who are really, really motivated to solve this problem and who see that it needs to be solved for society to reach its next stage. We have some amazing people who have joined, and more on the way in.
Bogdan Cristei (19:30)
That's very exciting. Congrats. I did hear some rumors about you stealing some people. I won't say names, but good job. What does the US entry look like? Which industries, what regions, and what has to be adjusted about the product or the sales motion for an American job site or warehouse, compared to one in Norway?
Christoffer Jørgenvåg (19:57)
We have Norwegian customers, UK customers, quite a lot of other European customers, and now also some very, very large corporations looking at massive deployments with us, which is very cool in the European market. All of them are chasing the same things: margins, being part of the future, labor scarcity - they're struggling to get people - and safety. In the US, we are working with customers in logistics, in mining, and very closely on a few energy cases. We are setting up our office now. I'm going back to Austin next week to build up the team and make sure we have all the components and machines in stock to service our customers.
I don't think the difference between Europe and the US is that big. It's a lot of the same things we need to solve. But when I talk to customers, I feel the labor shortage is a lot stronger in the US. I have so many customers who want to add another shift and simply can't. Which is very sad. This is something that really needs to be solved, and it's very motivating to see that you can help with it.
Bogdan Cristei (21:18)
To help with these kinds of things, you've been hiring a lot of folks internationally. You have people in Europe, in London, and you're going to hire AI engineers in Austin, San Francisco. What's your pitch to that breed - the San Francisco engineer, the Austin engineer?
Christoffer Jørgenvåg (21:41)
I hope it is exactly the same pitch, and the future will tell us whether it works. This is maybe a little personal, but I started coding when I was ten or eleven, in the macro editor of Excel, because I didn't have internet and I didn't have a compiler. That's what I had. After working with that, and then learning C and C++ and working with game engines, you get this feeling that you can build anything. Whatever you want to put into that virtual world, you can build. That is not true in the real world. There are so many constraints, and a lot of them come from labor being very, very expensive - one person's time is very expensive. If we can make one person's time a hundred times more productive, we can solve all of that, build a society of extreme abundance, and provide the tools for all the innovation that's going to happen. Similar to what drove the internet wave and the tech wave, this will all happen with physical things.
If you want to be at the forefront of that, we are the company. We know, because all of our customers benchmark us against competitors, that we are the best and we have the best tech stack. If you want to be part of that, come to us. Solve this for the better of humankind, and let's build this amazing society of abundance. Hopefully we will find the best people who want to do that.
Bogdan Cristei (23:19)
It's really interesting what you're saying. I often have this conversation with folks: why physical AI, why now? My answer is always: look at any business since the beginning of time. Why was it interesting? It saved somebody time, or it saved somebody's health. That's literally it. The washing machine, electricity, anything - you have more time at the end of the day and you're much more productive.
Christoffer Jørgenvåg (23:46)
And that's the basis of any economic growth: being more productive.
Bogdan Cristei (23:51)
And with physical AI, if you look at 3D printers two years ago, maybe 30 percent of what you tried to print came out well. Now it's just easy, at least for me. I do think we're at this inflection point where the technology is becoming good enough to start increasing productivity.
Christoffer Jørgenvåg (24:16)
That's right. We see it with our systems too. We are moving meaningful volumes of stuff for our customers - moving atoms - at a productivity level similar to a human. I agree with you, I think we are at that inflection point. There will be a lot of things that need to be fixed, because doing things in the physical world is hard. But I agree. I think we're there.
Bogdan Cristei (24:39)
Let's get back to you. What do you need in the next 30, 60, 90 days? Customers, hires, partners, introductions - how can people listening help?
Christoffer Jørgenvåg (24:50)
This is one of the strange things. Since I was very young, I've always been building things, and I've been running a company for many, many years, basically since wrapping up school. The key constraint has always been customer access. I've always worked in quite tough businesses, and it was always constrained by cash and by customer access. Now demand is just insane. Companies are paying for priority. We need to take customer access seriously, and we are working super hard to land the right customers, but that is not the limiting factor for me. Cash - of course we want the absolute best investors, we want to be very strategic about it, and we will need to run more rounds. But to solve this - and I call it solving physical labor, getting things done, moving and rearranging atoms - we need the best minds. So if I have one ask for the next 30, 60, 90 days, it's: bring me the best people who want to solve this, and we'll be super happy.
Bogdan Cristei (25:55)
That's amazing. All right, let's move into some hot takes. Without overthinking too much, I'll ask two or three questions and we'll end with that. First question: will we ever see humanoids on a construction site?
Christoffer Jørgenvåg (26:13)
Yes.
Bogdan Cristei (26:14)
When?
Christoffer Jørgenvåg (26:15)
Two to three years, and our silicon brain will power some of them.
Bogdan Cristei (26:18)
Really? Okay. Interesting.
Christoffer Jørgenvåg (26:21)
And I'll give you one more hot take: in five to ten years, there will be no humans on those construction sites at all.
Bogdan Cristei (26:27)
In five to ten.
Christoffer Jørgenvåg (26:28)
Yes. Because if you really want to challenge how things are done right now, you take all the people out, because then you can run your sites with a completely different operating profile. It has to happen.
Bogdan Cristei (26:39)
Especially in construction - such a fragmented industry.
Christoffer Jørgenvåg (26:44)
Exactly.
Bogdan Cristei (26:45)
Very interesting. That's why 3D printing of buildings had its try, for that reason. It didn't quite work; you still needed the people. Okay. What is the most wrong thing people believe about autonomy for heavy machinery?
Christoffer Jørgenvåg (27:03)
A lot of people talk about some kind of automation project and call it autonomy. To me that's the wrong thing, and it's what people get wrong the most. Even a lot of investors.
Bogdan Cristei (27:13)
Fair enough. I guess I already know the answer to this, but five years from now, what share of heavy machine hours in Europe and in the US are supervised by someone who is not sitting in the cab?
Christoffer Jørgenvåg (27:25)
Five years from now, 25 percent.
Bogdan Cristei (27:27)
25 percent. Okay, I agree. Last question. What's one thing you wish you had known before starting Hive?
Christoffer Jørgenvåg (27:35)
I wish I had known how to package this commercially from the beginning.
Bogdan Cristei (27:40)
Interesting. What makes you say that?
Christoffer Jørgenvåg (27:43)
Because I completely messed up how to sell what we sell.
Bogdan Cristei (27:48)
Interesting.
Christoffer Jørgenvåg (27:49)
And it was never about the tech. It was always about selling a service to our clients.
Bogdan Cristei (27:55)
I guess that would be your advice to founders starting now.
Christoffer Jørgenvåg (27:57)
Yes. That's right.
Bogdan Cristei (28:02)
Some very expensive learnings for you guys. Very good. All right, so with that - Christoffer, where can people learn more about Hive? Website, socials, where should people go?
Christoffer Jørgenvåg (28:15)
Add me or follow me on LinkedIn, and our website should be a good place to start.
Bogdan Cristei (28:24)
Awesome. Great to have you on, man. Thank you so much for your time today. Good to catch up.
Christoffer Jørgenvåg (28:27)
Likewise, so good to catch up. Thank you, Bogdan. Appreciate it. Have a good one.