"Hyperscale": Paris Marx on Excess of Big Tech Data Empires & Bernie Sanders' AI Bill

[00:00:00] Speaker 1: This is Democracy Now! I'm Amy Goodman with Nermeen Shaikh. [00:00:03] Speaker 2: With the midterms less than three weeks away, we continue to look at how AI data centers have emerged as a political flashpoint. [00:00:10] Speaker 1: We're joined now by the Canadian journalist Paris Marks, host of the award-winning Tech Won't Save Us podcast, writes the Disconnect newsletter, author of the new book Hyperscale, The Ambition and Excess of Big Tech's Data Empires. He's joining us from St. John's, Newfoundland, and we just talked to Justin Pearson, taking on Elon Musk's XAI data centers in Memphis. In fact, though, the largest concentration of data centers is in northern Virginia, as you write about. And Nermeen and I are speaking to you from New York, which has the only state moratorium on the building of data centers. Can you give us, can you put that battle in Memphis in the global picture? Why northern Virginia, for example? Is there this massive concentration of data centers? It's not one of the poorest areas in the country. [00:01:04] Speaker 3: No, certainly not. It's much more for historical reasons, right? Because a lot of that early work on the Internet was happening in northern Virginia. You know, obviously a lot of military spending that went into that. And of course, for that reason, a lot of infrastructure, a lot of, you know, fiber infrastructure, connectivity infrastructure, data centers, certainly over time, were built out in that area. And that, you know, kind of that legacy continues to this day. And so, you know, the people who live there in northern Virginia have really felt the effects of all of these data centers having been built out. And especially, you know, they have had data centers for a long time, right? But one of the things I try to get to in the book is really the shift from, you know, data centers as they used to be, you know, in industrial areas, maybe not so much larger than other tenants that would be in those, you know, kind of industrial parks and things like that, to the hyperscale data centers that are just on a whole other scale. You know, the resource demands are just immense, as, of course, you know, Representative Justin Pearson was just talking about in Memphis, where they have really been feeling this firsthand. And I think, you know, because Elon Musk was involved, this really brought it to the attention of a lot more people, you know, seeing the effects that come with one of these major data centers. But obviously, in northern Virginia, people are feeling a lot of that. But as we have seen this significant escalation in the construction of these hyperscale data centers, particularly in the past four to five years, and certainly since the generative AI boom, you know, really started to take off in 2023, we have seen not just many more communities in the United States seeing data centers planned or being constructed next to them and feeling, you know, what comes with that, whether that is the increases in their electricity prices, the pressures on their water system, or just, you know, kind of the noise and light pollution that can come from living next to one of these infrastructures. But many more communities around the world are feeling that, too. And I think that's part of the reason why, you know, we're seeing this kind of collective backlash, you know, throughout the United States, but in many other parts of the world against these data centers. [00:03:04] Speaker 2: So, Paris, let's just, you know, to give people a sense of the scale of these data centers, there are 3,000, more than 3,000 operational data centers in the U.S. now, and about 1,500 that are being planned. So, why don't you explain, if you could, why? Why so many and why so quickly? [00:03:26] Speaker 3: Yeah, absolutely. So, you know, we need data centers to process the types of things that we do online, you know, the digital services that we use. You know, of course, I'm connected to you digitally. That is likely going through a data center somewhere, right? So, it's not to say that we need absolutely no data centers, we need to get rid of them all. I think there have certainly been benefits from the digital technologies that have arisen over the past few decades. But we have seen this continued escalation in the amount of computing power that the tech companies in the tech industry are demanding, right? And they make the products that they're offering to people more computationally intensive. And so, the real reason, you know, one of the key reasons, at least, that we have seen this significant escalation over the past few years in the construction of data centers, because we really saw this start to take off in 2023, 2024, which is why it has become such a political flashpoint over the past 12 to 18 months, is really because of generative AI, right? Because of the chatbots and image generators that companies like OpenAI, Anthropic, but also Google, Microsoft, Amazon have been pushing for the past few years, because, you know, for regular people who are not really into the technical side of things, it might just be like, oh, okay, this is a new service that these companies are offering. But the reality is that these are much more computationally intensive than what, you know, these companies were offering before. It takes a lot more computation to kind of create, you know, kind of a text response to a prompt that you're putting in, or to generate an image or a video based on some prompts that you're putting into the system. And that has meant they have needed to build so many more data centers so much more quickly and make those data centers even larger than they have in the past, which, again, is why so many people are, you know, seeing data centers arrive in their communities, just because these tech companies are looking for so many places to be able to locate them, places that are willing to, you know, have them become their neighbors. And they're finding that increasingly, a lot of communities are saying no to that. [00:05:22] Speaker 2: Well, one of the chapters in your book, Paris, is devoted to the climate impacts, which we heard about earlier from Justin in Memphis. So the question, you know, obviously, America, the U.S. has exponentially more data centers than any country in the world. Now, China, which is also building, you know, massively data centers, though, again, nowhere near the number of what the U.S. has, they have been attempting to employ renewable energy to diminish the climate impacts of these data centers. In your research, do you see any indication that these tech companies are looking for more sustainable sources of energy? [00:06:08] Speaker 3: Not in any significant way, right? Like, if you listen to what the tech companies say, they will certainly tell you that they're investing a lot of money in renewable energy. You know, in the past, you know, last year in particular, they made a big push into nuclear energy, which, you know, obviously is very debatable whether it can be considered clean. But in their ledger, they would claim it is, right? Which is part of their way to say, like, you know, look, we're trying to do this, you know, efficiently. We're trying to do this sustainably. And, you know, they will certainly talk about how, you know, they have, they will reduce kind of the intensity of water use at a data center, right? So, you know, kind of the total quantity that a data center itself will use. But then when you look across the total number of data centers that they're building out, their water use and their electricity use as well is just absolutely soaring. But the real story here, if you think about the climate impact, is really their dependence on fossil fuels and their dependence on natural gas in particular. And so, you know, in the kind of 23, 24, we saw this real push for them to try to get connections to the grid in the various places where they were building out these data centers, right? But in many of those places, they have now exhausted those connections, right? The demand on the grid was so great that it became much more difficult to connect a data center in the kind of amount of time that they would need to connect it for it to come online to, you know, benefit their generative AI ambitions. And so what we have seen over the past year is a real pivot toward building on-site gas plants. You know, obviously, when we look at Memphis, there was the use of a lot of generators on-site in order to power the data center. And we have seen other data centers using that model as well. But now, of course, increasingly, they're moving toward actually just building dedicated gas plants for the data centers, such that there was a report out recently that it's something like 189 gigawatts of new gas capacity is in the pipeline in the United States just to power these data centers, which is just staggering to imagine, right? You know, we have already seen our governments roll back on their climate ambitions. We have seen the tech companies that back in 2019, 2020, wanted us to see them as climate leaders and, you know, as though they cared about climate change and taking action on that, completely throw their targets out the window in order to pursue generative AI and this data center build out, regardless of the consequences that come of that. And, of course, you know, those are consequences that will be felt by everybody through, you know, how it's going to make climate change worse. But, again, as Representative Pearson was saying, we also see very clearly the local impacts of local air pollution, you know, on the people who live next to these data centers, as the people in Memphis are feeling, but also in many other communities where this is happening. [00:08:50] Speaker 1: Paris Marks, I want to ask you about your critique of the legislation Senator Bernie Sanders is putting forward, certainly seen as an AI critic, a deep concern about what's happening and what's happening so fast. Senator Sanders and Congressman Greg Kassar announced the Ban Artificial Superintelligence Act to stop AI oligarchs from building machines humans cannot control. It would permanently ban the development and deployment of superintelligent AI and temporarily pause advanced AI development until a federal regulator has established safety rules. They would also direct the U.S. to pursue international agreements to prevent superintelligence from being developed anywhere in the world. You are critical of this approach. [00:09:48] Speaker 3: Yeah, absolutely. You know, I certainly welcome Senator Sanders' work on the data center moratorium and trying to move that forward on a national level. I think that's very important. And, you know, even when we look at the kind of superintelligence legislation and, you know, the discussion of, you know, kind of limiting or rolling back the further development of these AI models, I would also be on board with that as well. But probably not for the same reasons that are being laid out in this legislation. And so, to me, this feels like a response to these narratives that we have had for the past couple months now about, you know, the power and the threat that these AI models compose to humanity, right? You know, we have seen these tech companies and people who work at them come out and talk about how they feel or they believe and, you know, they want us to believe as well that, you know, these models are getting to a place that they are starting to learn on their own. And they're starting to become so powerful that within 10 years, they could decide that they want to wipe out humanity or a significant chunk of humanity. And that is a real threat to, you know, our collective species, right? And to me, quite frankly, I feel that that is a mix of science fiction and delusion on the part of the people who are running these companies. And it is also a very effective way to distract from the real harms that we're seeing from these generative AI products, from the chatbots and image generators today. You know, for me, I think that, you know, kind of this future focused stuff is an attempt to make sure that regulation is not targeting the things that people are actually feeling and that would actually restrict what the chatbots and the AI companies are doing right now. And, you know, rather having the discussion being about something that might happen 10 years down the line that I would argue is actually very unlikely and doesn't reflect how the technology really works. You know, what I would like to see from a regulatory perspective is not this focus on the super intelligent stuff, but, you know, the effects that, you know, as we're seeing more reports, the effects that dependence on chatbots can have on people's mental health, where we see them have mental health spirals because they have become dependent on these technologies. You know, the kind of deepfake crisis that we have seen and really saw very clearly on Elon Musk's platform and with the Grok chatbot at the beginning of the year. The questions about what dependence on chatbots and use of chatbots means for people's ability to learn, to think critically, to kind of have these cognitive skills and to develop them properly. You know, we can even go beyond that to the environmental impact of the data centers themselves, or, you know, the various other ways that we have seen chatbots use for people to try to harm themselves or to harm others, of course, in the case that we're seeing here in Canada, where the shooter out in British Columbia earlier this year, in one of the worst mass shootings in our history, was actually using ChatGPT for eight months to help plan violent scenarios and that attack in particular. Right up until the shooter, right up until the morning of the attack, when the shooter asked the chatbot what time school shootings tend to occur, and then, of course, went to carry out that act of mass violence. And so, these are the things that I think that regulation should really be targeting, not these kind of more fantastical ideas of super intelligence that, you know, I think are really the product of kind of delusional and deceptive tech people. [00:13:00] Speaker 2: So, Paris, would you say that there is a distinction between your critique, which is a political, economic critique, of the technology, in addition to, of course, these horrifying incidents that you've recounted of present-day harms, and the scientific and technological potential of this, say, generative artificial intelligence, or do you think even that potential that has been touted to us is misleading or an exaggeration? [00:13:38] Speaker 3: Yeah, in my view, I do think the potential is severely overstated, you know, I think that when we look, especially at how the tech industry tends to operate, it operates on exaggeration, right, in order to drive investment, in order to drive hype cycles, to get people interested in its products, to, you know, get a lot of media attention on what it's doing. Which is not to say that, you know, at the end of a shakeout, of course, there's a lot of talk of whether there is an AI bubble, and whether there's overinvestment into generative AI, and into data centers right now, and what might happen when there is a kind of market correction, you know, in that sector. And I think that what we'll find is that a lot of the places where the tech industry said that generative AI would be useful is actually not going to pan out in the long run. I think that there will certainly be some instances where, you know, certain sectors, certain companies find that it's useful to use generative AI products. And even then, I think their use of the products will look very different than what we have today. You know, there's obviously a big reliance and push to use centralized tools like ChatGPT or Claude from Anthropic. But actually, I think we'll see much more kind of use of the open weight models, the kind of smaller models that are targeted to certain use cases that are obviously also less resource demanding, which I think is how things are going to shake out. But even then, I would say that, you know, I think that the usefulness of generative AI as a technology has been overstated. I think that the pursuit of it comes out of certain kind of ideological ideas that the tech companies have, you know, they want to build this, you know, kind of human brain within the computer. And to do that, you need larger and larger models. But that's not necessarily the best way to actually develop AI for, you know, kind of benefiting the public, right? And one of the things that I took away from Karen Howe's work, who I know you've had on the show as well in her book, Empire of AI, is that there are a lot of other ways you can approach AI, right? But a lot of the money that goes into AI development right now is focused on generative AI rather than these, you know, many other techniques or approaches that could actually be much, much better for the public, you know, you know, be using far less resources. But the industry is not very interested in that right now. [00:15:51] Speaker 2: And finally, if you could talk about, because it seems to me that a number of people in the medical profession are very hopeful about the changes that generative AI will make possible in many, many different fields. And already, you know, the discovery of Google DeepMind's AlphaFold, which is considered one of the most significant scientific achievements in the history of artificial intelligence. We have 30 seconds. Your response. [00:16:19] Speaker 3: I'm quite skeptical of many of these claims. You know, I think that we have actually a history of thinking that new technologies are going to make big changes in medical advances and then they don't deliver. And I would just point to, you know, in Canada, the Auditor General in Ontario looked into these AI medical scribes, you know, the tools that transcribe conversations between patients and doctors, and found that in over half of cases, they were inserting false information, recording the wrong drugs. You know, and if this is kind of the basis of, you know, how a patient's medical history, that can really have some significant effects on, you know, how they are treated going forward if their medical file is filled with false information about their history. So, again, I'm a bit skeptical on that point as well.