Cleaning Up: Leadership in an Age of Climate Change

NVIDIA Rising: Will The AI Boom Bust The Planet? | Ep215: Josh Parker

Episode Notes

In 1849, when the Gold Rush hit California, the people who were certain to make money were not the miners, but the sellers of picks and shovels. Indeed, America's first millionaire, Samuel Brannan, made his fortune by adding huge margins to everyday items that suddenly become high in demand.

Today's sellers of picks and shovels are those providing the hardware and infrastructure to the software and platform providers, and one company stands apart as the beneficiary of the recent boom times in artificial intelligence: Nvidia.

They're currently among the top three listed companies in the US, alongside Apple and Microsoft, and are incredibly profitable, with estimated margins in excess of 40%. They've been around for 30 years, and are much more than simply chip fabricators.

This week on Cleaning Up, Bryony Worthington sits down with Josh Parker, Nvidia’s head of sustainability, to explore some of the challenges and opportunities he sees in the AI and Climate space.

Leadership Circle:

Cleaning Up is supported by the Leadership Circle, and its founding members: Actis, Alcazar Energy, Davidson Kempner, EcoPragma Capital, EDP of Portugal, Eurelectric, the Gilardini Foundation, KKR, National Grid, Octopus Energy, Quadrature Climate Foundation, SDCL and Wärtsilä. For more information on the Leadership Circle, please visit https://www.cleaningup.live.

Discover more:

Episode 204, the Sierra Leone special: https://www.youtube.com/watch?v=z-5QjSfy2SM
Nvidia’s Sustainability Report:
Earth2: https://www.nvidia.com/en-gb/high-performance-computing/earth-2/
cBottle: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/earth-2/models/cbottle
Michael’s piece on AI efficiency:
https://mliebreich.substack.com/p/ai-data-centre-power-and-glory-an

Episode Transcription

Bryony Worthington

This is the big debate, isn't it? Does a growing demand curve mean you see more innovation in clean faster, or is it just meaning more coal burn and gas burn and we're just never going to catch up with ourselves?

Josh Parker

I think about this question a lot, and I'm happy to say that looking at the data, I'm a very sincere optimist about where we're going to be in the medium term. One thing to keep in mind is that AI, once we really take advantage of it, I think, is going to end up being the best tool for sustainability that the world has ever seen. Because it's helping us innovate in ways that I think will lead to technological solutions to a lot of our sustainability challenges. You've seen so much happen just in the past year that the current generation of hardware and software and cooling looks, to be honest, very little like it did even a year, year and a half ago. From the previous generation to the current generation, Blackwell, we’re 25 to 30 times more energy efficient for the same inference workload. So whatever your inference energy looked like a year, year and a half ago, on top of the line, most efficient hardware, if you run that same workload now you're using 1/30 of the energy.

Hello, I'm Bryony Worthington, and this is Cleaning Up. In 1849, when the Gold Rush hit California, the people who were certain to make money were not the miners, but the sellers of picks and shovels. Indeed, America's first millionaire, Samuel Brannan, made his fortune by adding huge margins to everyday items that suddenly become high in demand. Sadly, he ended his days in poverty, a classic boom and bust story. Today's sellers of picks and shovels are those providing the hardware and infrastructure to the software and platform providers, and one company stands apart as the beneficiary of the recent boom times in artificial intelligence: Nvidia. They're currently among the top three listed companies in the US, alongside Apple and Microsoft, and are incredibly profitable, with estimated margins in excess of 40%. They've been around for 30 years, and are much more than simply chip fabricators. They recently issued their latest sustainability report, and I wanted to sit down with Josh Parker, Nvidia’s head of sustainability, to explore some of the challenges and opportunities he sees in this space. Please join me in welcoming Josh Parker to Cleaning Up.

Josh, thank you so much for joining us here on Cleaning Up today. I wanted to start, as we always do, by just asking you to introduce yourself in your own words. Please.

Sure. So Josh Parker is my name. I'm the head of sustainability at Nvidia. Been here for a couple of years now. Background in engineering and law, but I've been in sustainability for a long time and a techno-optimist about AI for sustainability. Happy to talk about that.

Well, it's a very topical subject, so we're very glad that you were able to make the time, especially as you've had a very busy time recently releasing your sustainability report. But before we get onto that, I just wanted to ask a little bit about your journey. So how did you come to be working in sustainability?

It's a great question, and I've had an unusual path to get here. At the previous company I was working at, Western Digital, I was there for quite a while, but had a couple of roles. One of them was doing ethics and compliance in Asia. And as I was coming back from that assignment to the United States, the Board of Directors was telling the management team that they needed a sustainability program because the company didn't have one yet. And you know, this is several years ago, and I was in the right place at the right time, excited to take the opportunity, and I haven't looked back since, it's been fantastic. I thoroughly enjoy the field, the specialty, and I'm still learning things every day, but I feel like several years in now, I'm finally figuring out a little bit of how to do the job well. It's a super fun field.

So Western Digital, I assume, was another tech company, with manufacturing in Asia. Was that the brief?

Yeah, they provide storage. IT storage, hard drives and flash, at the time, and, yeah, a lot of manufacturing in Asia.

So you started learning the craft there, and then what led to you arriving into Nvidia?

So I wasn't necessarily looking for a different role, but I was trying to recruit somebody from Nvidia to come work for me at Western Digital, and she told me that actually the head of sustainability at Nvidia was retiring, and so wondered if I wanted to apply there. So I did, and of course, I couldn't say no to that. It's a fantastic company, and a really unique opportunity to try to shape sustainability in a meaningful way at a company like Nvidia, that has this amazing technological solution, and also broad reach to try to help advance sustainability and advance AI for good, more broadly.

And I'm pretty sad that everybody knows who Nvidia are, but just in case there's one person out there who's not heard of you. Can you just give us a snapshot of what Nvidia is as a company and where it is today?

Sure. So we're just over 30 years old now. The company was founded by our current CEO, Jensen Huang, and he's been at the helm ever since then. Really brilliant leader. And we started out as a company focused on accelerating graphics, finding a way to make computers do graphics better. And the realization was that if you have dedicated hardware to do the math that's required for graphics, you could put that next to the CPU, the existing brains of the computer, and make everything faster and create beautiful graphics. And Nvidia was very successful at that for many years. Still is the market leader in computer graphics. We provide the GPU for the Switch 2 that just came out. But many years ago, we realized that the GPUs that we were creating, which stands for graphical processing unit, are really good. Because they're so good at math and matrix math and running all of that in parallel, it's really useful for other uses, like high performance computing, for simulations, for physics simulations and super computing, and also for AI. And so we, many years ago, started developing the software libraries and tools that would help people use those GPUs for those purposes. So we've really evolved as a company, and now the enterprise side of our business, providing those solutions for high performance computing. And AI is the majority of our business.

There must have been a moment when — was it OpenAI? Was it Microsoft? Or Google even, who just thought, ‘hang on a sec, this is going to transform our ability to do large language models.’ When was that moment? And who were the actors involved in that?

Well, yeah, that's an apt term. So transform is actually the term that was used in a paper several years ago, I think it was 2012, when the transformer technology was proposed. And so there have been a few milestones, going back to AlexNet, where people used GPUs to accelerate image classification. And then the transformer model came out. And then, of course, 2022 when ChatGPT burst onto the scene, was a real watershed, because it showed the world that AI was already very useful. And people had been saying that AI is going to be very useful soon in the future, there’s a lot of potential there, but people were shocked with how much it could do, especially for consumers, already in 2022. And then, of course, we've seen a lot of investment in AI since then, as companies try to take advantage of this new technology to improve their productivity and create new scientific discoveries and so much else.

But was it a shock to you as a company, how quickly this blew up?

So the company has been anticipating it and preparing for it for many, many years, and I honestly don't know if the sudden explosion of ChatGPT was a surprise to Jensen and our other executive leadership or not. In terms of how sudden it was, we knew it was coming. The company knew it was coming. And like I said, years and years of preparation to create the software so that we would be ready for that moment. But it really was an amazing explosion, and since then, I started calling it the Cambrian explosion of AI, because we see so much innovation, a little bit chaotic right now, but so much innovation that's driving really exciting new technologies, new discoveries. And it's really a fun time to be alive, if you take a step back.

And this kind of moment means that you're now the second largest company in the world?

Yeah. So the value of the company has grown significantly because of the huge value that everybody's seen in AI. And of course, we're selling more and more systems. So yeah, we're one of the most valuable companies in the world. And I think that's just a tribute to the potential for this technology and the potential for intelligence. When you are able to convert data and energy into intelligence, that's hugely valuable, and it’s really hard to put a limit on where that stops being valuable. Because is it useful in manufacturing? Yes. Is it useful in education, in transportation, in buildings, and anything that you want to do. Having more intelligence is typically a good thing, and we're at this stage where the sky is kind of the limit. And it's also one thing that a lot of people don't appreciate. It's a very democratizing technology. In the past, you needed a lot of technical skills to be able to write code or to understand technical topics, but AI helps a lot of consumers bridge that gap. And vibecoding is a thing where you're able to just talk about what you want to achieve, and AI helps you do it. So there’s huge potential, and I think that's why the company has been so successful.

I think maybe people have a perception of you as just making the widgets. You know, you're the manufacturer of the GPUs, but you do provide lots of services using your own products as well. The way I came to know you as a company was actually through meeting someone who worked on autonomous vehicles, mapping out the terrain, because the product you have and your understanding how to use the product means you're brilliant at spatial mapping. And that's got direct assistance into the autonomous vehicles market. So you're much more than just the manufacturer of a chip right, or of a processor?

That's right. And we really design and develop this integrated platform, this infrastructure that's used to create intelligence. So most of our engineers are actually software engineers, not hardware, and they're working on developing the ecosystem broadly, to enable us to create these tools for everybody to use, whether it's for autonomous vehicles, whether it's for manufacturing. And you mentioned, you know the spatial aspect to this, the fact that we have our roots in visualization and graphics means that it's easier for us than a lot of companies, potentially, to marry AI with that 3D modeling and the graphics. And it creates some really cool things, especially for autonomous vehicles, where you need to have tons and tons of data, you need to run simulations and 3D environments, to make sure that everything is safe with digital twins and manufacturing as well. If we can use AI and supplement that with 3D modeling, we're able to drive huge efficiencies in manufacturing that free up more resources and help us be more sustainable.

So those are the sort of positive things that you're excited about, because they are potentially transforming the way the physical world works. But most people's experience of AI to date probably might be, would you like to make your own emoji? Would you like a video of your cat on a skateboard? You know that's very, very sort of, it's almost a distraction rather than intelligence. So is there a disconnect in the public mind between what's actually the potential here and what's actually being shown to us?

I think so. I think there's a bit of an education gap in terms of what average consumers see in their day to day life that AI is doing, and what AI is doing behind the scenes at companies and in organizations and countries, that's perhaps more impactful and more substantial. Certainly, there's a lot of value in entertainment, and I use AI as a consumer for productivity in my role every day. It's made me much more efficient. It's gotten rid of some of the drudgery, and I think made me more informed. So it's been, I think, very useful personally for me in a professional context. But yeah, my kids love to use it to create— my son created a ton of images of cheetahs sitting on thrones, for some reason, surrounded by gold, something in his brain that he wanted to see. So the entertainment aspect of it is one that's very front and center for consumers. But behind the scenes, a lot of the work that AI is being used for is really monumental, and you see this in drug discovery efforts and material science by companies like Google and Microsoft and a bunch of startups. This, again, is another good aspect of the democratizing effect of AI, is that you can take one of these open source models and potentially use it for very, very beneficial goals. But yeah, to answer your question briefly, a lot of the work that AI is doing, especially the work that is creating new discoveries leading to significant sustainable outcomes, is really behind the scenes. It's not necessarily in the chat bots.

We should come to the potential negative externalities of all of this wonderful creativity that's being unlocked. Intelligence, but it comes at a price, right? Because ultimately, it feeds on electrons and on electricity, and it is quite an energy intensive way. I mean, tell me a little bit about how it compares to… In the old days, we all used to search on Google. Right now we're asking a bot questions, what's the relative change there in terms of energy?

It's really hard to put your finger on that, and nobody's as frustrated about that fact as I am, because I would love to have hard figures about it, but that question: how much energy and water and resources is each ChatGPT query using, or Gemini query using, it's really hard to put your finger on that again, because the pace of innovation here is so rapid. And you've seen so much happen just in the past year that the current generation of hardware and software and cooling looks, to be honest, very little like it did even a year, year and a half ago. And I'll give you one compelling example. So our previous platform that was top of the market a year ago was called Hopper, and very successful, most efficient, most performant platform ever

Named after a famous female mathematician, right?

Exactly. Yes, that's right. And then the current generation of hardware is called Blackwell that we've been ramping up now and shipping in volume. So just in one cycle from a previous generation to current generation, Blackwell, we’re 25 to 30 times more energy efficient for the same inference workload. So whatever your inference energy looked like a year, year and a half ago, on top of the line, most efficient hardware, if you run that same workload now, you're using 1/30 of the energy for that same workload, and that type of energy efficiency gain is something that we've seen very consistently over the past decade. It's actually 100,000 times more energy efficient than it was a decade ago. So it's hard to put your finger on that.

Talk us through jow you're achieving those kinds of efficiency gains? Are you now actively seeking to get as many flops per watt, whatever the term is? So it's not just about speed, it's about efficiency. How do you get those huge increases in efficiency?

It's definitely something that we're focused on. In fact, Jensen, our CEO, said a few weeks ago, performance per watt is our metric of success. So definitely there’s concerted efforts to achieve ongoing energy efficiency. And the reason for that is that's not only how we unlock the next level of performance, it also makes the platform much more valuable, because you're doing more with less energy. So it's really important, and it's something that market dynamics are driving, and we're focused on driving, and we're focused on. How we're doing it? It really is a multifaceted approach across the stack. Because we're a company that looks at GPU design, we look at data center architecture, we look at networking, we look at data center cooling, and then we look on the software side, at algorithms and efficiencies there… because we're looking across the stack, there are a ton of levers for us to pull to try to drive efficiencies. And when you combine them all, it really compounds to produce those very dramatic efficiency gains that you see. So it's really a mixture. I'll give you one example. When we moved from Hopper to Blackwell, so last generation to current generation platform, we transitioned from air cooled hardware to liquid cooled — direct to chip liquid cooled. And that's much, much more energy efficient, also much more water efficient. It's actually 300 times more water efficient. So that is one of the examples of very significant transitions that we've made that reduce our impacts and deliver more performance per watt and performance per liter of water.

Obviously, the pace of innovation is really fast in this sector, but how durable are the products you're making? How long do they last? And is there a lot of waste being generated here as they get replaced?

So on the waste issue, there's actually great news there, because what we're doing with our systems is creating much more compute out of a smaller and smaller physical package with each generation. So we're using a lot less resources to produce the same compute potential and compute capacity that we have in the past. So the waste is going down. In terms of the circularity, we're still seeing hardware from several generations ago being used very productively. Not for training a frontier AI model anymore, but maybe used for inference or for training small language models or other AI, or high performance computing workloads. So it's hard to know exactly what the ultimate circularity of the systems will be, because the hardware that we've been selling for the past several years is still being actively used. We do know that a lot of our customers are doing their best. Microsoft is a great example, they're trying to be as circular as they can with the hardware, recycling, repurposing wherever they can, and so they've got a lot of experience doing that. So if anybody can find a way to make sure that they're maximizing the recyclability and the reuse of the products, these companies are really well positioned to do it,

Some of the elements going into these products are rare, presumably, they're metals and some rare materials that there's a value if you can get them back out. But how much of the design can you design to be recyclable? Is that something you can work on?

There's some of that that you can do. But ultimately, the systems that we're designing are really cutting edge technologies, and performance generally dictates how they're designed. We are looking at opportunities where we can use recycled materials, but because these are so specialized, I don't want to overstate the potential for incorporating recycled content into new systems. But in terms of recovery of the materials, especially things like precious materials and critical minerals and so forth, rare earths. Those are things that our partners are very accustomed to doing, and they're very efficient at recovering those wherever possible.

And when it comes to siting your own infrastructure, are you seeking out the most energy efficient locations? Are you thinking about the carbon intensity of the grid that you're building into? Talk to me a bit about your decision making process.

So our place in the value chain is, we're generally providing the infrastructure to others who are building out their own data centers. So if you look at where most of our products are going, it's going to other customers. We do have some visibility into their siting decisions, and they've been very clear in general that access to energy is their number one concern when they're siting AI data centers nowadays. So energy is definitely at the top of mind in terms of siting. And if you look at the largest deployers of AI data centers, which includes the cloud service providers like Microsoft, Google, Meta, AWS, and then also look at the large co-location providers, so these are big companies that just create a ton of infrastructure and then they rent it out to others — like Equinix or Digital Realty — all of them have very ambitious clean energy goals and climate goals. And so the vast majority of the AI systems that are being deployed are being deployed by companies who are trying to pair them with clean energy and trying to minimize the emissions impact. So there's been a lot of press about potential impacts of operational use of AI, and we should absolutely be thinking about that and trying to manage those impacts. But a lot of people miss the fact that the vast majority of these systems are being deployed by companies who are doing their best to deploy them using clean energy and renewable resources.

And this is pretty a bit unfair, considering you have no control over it. But you know, there is a bit of a difference between say Elon Musk’s Grok, the huge park that he put up that's being powered by very poor efficiency open cycle gas turbines, it seems, or even diesel generators that are leading to quite a lot of environment impacts locally. Compared to maybe someone who's trying to do it correctly by putting it into a grid that's already clean, that's maybe got lower cooling demand. How much choice does a company have over these parameters? Are all companies under the same kind of pressure, or is it down to the choice of the CEO and that they're just going to go where they think is best for profitability, rather than sustainability?

So I would say there's a lot of discretion for companies to pursue whatever path they think is best for their company. We don't try to dictate that for our customers. Like I said, most of the infrastructure that we're building is being deployed by companies that are trying to use clean energy where they can. But we do need to give a little bit of grace to some of these companies, because they're put in a hard position where they're trying to build out infrastructure as quickly as they can. And at least in the near term, the energy supply is a little constrained, and so trying to build new infrastructure, bring new resources online as quickly as possible, so that we can unlock the next thing with AI and increase productivity and so forth. Sometimes companies are just scrambling to find whatever energy they can that's available. Now, the good news is that renewable technologies like solar are generally cheaper to deploy for new builds than anything else, so we're seeing a lot of new investment in that. Of course, there's new investment and talk about nuclear, which is a longer term solution, but it's something that's on the table now that that wasn't before. We're in a period of constraint where people are trying to find energy wherever they can. But I'm convinced that long term, because these companies are not changing their commitments, we'll end up marrying AI data centers more effectively with clean energy. And, in fact, not to get too lengthy here, but one of the cool things about AI data centers is that they're a little less sensitive to latency than traditional data centers. So with traditional data centers, like email servers or the company that's hosting your videos that you're streaming, you want those to be right next to the edge of the internet, so that when you ask for information, it comes back immediately. With AI, especially for training when you're just developing the model, it doesn't need to be anywhere near the internet. You could train your model in the desert, somewhere far away, or in the tundra in Canada, and then just bring it to wherever you need to use it. And even when you're using modern models, like reasoning models, they already have some delay built in, so the latency is not as much of a factor, which means that companies like Crusoe intentionally builds data centers where there is available capacity, especially clean energy, and so there's a lot more flexibility in siting a data centers than than there was for traditional data centers.

And it is important that we get this right. Because as you know, as we started the conversation, you said, ‘the sky's the limit.’ And most people are trying to project forward this demand curve. I think it is the IEA who said it could be that the demand is as high as the entire electricity consumption of Japan relatively quickly. There's a potentially massive increase here in demand coming at a time when the world is just teetering on the brink of pulling down global emissions from energy. We haven't managed it yet, partly because of demand growth. So at the margin, it's making the difference between are we continuing to grow the size of this problem? Or are we starting to finally see the curve coming down? This is the big debate, isn't it? Does a growing demand curve mean you see more innovation in clean faster, or is it just meaning more coal burn and gas burn, and we're just never going to catch up with ourselves?

Yeah, I think about this question a lot, and I'm happy to say that, looking at the data, I'm a very sincere optimist about where we're going to be in the medium term. So one thing to keep in mind is that AI, once we really take advantage of it, I think is going to end up being the best tool for sustainability that the world has ever seen largely because, not just because it's creating efficiencies, it's helping us save materials and so forth in certain circumstances, but because it's helping us innovate in ways that I think will lead to technological solutions to a lot of our sustainability challenges. So thinking of climate modeling, but more substantively, thinking about the efficiency of electric vehicles and material science, carbon capture and storage, unlocking cleaner energy sources. I mean, we're working to help facilitate the development of fusion power. And it's also important to put AI in context. So the IEA is forecasting strong growth for AI and the energy that's used by it, but AI is actually only the fifth largest driver of new load growth that the IEA is forecasting. So there are four other categories of energy growth that are more significant than AI. So it's not the lone cause of growth in energy, and it does have some very positive aspects to it. And then one other thing to consider is that a lot of the energy that's going to AI is going to new, efficient applications that are substitutions for existing energy consumption. So some data centers that are running old, less efficient software are being upgraded, basically, to run much more efficiently with accelerated servers that are boosted by GPUs. And we're also seeing more and more done in the digital world that saves us not only energy, but also materials in the real world, so running simulations on prototypes and so forth. Digital twins helps us to save resources before we actually need to go and build something and test it and spin all the cycles that we do, doing that. One other data point that I love to talk about is where you see the actual real world benefits of this. There's a manufacturing partner of ours called Foxconn, a very big manufacturer that does a lot of work for a bunch of companies. But they're building a manufacturing facility in Guadalajara, Mexico to manufacture some of our systems, and they used our technology and a digital twin to try to optimize that manufacturing facility for energy. And they're forecasting a 30% reduction in energy for that manufacturing facility across its lifetime because they used AI on the front end. So if we can keep doing that, if we can keep using AI to optimize energy in these other much larger energy sectors, then it can end up actually being a net positive for energy.

I like that example, but I can't help but think, yes, 30% more efficient is better than nothing, but it's still an additional load going on to what I presume is not yet a clean grid in a country which isn't the coolest of places. So I think we can always find all the positive examples, but some people have said, when we talk about this, the tendency is to just focus on the positive, right? Just the positive side of the ledger is the one we lead with. But that feels a little bit disingenuous. There's a negative side to the ledger as well. And it's not just, I suppose, the potential growth in demand at a time when we haven't quite got the technologies we need yet. It's also the social implications of AI. How do you square this, because there is a social element we're talking about as well as an environmental one.

Yeah, I'm glad you're hearing people focusing on the positive aspects. I hear a lot of people focusing on the negative.

What I'm doing is reflecting on your sustainability report, which we should talk about.

Absolutely, we should talk about it. We should bring the data, and we should try to figure out as best we can, what's the net impact likely to be. So I welcome it. So yeah, we'll get to the social thing. I do want to mention, though, that if you look at AI's contribution to things like energy demand and emissions… so the IEA estimated that all data centers used about 1.5% of global energy in 2024, so that's a tiny fraction of global energy. And AI is a small fraction of that.

1.5% today, at the moment, right?

Yeah, it's growing rapidly.

It is growing. And also, people just skip over the fact that cryptocurrencies have added probably that and more, and scarcely anyone is talking about that. And so I do feel that sometimes we can get over excited by this, but I think it's this thing that you said at the beginning, ‘ the sky's the limit.’ This just could keep growing and growing, and then clean energy really has to keep up, right? And that's the challenge.

Yeah, that's true. But at the end of the day, we need to think about, okay, say if AI grows significantly and it ends up being a full 1.5% of global energy, do we care enough about the benefits of AI to devote that much energy globally to that technology? Absolutely, we should manage the footprint. And if you run the numbers on the emissions the public numbers, I did this just a couple of weeks ago, including the IEAs numbers and other forecasts. A conservative ceiling for how much of global emissions come from AI is less than 0.1%. We should manage those impacts, of course, but I just want to put it in context in terms of potential benefits and costs. On the social question, this is something we should be thinking about and talking about, as well as the environmental impacts. At Nvidia, we've got a trustworthy AI team that focuses on this full time, to make sure that we're deploying AI responsibly. We really do want the world to get AI right, and that includes promoting sensible regulations that help, and promoting transparency. We have a model card of our own system to help promote transparency in models and so forth. So getting AI right is something we're deeply committed to, especially because of the potential social consequences. Again, I do think that AI has a lot of potential here, especially for things like education in the Global South. If we take these big LLMs, many of which are open sourced, and train them and utilize them for education, it's hard to think of a situation where there's more potential for economic development and education than with AI. It really is so democratizing. Actually, my teenage daughter was at a school that had a couple of awful teachers this past year, and we used AI very extensively to supplement her education, because she just wasn't getting it at school. And it really, really helped. It would have taken a ton more time and effort to try to find a way to supplement that education without it.

It comes back to your point about democratizing the language... I mean, these are large language models, and the ability to now go to a country, speak into your phone, have it auto translate. It's almost like the Babel fish that Douglas Adams believed was going to come. It does bring down barriers, but it'll be a function of how accessible it is to everybody, right? That and I guess there's a worry that the social impact is it will stratify society even further. Those who can will get better. And certainly in terms of employment, entry level jobs might be the first to go. So it's going to be quite hard to get onto the ladder. And then it might just consolidate. Rather than have countries leapfrogging, it might just consolidate power into the countries that have already got a head start. Do you think it's something that will feature in your sustainability reporting?

I’d love to feature more about our uses, Nvidia's own uses of AI for social sustainability, for social good. So I would expect us to report more on that. In the future, we'll continue to report on things like trustworthy AI, but they've got separate communications channels as well. It's definitely something that we're thinking about and working on. And I think we can and should promote it and talk about the risks as well as the benefits.

How do you see the role of government here, and you see the need for an AI regulator of some kind?

Well, that's probably a little bit out of my wheelhouse. We do have a government affairs team that focuses on policy issues. I mean, I can say Nvidia really believes that prudent regulation is useful and important to help get AI right. But there are many, many ways to get the regulation wrong, so we need to be very, very careful about it, especially during this period of rapid innovation. Because even if you come up with a fantastic regulation that fits today perfectly, it could end up creating really bad incentives in six months when people are using AI for different things, and the systems have completely changed. And even things like tracking energy efficiency of data centers, it's really hard to do that over an extended period of time now, because the innovation is just happening too rapidly. So one of the things that NVIDIA is trying to do is to be a technical resource to regulators in the UK. We've spoken with, actually we met with Ofcom a couple months ago. We talked to regulators in the EU, in the United States, across the world, to try to educate them, to help them understand how the technology works, how people are using it. Educate them about transparency and trustworthy AI, so that if they do want to regulate, they've at least got the right data to help them do it in an effective way.

I'm hoping that that's a dialog as well as a one way conversation, because there's also a lot that regulators have to take into consideration for society at large, which, whilst it's not the company's responsibility to do that, provides the context for why regulators are feeling under pressure. There's a dynamic there where a company has a very obvious bottom line to look after and a set of investors to look after, but politicians and regulators are trying to do that messy, complex job of keeping society on track and have the minimum amount of externalities. So as much as I'm really glad you're out there educating them, I hope you're also, I'm sure you are, as a sector listening to what society is sending back in terms of… You know, just as a parent, the proliferation of devices and distractions. And, yes, I love the fact that my son can use Duolingo and have a conversation in a foreign language. But I'm also very aware that he spends a lot of his time just watching crazy videos. It's something that as society, we're gonna have to grapple with, right? That this arrival is going to have a big impact.

Yeah, those are big issues. Absolutely, we should be working together in public- private partnerships to try to try to address those. And yeah, certainly part of our program is to listen as well as to educate. And hopefully we can come up with solutions that'll make AI as useful and non-harmful as possible.

To be fair to it as well, it is about surfacing those use cases that are perhaps hidden behind this barrage of infotainment that we're currently experiencing, and so surfacing things like your efforts to digital twin the earth, right? That's a big, bold idea that's enabled by access to this compute, and the cheapness of the compute now. So do you want to tell us a little bit about one of those big bet things that you're involved in, like Earth-2?

Sure, yeah, Earth-2 is really exciting. It's a significant commitment from Nvidia to try to facilitate climate modeling ultimately, and it's evolving. It's an ongoing project where we're trying to marry our infrastructure, especially our hardware, which is good for both high performance computing and also for AI, and trying to get that along with customized software tools into the hands of researchers so that they can make better use of existing data to make weather forecasts much more efficient, much more accurate and much quicker than they have in the past. The new system is at least 1,000 times more energy efficient, and I think 3,000 times faster than previous simulation only models. So it's really a step change in terms of what we can do. We're still in the early stages, because there's so much data out there, but not enough. We'd love to have more climate-related data, of course, atmospheric and oceanic, but the more data that we collect, and the more resources we devote to training these models, the closer we get to much more accurate weather forecasts and ultimately, climate. In fact, just a week ago, we released a new frontier model for generative AI for climate. It's called cBottle. We’ve published a little information about it, but we're definitely trying to contribute to that effort globally. To model the climate, which will help us, of course, take better actions and also help us with disaster response in the near term.

And this is so important. I mean, it feels like if you do need a counterpoint argument as to the benefits of AI, being able to properly harness it for those big, big challenges that we're facing into that are very disruptive, potentially, to the status quo for everyone — that's an understatement — for people to realize that there are these efforts underway and and the importance of that data, right? And the importance of data gathering/ I'm a little concerned that we don't do enough physical monitoring of the planet. You mentioned oceans. It's a kind of vast, huge part of the climate puzzle. And similarly, aerosols, the little tiny particles that are in the upper atmosphere, where it's so complex, understanding the ratio of sunlight that gets bounced back relative to absorbed. And the role of the oceans. I would love to see a kind of consortium of some of the world's biggest IT and tech companies saying, ‘Look, we can instrument the planet. We can do this now. We've got the hardware, we've got the software, we've got the capability.’ It's almost like a moonshot, to say planet Earth is actually the thing we really need to be focused on, let's instrument it, let's gather the data, and let's put the dashboard in place that allows us to see where the red lights are flashing, right? I feel like we don't have that at all. As policy makers, we get IPCC reports every five years, but they don't bring it to life in terms of what's actually happening as a planet, right? Where are we in this journey we're on.

Yeah, it's exceptionally complex. So you really need everybody at the table. You need the scientists, you need the governments, you need the tech companies. You need to bring everybody that you can, because it requires concerted effort amongst all of us.

Maybe this is a good time to ask you. You've just put out this sustainability report. Tell us a little bit about that, and some of the things you're proudest about, and some of the challenges you saw in that report. Then maybe we can think about where you'd like to be in a few years time.

I'll start with the one that I'm actually personally most proud of, even though it's not going to be the one that makes the headlines, and that's a new level of transparency around embodied emissions in our products. So we published a summary of a product carbon footprint that was third-party verified and ISO conformant, that looks at the emissions associated with the manufacturing of one of our accelerated servers, and we released that publicly. And this is the first, hopefully, of many that we're going to publish to try to provide more transparency, to help people understand, and again, make educated decisions about AI and about the net footprint and the handprint as well, so that we can be very rational and sensible about how we strike those balances. Also really excited about science-based targets we set. That was another big achievement. Happy to finally have that on the books. Now we're decarbonizing and working with our suppliers to do that as well, and those commitments are in our sustainability report. But excited to have those, those public and then the last thing I'll mention as a highlight is we achieved our goal of transitioning to 100% renewable energy last year. So you know, for a company like Nvidia, we don't have the largest energy footprint in the world, but we wanted to make sure that we were doing what we could and setting an example for others in our value chain of what we expect to see as we all work towards decarbonisation.

I’m interested in that point. As a kind of contribution to the problem, your electricity demand is not the big thing, it's your consumers, really your customers. But you've chosen to use a kind of certificate-based compliance, right? You're not saying you're purchasing renewables to match every single electron you use, right? You're using certificates, is that right?

So it's a mix. And yeah, our strategy is to adopt more and more PPAs long term, to try to have more and more additionality. And we're already in negotiation phases for a lot of that to cover our load. But because our footprint is so distributed, and we have a lot of rental offices and so forth, it's hard for us to manage PPAs for a ton of, R&D offices and smaller offices throughout the world. So I think certificates do have a role to play in kind of decarbonization for circumstances like these. But we are in consensus in terms of trying to transition to long term instruments like PPAs that have more additionality and will help green the grid more broadly.

It feels like that's where this demand curve can really drive innovation is if you're able to sign, as a number of your customers have done, with people who are trying to get through the valley of death in bringing commercial products, getting that first of a kind demonstration project built. If companies like yourselves and your customers can sign those PPAs and buy forward those technologies, that's when I think you get that marrying up of demand growth that equals a great new entry opportunity for new technology. Which is definitely going to be a more bankable way than maybe just a certificate-based system which doesn't provide that guarantee for investors.

Yeah, and it's a great point. I think this is a perfect crisis — they say, don't let a crisis go to waste — a perfect crisis where, because of the value of AI that the world is seeing — governments as well as companies and individuals. and because of the urgency for companies that are deploying these AI data centers to get their hands on energy and those companies generally having strong commitments to sustainability, we're seeing, I think, unprecedented demand for clean energy globally, specifically because of AI. So the other demands coming from buildings and electrification and manufacturing might not necessarily be as clean as what's being deployed and sought for AI data centers. So we're at this juncture where we need more energy. Tech companies are some of the ones that are coming and asking for more energy, also with strong commitments to sustainability, and they're the ones who are actually best positioned to help modernize the grid and to advance clean technologies for us to have a greener grid going forward. So it's really is a perfect storm for us to modernize the grid and green the grid at the same time, because of the parties that are involved in the load growth right now.

And adding to that, your point earlier about AI being involved in the design of these new technologies, so you might be bringing down the kind of R&D timelines for some of these new breakthroughs. And I guess optimizing grids, it's a nice, complex problem, which will be enhanced by us having more intelligence and more data. So, yeah, I can see a virtuous circle. But I can also see, if it doesn't live up to expectations, we don't get those technologies built, it could just be this massive boom in energy demand for us to not get more intelligent, but just get more distracted. That's what I worry about.

I hear you, but I think ultimately, again, we're in this period of chaotic innovation. But as much as we're driving the cost of computing and AI down, and that's a goal, it's never going to be free. So in the medium term, AI is going to be deployed where it's actually useful and where there's value created. So, yes, it's chaotic now, there is an energy constraint right now, it's driving more investment, but we will reach an equilibrium where we're deploying AI because it's useful, because it's creating value for individuals, for companies, for the economy, for sustainability, and that will be captured once we get there.

I've heard people describe these waves where you got these innovation crests where everyone's suddenly involved in chaos, and a lot of overvaluation, and then there will come an inevitable correction, when it all falls out and the tourists leave. But you'll then be left with the real job of getting the job done. I don't know where we are on that arc, but perhaps we're in the chaotic upcycle right now, but ultimately it will shake out. And I guess what you're saying is, there'll be an equilibrium that’s reached, and we'll see the real value of what's being done. I guess for me, for society's purposes, I think we just need to put in some good measurements. For me, any form of social regulation, it's got to be based on data of knowing what the impacts are, right? So perhaps we need to just collectively think about what are the parameters we should be measuring, and AI could probably help us measure those parameters. Let's do this evolution based on data and not sentiment, which can also sometimes just creep into this conversation.

Yeah, I love that. I think that's what we need. We need to have the data there, because it's so easy to be too much of an optimist or too much of a pessimist. We've seen a lot of both of those in the climate space in particular. So bringing the data is critical, and that's one of the reasons why we're increasing our transparency at Nvidia, to try to help people make informed decisions so that we can achieve that optimal outcome.

Well, thank you for bringing that transparency, and we look forward to future sustainability reports and to, I hope, people getting more comfortable and more aware of all of the various ways in which AI is being deployed. So seeing through the more shiny, distracting versions and then big moonshot plans like Earth-2 and where Earth-2 could go, I think will help with people feeling like this is not just about the risks and the downsides, but there's this big upside that could be very, very positive and very optimistic for our ability to get through this challenging period we're in.

We're in a period of chaos right now, but there's so much potential, and I think the data is pointing to the fact that we're going to get to a good place with this huge potential, especially for moonshots. But yeah, we need to keep talking, keep measuring those footprints, as well as the handprints, and do our best.

Great. Well, thank you, Josh. I'm really grateful for you to spend your time with us today. I'm sure this is a topic we'll keep coming back to, and we look forward to reading more about Nvidia's efforts in this space. So thank you.

My pleasure. Thank you.

So that was Josh Parker, Nvidia's head of sustainability. Whether our new found artificial intelligence powers end up being net positive for humanity feels like still a big question. Josh, as he says, is a natural optimist, and rising demand for electricity could lead to a much better investment climate for a range of clean technologies. But if a huge boom in the use of these new tools leads us to never being able to bend the curve of global emissions, then we may look back with regret that we were too focused on emojis and short-form videos and not enough on monitoring what we're doing to our planet and working out what to do about it. If there truly is a lot more going on in the background than we may perceive, then we should start to see some benefits. And if there isn't, then to avoid a planetary bust, it's incumbent on everyone benefiting from this boom time to try to correct the balance. We'll add relevant links in the show notes, including to Nvidia's sustainability report and Michael Liebriech’s recent article about the potential AI impact on energy demand. My thanks, as always, to Oscar Boyd our producer, Jamie Oliver our video editor, and to the rest of the Cleaning Up team and wonderful Leadership Circle members who make the Cleaning Up podcast possible. We hope you enjoyed the conversation, and please join us at the same time next week for another episode of Cleaning Up.