YouSaid · the spoken record

Justine Kalma

lines on the record: 37
first: 2024-08-01
most recent: 2024-08-01
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“What I love to see nuclear fusion, I'd be super excited about that. A lot of the folks that I've talked to who are even skeptical about it happening anytime soon say the same thing. But it's generally not really thought of as a solution for the most pressing issues we're facing today.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“So nuclear energy is a heavy topic. I mean, that's really controversial. On the one hand, you have some environmental advocates that say that it's the biggest single source of carbon dioxide-free energy that we have in the US today. It also doesn't fluctuate with the weather and time of day like solar and wind do in hydro. And so it's seen as something that can provide a stable source of carbon dioxide-free energy and the biden administration has been really all on board with nuclear energy as well as a way to hit its climate goals. You also have on the other side of the issue, you have a lot of environmental justice advocates that are really wary and skeptical of nuclear energy because it still comes with nuclear waste issues, right? And there are concerns about other environmental impacts from things like mining, uranium, which has left the toxic legacy.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“One other big lever to push here, and that's cost. Energy isn't cheap. And the more constrained the supply gets, the more expensive it will be to use. There's one thing all of these companies don't love to do. It's spending more money than they have to. Money itself is a pretty big incentive for the companies investing in AI to also invest in making that AI as energy efficient as possible in the long run. Of course, the solution everyone would prefer would be to magically develop an abundant source of cheap energy. Inside the Silicon Valley VC set, the crypto boosters have all started talking about nuclear power as a solution. They started proposing we build lots more nuclear power plants. Is that a reasonable idea? Is that something that's on the horizon in the United States?”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“I don't know that regulators are really trying to single out any particular industry just because of their whatever moral sense is on that industry. I think really what it comes down to is making sure that there are science-based targets to hit. And honestly, even with just what we have right now, even just with EVs and electrifying homes, that's already going to be a huge difficult lift that power grids and policymakers are struggling to implement. And so again, we're talking about a system that's really teetering here. And that's where a lot of the concern comes when there are these big unexpected factors that add more pressure to an already, really, really strained system and energy transition.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Pressure for large scale change, notoriously needs to be top down, not bottom up, because consumers only have so much power. Regulators and governments in theory have a lot more. But they also have incentive to think more broadly than just one sector.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Try to walk this line with consumers because what we do individually as consumers, as gamers, as whatever, is important, but I don't think we should let regulators and big companies off the hook because what folks are really concerned about are these giant new hyperscale data centers, right? Not necessarily what you're doing in your home. The bigger impact is what's happening on that much larger scale.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Scale really matters. Sure, your energy use at home will go up if you start charging a new EV, or if you have a high powered GPU and a gaming PC and you play games for 10 hours a day. No judgment. And AI will start adding to that use too in subtle ways. For example, Apple is promising that most of its upcoming AI features will run locally on their phones, but the laws of physics still apply and all that extra computation will still use energy. It will just be the energy in your phone's battery that you might have to charge more often. It's hard to know how all of this will play out. Those features haven't even shipped yet. And it's hard for us as individual consumers to evaluate our choices at home compared to the massive scale of the problems out there in the world.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Energy use. For example, the EV transition is broadly considered to be good, even though all those cars have big batteries that need charging, which will add up to more strain on the grid. Anti-EV activists, including Donald Trump, bring this up all the time. But the arguments don't have the same impact.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Ever we cover the subject. Artists, writers, and other creatives are mad about having their work used as training data and their filing copyright infringement lawsuits left and right. People who care about democracy and elections are really worried about deepfakes, a problem which is being exacerbated by the very leaders of the tech industry on their own platforms like X. A lot of people are deeply concerned about losing their jobs. And executives and some of the major AI companies have spoken pretty callously about that idea. There's just a long list of moral objections you can raise about AI. And a lot of people who do raise objections feel pretty powerless in the face of systems built by billionaires. And because climate concerns are inherently existential, they can feel like a powerful argument against a technology that might cause a lot of other disruptions along the way. So it's useful to compare how we talk about AI energy use and its climate impact with how we talk about other big energy transitions to see how different the conversation is. Because as a society, we've assigned very different moral values to different kinds of energy.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“complicated because AI is a pretty bubbly industry. The energy use estimates we're talking about are based on things like NVIDIA's own sales projections on the market saying the line will just keep going up forever. If that's true, we'll then figuring out how much energy will be required to support AI as a grows is a huge challenge. But is that line going to keep going up forever? If you read some of the emails from our audience, it definitely won't. But NVIDIA and the rest of the industry are betting that everything will be successful and all those AI server farms will keep running at 100% forever. And the answer is somewhere in the middle. We just don't know. And that brings us to an even more complicated problem, which is the moral question. You can gauge financial value pretty easily. I mean, you can look at something like monthly active users as a metric for success. But what good is AI to me or to humanity is a lot less quantifiable. And we're going to be arguing about it for a long time. There are a lot of moral objections to AI. We hear them every day at the Verge.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“I think right now, again, a lot of the research has been focused on the energy consumption of training AI. There are really big questions about actually using all of these AI tools, the energy intensity of the inference part of AI. And that's where we're still seeing research kind of starting to come out now. And there are questions about how that might materialize. The question that a lot of energy experts and researchers are posing is, what do we want to use AI for and what do we see as something that's worth the energy and environmental costs and what is potentially something that maybe is just not worth it?”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Especially Bitcoin mining, is that it's guaranteed to use 100% of the computers you throw at it since the underlying problems are designed to get harder to solve over time. The more computers you throw at Bitcoin, the harder they all have to work, and you can basically predict their usage in a straight line. But we don't actually know if anyone's going to be using AI or if they even will at a meaningful scale. We can't really quantify how correct or incorrect our projections are because we don't really know what the value of AI is going to be for either individuals or businesses over the long run. We might be stuffing H100s and data centers to create an enormous supply of AI compute power that there isn't enough demand for. And that's really affecting the quality of the research and the estimates we can get.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Welcome back. I'm talking to the Virge Senior Science Reporter Justine Kama about the incredibly messy collision of climate goals and our reliance in the cloud, especially for AI. Right before the break, Justine asked a really important question. What are we using these AI tools for? And are those uses worth the risk? And that's where we really start to feel the tension around AI and energy. We all understand there's value in being able to watch Netflix in your living room or talking to your co-workers or friends on Zoom or FaceTime. But when we ask about AI's value, we're really talking about two different things. One is its financial and economic value. Big tech is all but tripping over its own feet to invest in AI as fast as it can. But so far, they're not getting any returns. Every time I've talked to a CEO about AI and Dakota, they all say the same thing. The AI investments are for the long term. Everyone's just making a huge bet. They hope will pay off in the long run. It might be that all those H100s just sit idle. This is where the comparison to crypto gets really complicated. An unusual aspect of crypto mining.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“That is a big part of the question is just what are we using these AI tools for, and is it seen as something that is worth the risk? But I do think there are some other factors at play here too. Where we are now, as we're seeing the world just had four of the hottest days on record just in the past week. We're seeing more intense heat waves, storms, fires. We're running really close to not being able to meet a lot of climate goals to keep those things from growing much worse. And so right now it's just that the pressure is on. And we are really supposed to be using less energy. That is a big piece of the puzzle too, is trying to be more strategic about our energy use and figure out how we can use energy much more efficiently.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“The complexity for me. I can't think of a point in modern human history, especially the history of humans with computers where energy usage has gone down. It only ever goes up. That's what we do here. And it feels like AI might accelerate that. It feels like crypto definitely accelerated how much energy we're using to power computers. But underneath it is just a linear increase. And we're not getting too mad about Netflix every so often re-encoding its entire library, which uses a lot of power, runs a lot of computers at 100%. Gmail just gets bigger. Google's not making Gmail smaller. YouTube, we just keep adding an infinite amount of video to YouTube every single day. Our cell networks are getting bigger and we're putting more computers on them every day. There's that and then there's AI and crypto. And it feels like the difference to me is, yes, maybe they're going faster. But we as a society have not assigned them a value that evens out the amount of usage that they take up. I feel like we could get really mad about how much power YouTube runs every day.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“That as well as what happens if there is an extreme heat wave and you have the need to power a lot of people's air conditioning, but also all of these new data centers and you don't have enough electricity supply during and demand peak. I've definitely seen a lot more concern about what's going on with AI and crypto. Data centers have been energy hogs for a long time, right? And these companies have done a lot to say that they're addressing that. They've looked for more efficient ways to cool down these data centers because cooling is a huge electricity cost. They've tried to make more efficient chips. I think right now it's just that we're seeing AI throw our wrench into those kind of best”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Location matters. And so you're seeing a lot of concentration of data centers being built in, say, Virginia, for example, or for crypto mining in Texas. So when you have a lot of data centers that are pulling electricity from the same local power grid, that has a much bigger impact than diffuse across the US, right? So you do run into more issues if all of this is going down in one location you have a lot more stress on that local power grid you have that utility potentially looking to extend the life or bring on more fossil fuel power plants and then also facing more energy crunches when it's trying to balance the energy use for those data centers with what other businesses and residents need in their day-to-day lives we've seen in some places folks electricity bills go up a little bit you also see concerns again about building out new fossil fuel plants more pollution from”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“In a data center rack is drawing a comparable amount of power to a consumer grade NVIDIA 4090 in a gaming PC in someone's living room. But the H100 in the data center is sitting next to hundreds of other H100s, and that's where it starts to become a problem. On top of that, data centers themselves tend to be clustered in certain regions, or various financial and logistical reasons. That means all the energy usage isn't distributed evenly around the country. Instead, it's putting a deep strain on certain parts of the grid.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“The companies have set these big sustainable goals have also absolutely whiffed those goals as they build out more and more data centers. Microsoft's emissions, for example, were 30% higher in 2023 than they were in 2020 when the company made its first big climate pledge. Amazon is pumping out 34% more emissions than it made its sustainability pledge also in 2019. And Google's 2023 emissions were up by 48% compared to 2019. So these companies are just going for it. There are not going to be fewer data centers anytime soon. And the hard part is that they're not just for AI. Pretty much everything we do these days happens in the cloud. And the cloud is really just someone else's computer. If you're watching video on Netflix or YouTube or listening to music on Spotify, talking to someone on Zoom or playing an online multiplayer game, it's all using some amount of computing power in a data center somewhere. But the where specifically is really important. When you get right down to it in Nvidia H100.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Bring more renewable energy capacity online. But yeah, generally, you're not seeing a data center that is just running off of solar power on its rooftop. It's important to say that these are part of broader climate goals. So you have under the Paris Agreement a commitment to stop global warming from moving beyond roughly 1.5 degrees Celsius above where they were before the Industrial Revolution. And that is really meant to save homes. That's meant to save lives. That's meant to save species. You know, if you move beyond that, you're looking at things like losing 99% of the world's coral reefs, losing a lot of coastlines to sea level rise. And so the goals that they've set aren't out of nowhere. They're meant to help get us these science facts target.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“A lot of these companies have made really big climate pledges, particularly to reach net zero or net negative carbon dioxide emissions. And a big part of that is to be able to run on renewable electricity for them to say that they are running on renewable energy. Again, if they're plugging into the grid, they don't know where that electricity is coming from. All the electrons get jumbled up right in the grid. We don't know if it's coming from gas-fired power plant or if it's coming from a solar farm. And so typically companies will often buy renewable energy certificates to claim the environmental benefits of generating renewable energy. And they buy enough of those to match their renewable electricity use. That's generally how a lot of companies are meeting their sustainability goals. They're also trying to enter into power purchase agreements to facilitate the building of new solar wind farms.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“When you hear a big company say we're going to have 100% renewable data center, is it actually they've got enough solar panels on the roof to run the data center? Are they just trading credits back and forth? Are they just making these power purchase agreements? Or is it adding up and they get to claim it? Who audits that?”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Right now, the grid in the US and globally power grids are still running on a lot of fossil fuels. In the US, it's still a majority fossil fuels gas in particular. And so while we do see solar and wind energy ramping up, it's not ramping up at a pace that right now is going to get the U.S. to meet its climate goals under the Paris Agreement. We honestly just don't have enough renewable energy on the grid right now. When you plug into the grid, you don't really know where that electricity is coming from, whether it's coming from a gas-fired power plant or from a solar farm. That makes it hard to make sustainability claims.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Take a short break. When we come back, we'll dig into where all that energy in the grid comes from. And if big tech companies are keeping their promises for how to deal with it. Welcome back. I'm talking with Verge, Senior Science Reporter Justine Kama about how energy use and climate commitments are colliding with the growth in generative AI. So we've talked about the grid a little bit and how there's not a lot of storage in it. Supply is constrained, demand is going up. The big companies like Amazon, Microsoft, and Google, which are building all of these data centers, love to talk a big game about how they're using renewable or sustainable energy and reducing their emissions and fossil fuel usage. What kind of power are most data centers using right now?”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“I think it's important to understand the broader context here of why we're sweating over small single digit percentages of electricity demand increase. And that part of the reason why is because our power grid is already operating on a razor blade. You're constantly trying to generate just enough electricity to meet demand to avoid power outages, right? We also don't have a whole lot of energy storage on the grid to safeguard from an energy crisis that we're starting to see more of whenever we see these stronger storms or heat waves that are putting a lot more pressure on the grid as well. We also have really not a lot of time to meet climate goals that have been set at the global level for the US and also that many of these tech companies have set for themselves. And so when we're talking about trying to thread the needle, there's a lot at stake. That's where people start to sweat over numbers like this.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Are some pretty obvious examples of this, like the blackouts or brownouts you see when there's a severe cold snap or heat wave in some local part of the grid just can't keep up with heating or cooling demand. All of our data centers basically just plug into the grid the same way that any house or store office building might. And there's just not a lot of wiggle room left for all that extra demand.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Put that into some context. Here in the United States, there's an entire federal agency, the Energy Information Administration, that exists to gather, analyze, and publish information about energy usage in the United States. They publish all the reports you could ever want about how much electricity gets used in this country, how it's made, and where it goes. In 2022, the EIA says that total energy production in the United States is about 4,180 terawatt hours, and total energy consumption was 4,070 terawatt hours. Those numbers are very close, and the amount we use is going up faster than the amount we produce, which leaves very little margin for error. This is what people mean when they worry about the grid. Companies make power plants of whatever kind, gas, solar, nuclear, hydroelectric, and so on, and then all of that electricity gets transmitted through a bunch of systems up until the last mile utility cables on your street bring it to your house. The grid”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Looking at a few different things when it comes to a data center, they're looking at the power that the machines are actually using. So that accounts for about 40% of the electricity demand of a data center. Cooling requirements are another 40%. And then the remaining 20% comes from other associated IT equipment. And then they do put an estimate, a lower band and a higher band in terms of how much electricity data centers could use in the future. And so that's where you get a couple different estimates in terms of what that might look like. So upper band, the most would be equivalent to adding at least the electricity demand equivalent of Germany that's on the larger end, an additional 590 terawatt hours of electricity demand. And then on the lower end would be around 160 terawatt hours of electricity demand, which would be more in line with the electricity that a country like Sweden would use.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“That's going from the International Energy Agency, which is an international organization that was started back in the 1970s to monitor the world's energy supply use. It was founded in the wake of the 1970s oil crisis. And so they're really concerned with making sure that we have enough energy to make the world run. Lately, they've also done a lot of tracking around the clean energy transition as well.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Data centers have been around for a long time, but I think what is worrying some energy analysts now is where there's a lot of fast growth, which is outpacing both the capabilities of our aging grid, as well as our abilities to try to meet climate goals that have been set in the US and globally. So data centers are expected to use about twice as much energy, electricity in particular by 2026 as they did last year. And AI in particular could use about 10 times as much electricity in 2026 as it did last year. So that's huge growth in demand.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“When you say data center, does that encompass everything from Amazon's big fancy AWS data center or Microsoft's big fancy Azure data center all the way down to a guy in Texas who rented out a barn and stuffed it full of crypto ASICs?”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“It is. I mean, we've seen data centers become more energy efficient, but we've also seen their energy use expand since 2000, and that's expected to grow a whole lot more in the coming years, particularly with data centers that are used for crypto mining and for AI.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“There are about 8,000 data centers around the world as of the latest estimates from the International Energy Agency from this year. And about a third of those are in the US. So the US has more data centers than any other country in the world. We know that globally data centers use about 2% of electricity. Again, that's for those 8,000 data centers globally. The US has about a third of those data centers that use roughly 4% of US electricity. It's a small but meaningful chunk”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“A lot going on here, and we're going to get into a lot of the moving parts, but I want to start with some really basic facts just to set the stage, because we can't talk about a problem getting worse or not getting worse. If we can't name it to begin with. Are a lot of data centers in the United States. There are more of them every day. How much power are they using right now? Is that a reasonable baseline to set?”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“In terms of power draw. What framework should we use to evaluate the climate impact of those cards and how we feel about how they're used? Messy and it's complicated, and there are a lot of apparent contradictions along the way. So, really, it's perfect for Dakota. To help sort it out, I've invited Verge Senior Science reporter Justine Kalma on the show to see if we can untangle this knot. Welcome to Dakota.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“Bad instead of catastrophic. But now with AI, all of them are blowing past their emissions targets, and they're actually getting worse over time instead of better. It's not great. But putting a bunch of computers in a data center and running them at full tilt is how basically everything works now. If you have a moral objection to AI based on climate concerns, you might also have a moral objection to TikTok and YouTube, which are constantly ingesting and encoding millions of hours of video. You might have a moral objection to video games, which both run on Power Hungry GPUs in people's homes and often also require intense data center workloads for online multiplayer. And I'll take a guess, but I feel pretty certain that anyone with climate concerns about AI has a pretty harsh assessment of crypto as well. I mean, think about it this way. The NVIDIA H100, which is the gold standard for AI GPUs, is pretty similar to the gaming-focused NVIDIA RTX490.”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT
“And welcome to Decoder. I'm Nilai Patel, editor in chief of the Verge, and Decoder is my show about big ideas and other problems. We've been talking about AI a lot on the show lately. It's unavoidable. And there's one piece of feedback we keep getting that I really wanted to spend some time on, how the lightning fast explosion of AI affects the climate. After all, to run AI at scale, we need to build a lot of data centers and pack them full of power-hungry GPUs. That takes a lot of energy. And whether using all that juice is worth it comes up frequently whenever we talk about AI, both as a matter of practical concern, given our aging grid support all of this, and as a moral objection, we shouldn't build these systems because they'll wreck the planet. And what's particularly complicated is that big tech companies like Amazon and Google and Microsoft have also spent the last few years working with governments around the world to set ambitious goals around sustainable energy usage so that we might slow the rate of climate change to simply”
2024-08-01 · Decoder with Nilay Patel · AI has a climate problem — but so does all of tech · IDENTIFIED FROM THE TRANSCRIPT