AI is coming for cement, steel and glass production

0:01Mike Butcher:Hello, welcome to Path Founders with me, Mike Butcher. We like to unpack the code, the capital and the consequences of the startup ecosystem.

0:12Mike Butcher:Plants, not the garden variety, make cement, glass, steel and chemicals, and they are very complex, volatile and expensive to run. And they currently run on software built for a very much another era. Could AI help? Well, today I'm joined by Josh Vernon, who's CEO of Gigaton, the company formerly known as Carbon Ray, which builds autonomous control software for some of the world's most energy intensive industries. Gigaton is starting with cement, working with major producers, but its ambition is to replace the ageing control systems inside industrial plants and move heavy industry towards a future of AI driven, self-learning, autonomous operations.

1:01Mike Butcher:Josh, the old carbon ray story was AI for industrial decarbonisation, but it looks like the new gigaton story sounds more like replacing control software inside plants. So what changed? Mike, it is a pleasure to be here today. Nothing has changed in terms of our long-term mission. So long-term mission remains the exact same. End objective of the company is to reduce gigatons of CO2 emissions from heavy industrial manufacturing. We learned that the way we were trying to do that over the past couple of years was theoretically correct, but ultimately naive. And what I mean by that was that we were trying to put artificial intelligence on top of existing control systems, trying to own the intelligence stack above the thing that actually controlled the actuators in a plant and spent years banging our heads against the wall trying to understand how to get those systems to work effectively.

2:00We ultimately came to the conclusion that if you want a future which is fully autonomous plants, which in 10 or 20 years or industrial your processes will be fully autonomous. You cannot build that future on heuristic-based or rules-based logic that were written by people in the 1980s. So Gigaton is, I would say, an evolution of that mission and an increase in our ambition to go after a stack of the technology that needs to be rebuilt for that future.

2:32Mike Butcher:It sounds actually quite similar to how the new neobank movement operated. They decided they couldn't rewrite the software of existing banks and had to start all over again. Why do you think it took you so long to come to that conclusion? If only I had the answer to that, we would have made the decision earlier. I think we had a huge amount of respect for what the industry had built previously. It's not easy controlling one of these industrial manufacturing plants. You're talking about one of the most complex chemical processes on earth you're talking about 1500 degrees celsius 8 million ton production per year completely non-linear environments and to go in and off the bat think that you are going to control one of the largest manufacturing sites on earth the very actuator level is very ambitious um i think we thought our value as ai scientists was to sit a layer above that and we thought that was also the right go-to-market strategy from a sales perspective.

3:35You could use the networks of the incumbent control providers, add the AI on top of it and go. Ultimately, it just works out that that isn't as effective from a technology point of view.

3:46Mike Butcher:So how is it that you're able to supplant the existing software or are you doing something else? We supplant the existing software. So we'll, we've got a piece of hardware that will sit in our customers' IT or operations technology environment inside their plant and we will have models that run on the edge and those models will connect directly to the plant and will take autonomous control of the plant to drive them in the direction that's optimal for cost and carbon. Well okay so it sounds like quite a big job but part of the reason we're talking today is that you've raised 26 million dollars to fully autonomise these industries.

4:31Mike Butcher:You've raised from companies including let me see, Plural. Funding Round was led by Plural Series A. You've also got 2150 which is a VC based in the UK. Planet A Ventures, a lot of European VCs here. What sort of storytelling did you do in order to get to that point? Storytelling. So maybe I will share the story with you. You can judge the storytelling off the back of that. But the storytelling really is that the upheaval that our industrial plants have faced over the last decade is more than they have ever faced. You used to have these industrial plants which had a single fuel type coming in, which was predominantly coal and a single raw material, which let's take cement, for example, predominantly limestone.

5:22And you put that heat or that energy source and the material into a big kiln and that kiln would get to temperatures where there's a chemical reaction and that limestone turns into a different material which is clinker pretty much cement. What happened over the last 10 years is that it became economically unsustainable to use coal as your primary energy source all costs about 150 euros per ton and it's very bad for the environment, the amount of CO2 it emits. And it is much, much cheaper, or actually economically beneficial to use recycled waste material as your fuel source. So that could be human feces, it could be shredded tires, sewage sludge, it could be bone material, could be sesame husks, anything that the economy produces, which they don't use, can be put into an industrial plant and used as an energy source.

6:17But once you start putting that energy source into the environment or into the process, the complexity of that process rises very rapidly. You can imagine a process that was made to burn coal, now it's burning sewerage, just burns completely differently and the end product is different. So in that environment, you used to have these fixed, those rules-based control systems that we were talking about, trying to understand the complexity coming in and completely failing to do so in many cases. So what we do is we go in and we build simulations of that environment and predict what's going to happen regardless of what you put into the front end of the plant.

6:56So you can put in as much variability as you want, we'll predict what's going to happen and control the plant so that you get the exact same output regardless of what comes in. And AI is perfectly placed to do that. And customers today do that, run our system in autonomous control and you can save up to 3 million euros per plant per year just by switching out those fuel types and running a system like this.

7:21Mike Butcher:I see. So obviously you convinced your investors that this was the best approach, but you're also starting with cement. So what was the decision making behind that? So cement, we started by looking at cement, steel and glass. And the end objective is to build a generalizable platform. This is a problem that exists across all industrial manufacturing environments. Cement is interesting because it's the most used product on earth after water, so it is ubiquitous. It is also extremely closely correlated with GDP growth. So when we're talking about national sovereignty and the ability for an economy to grow, their ability to produce cement for roads, schools, hospitals is critical.

8:04So ubiquitous, very closely tied to the economy's growth, but most importantly produces one of the largest amounts of CO2 on earth. So 8 % of all CO2 emissions comes from 3000 odd plants scattered all over the world. And that is a pressing issue, not just on the carbon side, but on the cost side that needs to be solved. So there is a, there's a very real burning platform when we go to speak to our customers about what are you going to do to substitute the inputs to the process out to remain competitive. It was the biggest burning platform.

8:40Mike Butcher:I see, right. So you're going for the most ubiquitous, but you're also competing with some very big companies, Rockwell, Aspen Tech, Honeywell, ABB. Is the 26 million to give you the war chest to compete with these companies, or is it for something else? And also, how do you feel you're going to be able to go up against these very established companies? I think those companies have done a wonderful thing historically. I mean, they built the industrial world and still also control a huge amount of the hardware that exists in the plants. But the software that some of them, or the software that incumbents in general have built, aren't fit for what the plants look like today.

9:27If I were to run you through the control logic of some of their software, it is very similar to what you would imagine Excel to be, where it's if then that, if temperature drops below X degrees, then do that. And as you start adding more complexity, those systems start to fail. So yes, part of this capital is to bring the world's best talent into Gigaton and to give them a opportunity to go and rethink or re-architect the control systems of tomorrow.

9:59Mike Butcher:I see. So is it going to be a case of you having to work alongside these systems? Will you have you said you had your own hardware? I mean, how deep do you have to go into these companies to, you know, to swap out the software? We will absolutely work with and in partnership with a bunch of those industrial incumbents. We have some of our own hardware, but we do. We definitely do not have enough hardware to build a cement plant ourselves. And some of those companies will build the actual infrastructure for a cement plant. So there's a huge amount of collaboration there. I see. Right. I see. So you're certainly you're sideling alongside the industry and working a lot with it.

10:44Mike Butcher:Can you give us a lot of people won't be familiar with this this world whatsoever, of course. So, I mean, could you give us a little bit of a picture? Is it is what you're building closer to a sort of a digital twin? or give us a picture of how the AI is going to operate. Yeah, so a digital twin is a component of it. So if I take you through from the beginning of the process, you'll walk into a plant, you'll try and understand what the process looks like by collecting as much data as possible. And this is relatively proprietary data. It's not like you can just Google for it. So it exists in the mess of an industrial world.

11:26You pull all of that data together and off the back of that, you build simulations or digital twins or predictions of what the plant process is going to look like over a given period of time. And then you use those predictions or that digital twin to put a controller on top of. And that controller can be thought of like a reinforcement learning controller where it's looking at all the possible predicted outcomes based on your inputs. So as an example, if I were to increase alternative fuels by half a ton, what do I predict is going to happen to the combustion state in the chemical process? Or what will happen if I increase the fan speed?

12:12Whatever that prediction may be, the reinforcement learning controller will look over that simulated environment, select the optimal targets for the plant to hit. and then the AI will plug in in closed loop control and actually control the plant autonomously. We talk about it internally as us controlling the world's largest robots. And I think it is a marvel that there's 3 ,000 odd cement kilns sitting all over the world that are spitting out terabytes of data every year. And we can get the best and brightest machine learning engineers to come here and play in the real world with these enormous robots and test adding their gem L1 control theory.

12:52Mike Butcher:So China apparently is already building fully autonomous dark plants. How close are we able to do that in the West? And how do you think your company is going to help us get there? I'd say China realistically are probably half a decade to a decade away, if not a little bit more. but that is meaningfully closer to that outcome than the rest of the West is or rest of the West are. Some of the largest producers in many industrial fields are now chasing after the same thing of fully autonomous dark plants. I think that is a necessary end game, but it's years away. The closest we are in the West now is a stepping stone of remote controlled operations.

13:41So where you have a hub, let's say a North American hub of a large cement producer, and from that hub, they're controlling multiple different industrial assets all over the country. We are certain that you cannot get to that fully autonomous future if you are stuck adjusting the software that controls the plant, which is fixed rules and logic that you have to continually tune by hand. So our long-term vision is to be the technological infrastructure that enables fully autonomous plants. There's no way that the existing infrastructure gets you there. But we are also conscious that you can't go into a cement producer today and sell them on a dream in 10 or 20 years time.

14:25They are worried about survival over the next year. And to do that, we have to be able to provide immediate ROI to customers today. So saving them cost and carbon today while rebuilding that infrastructure.

14:38Mike Butcher:Well, it's a fascinating area and I'm sure we'll be watching it with great intent. That was Josh Vernon, CEO of Gigaton, which has just raised$26 million, Series A funding led by Plural. Thanks for joining Path Founders.