Vinod, thank you so much for joining us.
MR. KHOSLA: Oh, it’s great to be here.
MR. PUKO: If we could start with a topic that I know, especially for a Washington Post reporter–that’s Washington–I just want to ask you a little bit about some things that the Biden administration says about the private sector and about your business, in particular, the business that you’re in. These guys are really big on climate change, but a huge premise for them is that, in many cases, it’s the private sector that really has to take the lead. It’s giant investment firms like BlackRock, big Wall Street banks and VC firms like yours that according to them are the ones that really have the capability and maybe the only capability to muster the vast sums required for the energy transition. I just want to know–from where you sit in the private sector, they’re talking about you–how much of that do you think is true?
MR. KHOSLA: I think it’s very, very true, and I think there’s a large difference between the venture community and what the BlackRocks of the world do. Technology breakthroughs are absolutely essential to have the solutions we need. We do not have solutions today that are scalable or economic. You can’t subsidize energy at scale, and so the administration is right. The technology breakthroughs have to come from the venture community. They have to come from the universities where research is going on, and these have to be large innovations, not minor innovations. We’ve spent too much time thinking about very minor innovations, so I do agree with that.
Having said that technology is absolutely necessary but not sufficient, and policy has to facilitate this. And I think the government with the IRA and the Net-Zero Act in Europe have done a very good job of incentivizing early technology development.
MR. PUKO: I’m glad you brought up the IRA, or the Inflation Reduction Act. We’re going to get there in a second, but to help us get into it, I want to put some of this policy talk in the context of your own career. You’ve been active in Silicon Valley for decades now, and at first, your work was not in this space. I think if I could say it was in more standard investments for VC firms, you know, computing, networking technology, and then that’s changed over the years. So I just would love to hear from you about you and your peers who have moved into climate tech and get a better understanding of what’s behind that. How much of that is because you’re doing the investment and money-making work that you’ve already done, and how much of it is because of a moral imperative?
MR. KHOSLA: I think moral imperatives are personal for people. I think it’s an important critical problem. I want to work on it, but none of that works unless you’re getting rates of return for your limited partners. So the economics have to work. As I say, you can’t defy the laws of economic gravity, whether it comes to investment capital going into these areas or it comes to the products you produce. In the end, green steel is nice, but if it doesn’t meet what I call the “Chindia price,” the price at which everybody in India and China would buy it, you’re not going to have scale. And too many of the climate technologies work at small scale under subsidy, but they can’t work unsubsidized and be market comparative at large scale, and hence, the demand won’t be there. So that’s what the breakthrough technologies are needed for.
Now, let me separate what we do. Creating a breakthrough so something is economic, where a plant protein hamburger is the same cost or maybe cheaper over time than, say, animal hamburger, that’s the role of technology. It has to taste as good or better, and it has to be cheaper. And then later-stage capital like BlackRock can come in and scale it, and you’re seeing that in solar and wind, that latest-stage capital, probably solar is about half a trillion dollars a year now being deployed is needed for scaling. But the high-risk part, the early technology breakthroughs is where there’s a bottleneck and not enough of it going on.
MR. PUKO: That’s interesting that you say that because there’s a ton of money coming into the space right now. A lot of it is government money, but as you sort of allude to, it goes into different areas.
I still want to keep the conversation on this a little bit broad for now, while we’re talking about motivation of different businesses in the financial sector and what Washington is trying to accomplish. The Biden administration officials will talk in some of the terms that you’re discussing. We noted that Special Climate Envoy John Kerry recently said that–well, he said a few times that, quote, “This is a new industrial revolution,” and in a recent interview, he made the specific case that there’s a lot of money to be made. But that’s not always the case, I think, as you were sort of alluding to. The IEA has pointed out that a lot of climate friendly tech, the margins are much lower for investors than fossil fuel has historically been in fossil fuel investments. So could you sort that out for me right now? Do you–
MR. PUKO: How far have we come along in making sure that there really is big money to be made for investors like yourself?
MR. KHOSLA: Yeah. You know, if you look at coal-powered plant, it is fully amortized. It’s 40 years old. Its marginal costs are low, and that’s what new technology like fusion technology or solar have to compete with. And it has to be just as reliable, and solar is not dispatchable electricity. You can’t have solar energy when you watch–when you want to watch Monday night football. You don’t want electricity when it’s being–when the sun shining or the wind’s blowing. You want it when you want to watch television or want air conditioning. So it’s important that we talk about the right thing.
Early technologies–and solar was the same way, fusion is definitely going to be this way–are expensive when introduced. When they scale, when they become 3, 4, 5 percent of a market, they should reach market competitiveness, and at that point, they don’t and shouldn’t get support. They don’t need it, and they shouldn’t get it. And if they do need it, it’s the wrong technology. And so this is the tricky part from the government point of view, when to support things, when to have subsidies decline and have the technology stand on its own.
And things like IRA get past the hump of the first 2, 3, 4 percent of production in the market, but it is up to technologists, not even investors, to have the breakthroughs that make these costs competitive. Can you make cement cheaper than current cement? I believe you can, and by the tenth plant being built, that will be the case with the kinds of technologies a company like Fortera is doing.
At that point, it doesn’t need any more support and should be off and running, and that’s sort of the whole purpose, I think, of both the technology piece and the early support these technologies need to take off.
MR. PUKO: So let’s talk about the IRA a little bit more, again, the Inflation Reduction Act. To remind everyone, it is the largest climate bill in history, by some measures, unlocking potentially a trillion dollars in government investment. But a lot of it goes into traditional solar wind manufacturing of those technologies and infrastructure, that you’re sort of suggesting there might be, you know, a window closing soon for which you don’t think those subsidies are necessary. So I’d just love to hear your thoughts, your broader thoughts on this spending from the government. Did Washington get the mix right of new technologies, emerging technologies, and just how to fund them broadly?
MR. KHOSLA: I do believe a lot of things are in the Inflation Reduction Act, but something like solar is at about a trillion dollars of investment a year, and it should not need any subsidy. So it’s time to stop subsidies in solar and wind, for example, possibly even electric cars because they are well on their way. We should be funding university research in these areas and nascent technologies which haven’t scaled yet to help them get to scale and prove for themselves that they are market competitive unsubsidized.
And that’s when, you know, the BlackRocks of the world come in and these technologies. They compute the IRR on the wind farm or a solar farm, and that’s the way it should be. Government should not subsidize anything at scale. It should only subsidize the kickstart of these technologies, because early on, there are higher production costs because they’re such low scale, they haven’t matured as technologies. Production methods aren’t right, but we’ve seen the cost of solar decline, and that should happen in every new technology. Whether it’s hydrogen and the IRA or the Inflation Reduction Act really supports hydrogen, it should be in cement and steel, which are new areas. So I do believe that should be the right approach to subsidies and incentives.
MR. PUKO: Well, Vinod, if I could ask you bluntly then, do you think that the IRA is the right climate policy? Did they do this bill the way the world really needs it to address climate change?
MR. KHOSLA: I believe the IRA is much more good than bad. If I were king, I would reallocate the dollars more towards earlier state technologies and fundamental university research.
MR. PUKO: That would be pretty helpful for your line of work, right?
MR. KHOSLA: Well, it would be, but it’s what’s needed. We need fundamental innovation, so things are market competitive.
And manufacturing is an area where we still need work on the West. China has captured most of the solar market. It’s unfortunate, and we have to be competitive with these new technologies. So one has to be smart about what to fund and what not to. More solar farms, more wind farms should be strictly market comparative, and they should raise funds in the public markets or from investors like BlackRock, not from the government.
MR. PUKO: Well, let’s talk about it in real-world examples. The IRA–and I should also bring up the big bipartisan infrastructure package from the prior year, which also put a lot of money, billions and billions of dollars, into different programs related to climate change. How has all that changed the decisions you’re making and what you’re investing in, if at all?
MR. KHOSLA: Well, it has changed. The IRA lets us invest in, say, a new steel technology, a radically different approach, which, if successful, would make steel cheaper than “fossil steel,” if I can call it that. Those are the kinds of bets the government should encourage, and there should be no–more university research in that kind of steel production or better fertilizer production, better hydrogen production. So these are–industrial heat is a huge area where we haven’t had the breakthroughs we need, so more research is needed there, but as is more venture capital investment on high risk, likely to fail, but highly consequential if successful kinds of technologies.
MR. PUKO: I will point out that those types of investments are often what get policymakers, lawmakers in the most trouble with voters when they tend to not work out. So could you tell us a little bit about some of the risks there, both from your side and maybe the political risk for these policymakers? Because a lot of what you’re talking about is–you mentioned a lot of these investments are probably going to fail, and they have long timelines too. We’re talking about technologies that still need 10 20–what?–maybe in some cases, 30 years to pay off to work?
MR. KHOSLA: Yes, they do take a long time. You know, fusion will be a dozen years from the start. Now we are halfway there, I think, and fusion is very, very promising technology. I think we have to do the messaging right.
University research doesn’t always work out, and we should be funding more of it because the things that do work out more than pay for the things that don’t. And that is the venture capital model too, and in deploying these technologies, I think getting plants started and subsidizing early product is very, very valuable for venture capitalists to take the risk. Without it, it would not be possible.
You know, if you have one fusion plant and 5,000 coal plants, the one plant is going to have a higher cost than it would if there were 5,000 fusion plants, and so we have to understand that new technologies don’t have scale and, hence, have higher costs. But we have to make sure that they actually get to economic unsubsidized market competitiveness if they scale.
And, you know, venture capitalists take a lot of risk and lose a lot of money, and we’ve seen that repeatedly in climate. And the government has to support the products they produce when they’re nascent, but only when they’re nascent. Otherwise, we won’t make this energy–you know, I want to emphasize the point that in solar, there’s–
MR. KHOSLA: –half-a-trillion-dollar business. That’s a pretty large economic chunk of the GDP that mostly goes to China. We can’t afford to do that with steel and cement and water and fertilizer and industrial heat or hydrogen, huge pieces of the economic pie at stake that we in America have to win. And that’s where government can play a smart role.
Unfortunately, too many dollars get allocated downstream to technologies that have already been de-risked and already scaled, and environmentalists play a big role in that. And I think they’re misinformed as to market economics and how things should work. In fact, environmentalists often do more damage than good in these kinds of decisions. They look for idealized solutions instead of practical solutions.
But I do think overall, the IRA will do much more good than bad.
MR. PUKO: I’ll also point out that a lot of the issues that you’re grappling with, in particular, are some of the hardest that there are in climate policy. You’re talking a lot about making steel and doing other heavy industrial activities, which are often–you have to do them at high heat. It’s very difficult to do them without some type of greenhouse gas emissions. The solutions there are much fewer and less proven than in your standard solar and wind power, just power generation that we’ve been talking about. You’re also talking about fusion, which has been sort of a Shangri-la for people for a long time, mass quantities of emissions, free power. Could you maybe explain to us one or two that are your favorites? And maybe if I could get more specific than just your favorites, I’m wondering, which one do you think can actually make the most money for investors over time, and then which one, if it’s the same or different one, that has the biggest capacity to reduce emissions, especially quickly?
MR. KHOSLA: Well, those two go together. Fusion is a classic example of something that would be a humongous market at scale and very doable.
You know, the experts have generally been wrong on fusion, but I do think if we solve that problem–and there’s probably a dozen good efforts at fusion, including our effort, Commonwealth Fusion, that if successful would literally change electricity production globally and the superior solutions to even solar and wind, though solar and wind will be very complementary. And we have investments in solar too. So I do think they can make a large difference.
I’m seeing a lot of innovation in fusion but also in areas that are hard, like fertilizer, steel, cement. I think we have solutions coming that I’m relatively optimistic about if we can get the kind of support IRA provided and the Net-Zero Act in Europe provided. So I am optimistic. I’m seeing the innovation, and, you know, people used to say about the electric cars, done that, tried that, General Motors did it, failed, it can’t happen. Then a lone entrepreneur like Elon Musk at Tesla made that happen. He singularly changed global view, and that’s why I argue we need a few instigators like that of change. And we are seeing that in all these areas, from plant proteins to steel.
MR. PUKO: I know you were also a little bit critical of environmentalists a minute ago, and I think we have time for one more question. I want to touch on that because–I get a lot of your points, but one real big issue for them that they have a lot of evidence for is that there is an urgency here with time. A lot of the investments that you’re talking about, the technologies that you’re talking about that could make a huge impact undeniably, are still, as we’ve talked about, maybe a decade or potentially, in some cases, a few decades away, but this has been the hottest year on record. There are worrying forecasts that it could keep rising even faster, and some scientists are worried about tipping points that could accelerate that even further. So I’m wondering how you square some of those imperatives and the need to make progress between now and 2030 with some of the more long-term thinking that you’re applying to the space.
MR. PUKO: Well, I had an op-ed just before the last COP conference in The Economist that argued that trying to get reductions by 2030 is a really bad idea. We don’t have the technologies that can market competitively scale, and we are better off developing those technologies–and those are well in process–and then trying to scale them in 2030s and 2040s, not now when they’re uneconomic.
Look, the environmentalists have done a fabulous job of identifying the problem here, and I think we owe them a lot for identifying the problem. When they come up with solutions, they’ve generally done a poor job. I think when they were pushing hydrogen cars, that would make no economic sense, and California tried to do that. Others did. So I think the solutions have to be based on technology breakthroughs, and environmentalists have generally not been good at predicting it or said let’s apply the solution at any cost, even if it’s uneconomic.
They’ve also been misguided. GMO is the classic area where the environmentalists have been misguided and done a huge amount of damage. I also argue that environmentalists caused more coal plants to be built decades ago because of the opposition to nuclear, which was a bad policy thing that they pushed.
So I do think we–there’s a large role for environmentalists and activists to make the world aware of the problem, because it is critical and urgent, but that doesn’t mean we have the solution. We do not have those solutions. And we will not scale to solutions if we try and apply them and have reductions by 2030.
MR. PUKO: Vinod, you mentioned COP. That’s the UN Climate Conference. It’s only a few weeks away now, and all of these disputes are going to be hot button issues there. So stay tuned for more coverage on that from us in the weeks to come.
For now, Vinod, thank you so much for joining us on Washington Post Live.
MR. PUKO: And everyone else, just please stay with us for a few minutes. We have more interviews coming up.
MS. KELLY: I’m Suzanne Kelly, CEO and publisher of The Cipher Brief, a national security-focused media organization.
We talk a lot at The Cipher Brief about the impact that artificial intelligence is having on our world, and I’m delighted today to spend a couple of minutes talking about how AI is impacting environmental sustainability. And joining me in this conversation is Andy Markus. He is the senior vice president and chief data officer at AT&T. Andy, welcome.
MR. MARKUS: Thanks, Suzanne. Thanks for having me.
MS. KELLY: You know, I thought we’d start kind of with the basics. Can you tell me what role AI is playing in helping companies reduce emissions?
MR. MARKUS: It’s a really good question. AI–and when I talk about AI, it’s data, it’s automation. It’s all types of AI. It’s traditional AI as well as generative AI. It’s a really important tool in helping companies, you know, really fight, you know, emissions and get to where they want to be.
I’ll give you a couple examples. A really great example is energy efficiency. Using trends and patterns and how, you know, demand and how machines and products, devices use energy, you know, we can either draw down or even cut off, you know, the energy consumption based on those patterns, right, and to be a lot smarter about how we leverage energy and, you know, create emissions.
Another good example would be proactive incident management, right, using trends and patterns with AI and finding the problems before they occur, so we don’t have the swirl of fixing the problem in a crisis situation.
A third example could be supply chain optimization, right, the ability to use AI to design, to create, to order, to ship products in a much more efficient way so that we really minimize emissions.
And a last example maybe would be climate forecasting and climate modeling, so that we use both natural severe weather as well as climate change in our planning so that when those occur, we’re not scrambling, and we can reduce emissions in the process.
MS. KELLY: I’m curious to know what kind of an impact are a lot of these applications having on a company’s sort of overall emissions?
MR. MARKUS: At AT&T, we believe connectivity can have a really big impact, you know, through fiber, through 5G, through our Internet of Things, through our connectivity services. We believe that, you know, that can all really help our clients, our business clients, reduce their emissions.
At AT&T, we have a goal that by 2035, that, you know, we would like to help our partners reduce their emissions by a gigaton, and a gigaton is, I think, equivalent to one billion flights between New York and Los Angeles, so that’s a lot. So we think it have–can have a really tremendous impact.
MS. KELLY: All right. Let’s get personal for a minute. I’m really curious to know how AT&T itself is applying AI to advance its own sustainability commitments.
MR. MARKUS: Excellent question. So all the things we talked about above, you know, we’re doing, so we’re on our path for sure. I’ll go deeper in one area, and that’s our dispatch optimization for how we dispatch our vehicles to serve our customers. So AT&T has about 50,000 vehicles in our fleet. It’s one of the largest fleets in North America, and if we think about, you know, all the different options that we could dispatch those vehicles and those technicians on a daily basis, it’s trillions. So there’s a lot of options we have to work through. So it’s a great AI problem.
We solve that, you know, with some constraints, right? We want to make sure that our technicians don’t drive too far from their home. We want to make sure they have their breaks, you know, that they can take lunch. We want to make sure that we solve our most pressing customer problems first.
So we take all that into an AI solution, and again, there are trillions of options that we could look through, and we optimize that, you know, leveraging AI. And when we do so, it has a lot of impact. We drive 20 percent fewer miles, you know, with an AI-produced solution, and we actually reduce our emissions by over 51 million pounds. So, you know, it’s a really great use of AI, and it’s really making a difference. And it’s allowing us to do a couple of things. It’s allowing us to, of course, save, you know, money for the company, but it’s also allowing us to be good stewards of the climate as well as, most importantly, to serve our customers in the most effective way possible.
MS. KELLY: It does seem to be the consensus that every single business can somehow benefit from using AI, but I’m curious. What are the network requirements for businesses, whether they’re small, medium, or large, to be able to use AI in their business?
MR. MARKUS: That’s another great question. You know, at AT&T, we’ve moved a lot of our network and data operations to the cloud, and that’s helping us, because I think in the future, the latency requirements of AI is going to require the inferencing of AI to be a lot closer to the edge. And so the movements that we’ve made put us in a good position, and I think other businesses have to think about that as they think about the future of AI. Latency is going to be key and having the inferencing of that AI closer to the application is going to be paramount.
MS. KELLY: I’m curious to know. It’s obvious AI is really just in the beginning stages, and it’s going to be scaling, whether we want it to or not, and I think there are a lot of reasons for a yes on that. How is AT&T preparing to meet those expectations as businesses start to get a sense of what AI can do for them?
MR. MARKUS: At AT&T, you know, we really believe that AI is going to allow us to reimagine how we operate the company, to become kind of a core fabric of what we do in so many parts of AT&T.
We’re leveraging AI across things like–I mentioned dispatch optimization–across how we manage fraud for our customers and our company, how we develop software, how we manage and plan the network. So it’s a very important part of what we do AT&T.
We just launched what we call “Ask AT&T,” which is our generative AI enterprise orchestration layer, which allows all parts of the company to leverage generative AI to make what they do in their area, you know, better and more efficient, more productive. So, you know, we believe AI is super important and has a really big impact on the future of AT&T.
MS. KELLY: I think just being able to think about how all these things are going to impact the future has got to make your day job the best day job ever.
Andy Markus, senior Vice president and chief data officer AT&T, thanks so much for joining us.
MR. MARKUS: Thanks for having me.
MS. KELLY: Now back to my colleagues at The Washington Post.
MR. PUKO: Hello, and welcome back to Washington Post Live. For those of you just joining us, I’m Timothy Puko, The Post’s climate correspondent.
Now we have up our next interviews. First off, the CEO of Gro Intelligence, Sara Menker, and along with her, we have the CEO of FreeWire Technologies, Arcady Sosinov–sorry about that, Arcady. But thank you, Arcady and Sara, both for joining us. Welcome to Washington Post Live.
MR. SOSINOV: Thanks for having us.
MS. MENKER: Thanks for having us.
MR. PUKO: Sara, if I could start with you. I was really impressed by the last slide that said that Gro Intelligence can predict anything. Tell us about the artificial intelligence you use to do all these forecasts, especially potential food shortages tied to global warming. Could you just explain how all that works?
MS. MENKER: So I think Gro Intelligence can predict anything is–you know, it’s a little broad. It’s anything in sort of our complex real-world systems and sort of how, you know, the interplay between them. But food and agriculture is such a central part of sort of our world, so it ends up sort of linking to many different things.
But, you know, we use sort of artificial intelligence on sort of multiple layers of what we do at Gro. You know, the first layer of sort of building predictive models is obviously having really good data to power these predictive models, and so the first layer of AI that we use is actually in organizing information that we take from, today, over 170,000 different data sets around the world that come in many different formats and languages to actually organize it, so just literally taking the definitions of different products as labeled in different countries, making it the same. So it’s really a mapping and a definition sort of process, and that mapping process initially was human-centric and is now highly automated. And that’s the first layer of AI.
The second layer of AI that we use is actually in creating new data sets from that data. So before, again, you build predictive models, we have to complete gaps. In many parts of the world, as you can imagine, data sets are–there are big gaps to data sets or countries don’t report data sets like what fields are growing, you know, what crops. So you have to build algorithms to actually detect that first before you can then say, “Then how can I predict how much of that is growing there?” So we use AI to actually develop new data sets to fill gaps, whether it’s again about countries are actually the whole world. Satellites go down often. So we have a generative AI model that creates synthetic temperature data sets at times when NASA is not reporting data.
And then finally, you then use it on a final layer, which is actually building predictive models, so everything from forecasting the supply of–the production of a particular crop or product in a country or in a specific district in any part of the world to predicting demand patterns, trade patterns, et cetera.
So we have over 2.9 million models that we’ve developed that are in the system today, and as you can imagine, because agriculture sits at the intersection of the, you know, ecological preservation and economic growth, a lot of these models are not just about agricultural markets, but they’re actually about, you know, economies and economic indicators as much as they are about actually climate data sets. And that’s really where the “everything” comes from is because you quickly then realize that’s a big part of the real world.
MR. PUKO: Arcady, I want to bring you in here too because you’re also in the artificial intelligence and predictive modeling business. Your particular role or niche is helping gas station owners figure out which ones would benefit from having EV charging stations at their businesses. So basically the same question for you. I’d just love to understand a little bit better about how your system helps make those decisions.
MR. SOSINOV: That’s right. And thanks for having me.
So fundamentally, we’re solving the two pain points within the world of charging infrastructure. The first is, where do I deploy charging infrastructure for the best profitability, the best cash flows? We have an AI model called “Mobilyze,” and it’s a system that allows us to really predict out three things; one, where do I put charging infrastructure today? And I don’t mean, where do I put it in a state or in a certain city. I mean specifically down to a three-by-three-foot square plot of land. Where do I place charging infrastructure for the best cash flows, the best returns, and for the lowest impact on the utility grid? As part of that, the AI model actually forecasts utility utilization over the next ten years on five-minute intervals. I can tell you six years from now, exactly how many vehicles are going to arrive and use that charging infrastructure within that five-minute window. It also then predicts out what kind of impact that utilization is going to have on the utility grid and, more specifically, what impact that has on your utility bill.
The site hosts have to really understand both utilization, which drives revenue, as well as the impact on the utility and their utility bill, which drives costs, and that helps determine the return on investment, or the IRR if you want to look at it that way, of deploying charging infrastructure. And so we call our AI models “AI for Main Street America.” We’re helping gas stations, convenience stores, retail owners and operators determine whether charging is the right thing and new business model for them to deploy and exactly where to place that charging infrastructure in order to create the best returns on a portfolio basis.
Now, this is on the back of really compelling technology that we built that integrates battery systems with charging infrastructure, and the combination of the two allows us to deploy charging anywhere you want without requiring utility upgrades.
So, as we know, high-power, ultra-fast charging requires heavy utility infrastructure. That being said, we have mitigated the need for that utility infrastructure by pairing a battery system with the charger, which allows it to deploy anywhere, which then paired with the AI model allows us to deploy the very targeted network of chargers with these gas station operators and maximize their cash flows over time.
MR. PUKO: So I have a question that I want to throw at both of you. Whichever one of you wants to jump in, you know, please do it. Predictive modeling has been around for a fairly long time now in different forms. It’s become, of course, more sophisticated over the years. But when we talk about artificial intelligence, I think people are thinking of like ChatGPT and some of the very recent lightning-fast advancements in that field. Are your models incorporating some of those very recent advancements, or when you talk about being AI driven, how exactly has that changed in recent years, and what’s it mean for your businesses?
MR. SOSINOV: Yeah. I mean, it’s interesting. As we call it “classical AI,” and I think my colleague here would agree. The fact is we’re not using large language models or generative AI as they call ChatGPT and other services like that today. Instead, we’re using, I guess, a form of classical AI, which seems like an interesting nomer. We are using predictive algorithms to ingest the large data sets and make decisions that otherwise a human would have found impossible or would have taken years to make that decision, and that is still a version of AI, but it is a little bit different than the generative technologies that are catching people’s attention.
That being said, there is AI being used up and down the stacks of technology companies like ours. In that, what we just talked about, the Mobilyze system is on the product side, is customer facing, but our software engineering teams are using really impressive AI tools in order to speed up the development, to speed up their work.
When we look at the output of our software development teams today, it has gone up by an order of magnitude because they’re able to use some of the more recent developments around ChatGPT, not customer facing, but internally facing, to be able to write code much more quickly with fewer bugs and that we can get out to market very quickly. So there is some generative AI being used on the internal side, but on the external side, it’s still a lot of predictive models, similar to the models we’ve seen over the last couple of years.
MR. PUKO: Sara, is that the same for Gro Intelligence?
MS. MENKER: Yeah, I think it’s a little different in terms of sort of based on our domain and sort of how we use it.
So, you know, the way I describe it is that Gro is a vertical AI company. We have a lot of depth in specific areas and obviously started out in the food and agriculture space and sort of by doing that ended up having models on climate and economics, et cetera. But our system is a really good AI for the things it knows about. It’s a knowledge graph, and it’s a knowledge database, and it’s about knowledge. It’s a knowledge engine.
And if you think of things like ChatGPT, their horizontal AI place, they’re great reasoning engines, but they’re actually not knowledge engines. That’s why they hallucinate, and they hallucinate for many other reasons as well. But essentially, it cannot give you a factual answer because it’s not smart about specific issues, but it can reason its ways into an answer. And it’s–that’s why it’s a large language model.
So for us, the intersection of the horizontal and the vertical world is really where it’s gotten exciting, you know, over the past sort of six months, which is, you know, plugging in our vertical AI systems into horizontal AI systems actually allows us to interact with very complex data sets differently. And it allows our customers to interact and access very complex data sets more accurately, but essentially, think of a world where if you go to ChatGPT for today and you ask a question about, you know, what will wheat yields in France be, you know, at the end of this year, it can’t give you that answer. That’s just not what it’s designed to do.
But a version ring-fenced into our knowledge graph, which is a vertical AI, actually can give you that answer, and so it changes the user experience fundamentally. And so those worlds coming together is one that’s particularly really exciting, and I think actually will become groundbreaking for verticals and for specific sort of industry applications. And that’s where we’re really sort of headed as a world, and I think that particularly is really exciting because it presents opportunities of delivering complex products to users and customers that are actually not data scientists or not data nerds and geeks. And I think it develops an understanding and a knowledge base that’s much broader. So I’m particularly really excited about it.
And then I think the term “generative AI” today is used in sort of the LLM concept, but as I mentioned, you know, at Gro, we use generative AI for developing, you know, land surface temperature data sets, and it’s just a–it’s a form of AI but, in this case, applied to imagery and remote sensing. So I think there’s many other applications as well, but it is, to me, the most exciting thing is–the reasoning world coming together with sort of the knowledge world, and those two AIs combined, I think, will provide an endless number of opportunities and over time actually drastically reduce the cost of delivery of everything from sort of knowledge and information for managing food security risks to delivery of health care. So it will be, I think, really transformational in that way.
MR. PUKO: Well, this is all pretty high-level stuff, and I don’t want to overlook the old-fashioned human intelligence that often goes into making these systems. So, Sara, let’s start with you here. I’d love to hear a little bit more about your personal story. You grew up in Ethiopia, went to college in the U.S., worked on a pretty vaunted commodity trading team at Morgan Stanley. What led you ultimately to strike out on your own and develop Gro Intelligence?
MS. MENKER: Other than insanity, I would say it’s–you know, being a commodities trader, you see how sort of–you know, the one thing about trading commodities versus, say, like credit risk or something like that is, you know, when you’re trading oil or you’re trading natural gas or you’re trading power, you are trading these real assets, and there are things being delivered, And there are sort of these real-world consequences, and agricultural products are in sort of the world of commodities.
I had personally seen how the energy markets, in particular, because I was an energy trader, had been transformed due to better data, data availability, and I was very sort of interested in how do we bring that type of sort of access and transparency to high-quality data and information to agricultural markets.
Now, significantly more complicated because, you know, very few agricultural products are actually traded in a sort of a financial market context. You know, I could count them in both hands, less than 10 fingers, in terms of what’s liquid and traded. Whereas there’s tens of thousands of different agricultural products, and so we’re–you know, we had to do this very differently and had to do it way more efficiently in terms of data access information. But to me, I don’t think we fix systems we don’t fully understand.
And so the reason I left and started grow was, you know, I was really interested in how do we solve for food security? Like why have we been talking about it for decades, and yet we still haven’t solved this? And to me, it became very clear that all the major decision-makers around the world were relying on pretty, you know, stale data sets, very macro, not micro enough, definitely not predictive, to make these decisions. And it was like, we have to start by building infrastructure to help the world better understand the system, and once we do that, then we can start to sort of advance the knowledge around it and how we fix it. So it was just, I guess, a desire to solve that problem, and then it was finding all the right tools along the way to do that.
And so, you know, like I said, we’re very much a problem–like we–we didn’t have a technology that we’re looking to sort of apply to a specific domain. We started with we need to solve a set of problems, and then it’s like, what do we do to do that? What do we use to do that?
MR. PUKO: Now, Arcady, you were born in Ukraine and also had a successful career in finance before launching FreeWire. Was it insanity for you too, or why electric vehicles?
MR. SOSINOV: I mean, that’s certainly part of it. As any good entrepreneur will tell you, you set out to solve hard problems, and the harder the problem, the more exciting it gets and the more motivated you become. So that’s right. I emigrated to the U.S. in 1991 during the fall of the Soviet Union, and my family, as we settled in New York and then eventually Boston, we took very blue-collar jobs. And so my father for 26 years drove a cab in New York and then in Boston, and so, you know, that sort of started my passion and love for anything automotive.
I did eventually graduate in Boston. I went and worked in the finance industry. I was a quant at a hedge funds, which is, frankly, just a fancy term for data scientist, which I think helps you understand my sort of passion for anything AI, including all the data science underneath that. But eventually, after I spent some years in hedge funds, I started to see this really interesting new market forming around electrification of transportation and thought this could be an opportunity for me to get into the automotive space, where I do hold a passion. And I wanted to find the hardest possible problem I could find to solve, and the hardest problem wasn’t building the vehicles. The vehicles were being built, and there were great companies out there, including Tesla, including some of the more traditional OEMs who had compelling products coming out. But the hardest problem to me was the interface of that electric vehicle with the utility grid.
I did the back-of-the-envelope analysis, and I found that we don’t have the available power to support ubiquitous fast and ultra-fast charging to support the growth and adoption of electric vehicles that we see today and that we expect to see by 2030, and the utilities weren’t moving fast enough to upgrade their aging grid infrastructure and their legacy systems to allow us to get there.
And at the meantime, I saw this burgeoning economy and market around distributed energy resources, battery systems, renewable resources, and I thought there had to be a way to pair these two technologies together. And so that’s what I set out doing. It was a pairing of battery storage yields, mitigating utility build-out, and it was supporting this electric vehicle economy and market that was building. And that seemed like a very challenging problem to me, but it also seemed like a great opportunity because I truly believed that the company that could figure out how to get all of that power, all that power to electric vehicles without requiring billions of dollars, what we estimate is $50 billion of infrastructure build-out costs, that company would become very valuable.
And so that’s what we set out to do. We’ve become very advanced in it. We now have products deployed in over 30 U.S. states and seven countries. We’ve raised $250 million of venture to private equity to sovereign wealth dollars to sort of help propel our mission, and it’s still, I can promise you, a challenging problem that I love to do every day.
MR. PUKO: When the two of you talk about wanting to solve the hard problems, it reminds me of some of my own work covering financial markets before I got to The Washington Post, and for a long time there, especially throughout the 2010s, it seemed like leaders, business executives, traders who operate in these markets, either as investors or as the companies that are getting invested in, a lot of them had–it just took a long time for many of them to get their minds wrapped around or to even prioritize the systems that could address some of the real changes coming along with climate change and greenhouse gas emissions. So, Sara, I want you to talk in that context about the Carbon Barometer partnership that Gro Intelligence recently launched. Tell us a little bit about how that fits in, the partner that you have working with you there, and just broadly again about how that system works.
MS. MENKER: Yeah. So the Carbon Barometer is a partnership that we did with Kepos Capital, which is actually a quant hedge fund that was started by Bob Litterman and Mark Carhart, two, you know, very well-known figures in the quantitative finance and sort of pure investing world, really. But both Bob and Mark, you know, really sort of believed that, you know, having a systematic lens to understanding the price of carbon was the way to do better investing in public markets on sort of–on climate itself, right?
So the work really started originally when they had developed and had started working on the concept of a carbon barometer as part of their work to launch a green and sort of a climate fund that was just a pure hedge fund but had a systematic lens to it. And we knew each other from our regular work at Gro, where we actually sell data to systematic and sort of quant funds. That is one of the segments of our customers, and Bob and I had gotten to know each other. And he had this idea which was, “Hey, why don’t we partner together, and we will bring sort of what we’ve done with the Carbon Barometer, but we’re not a data company. And you guys have obviously coverage of all sorts of, you know, climate data sets, but you’re not doing as much work on the carbon space. Why don’t you take sort of this methodology, let’s systematize it and actually automate the ingestion of these very complex policy paperworks to actually determine systematically, what is the implied price of carbon in every part of the world?” Then we did this for 85 percent of emitters around the world to say, you know, what countries are actually doing over time to sort of–to price carbon?
And so we set up this partnership. We put together an independent advisory board to vet the methodology and essentially launched it earlier this year, and, you know, what it does–and it’s an open–it’s actually open and available for people to use on our website. And what it is is, it lets you look at what the global price is, which is about $18.50 as of 2021–2022 numbers will actually be released very soon because all the policy documents are almost fully loaded and in all the different sources, like the OECD and the IEA, et cetera, which we tap into. But what we look at essentially is what–how much greenhouse gas emissions is every country sort of producing, and then what kind of subsidies do they have on actually fossil fuels, which many countries have subsidies? And then what types of mechanisms do you have for reducing that? And it could be everything from trading schemes to outright taxes, et cetera.
And so there’s seven policy levers, and we rank every country in the world. And what you see is–you know, you get to a global price of about $18.50 or so, but the distribution is everywhere from the highest price in the world is in Spain at about $129 a ton to, you know, the lowest are actually the high-emitting countries that are large fossil fuel producers and have no sort of carbon policies in place like Iran, for example, where you’re emitting a lot, you’re producing a lot, and there’s lots of subsidies and essentially no–nothing to mitigate it. So it’s actually a subsidy, not even a price on carbon. It’s a negative price, essentially.
And so it’s just–it’s a really interactive way of looking at it, but also measuring what policy measures countries are taking over time to actually put a price on carbon, which is what it’s fundamentally going to take to transition our economies. It’s a very expensive process.
MR. PUKO: Sara, it is a huge challenge.
Unfortunately, that’s all the time we have today to discuss it. I just want to thank you both, Arcady and Sara, for joining us, and for all of the viewers too, thank you so much for joining us on Washington Post Live.