Welcome to NoMore Risk Better, a Credit Sites Podcast. Across the global strategy team, we aim to make sense of the macro and the micro, highlighting opportunities and the risks facing the fixed income markets. As the macro makes headlines, we leverage our network of experts across fit solutions to better understand economic trends, rates, durations, geopolitical events, and how these factors impact corporates. At Credit Sites, we understand that credit investing comes down to picking winners to generate alpha and avoiding losers.
Our team over 100 analysts across the US, Europe, and Asia provide unmatched expertise and fundamental knowledge. In our weekly podcast, the strategy team offers a look at the conversations we have with our colleagues, including analysts, fellow strategists, economists, and leverage finance market experts. If you want to know more so that you can risk better, you'll want to give this podcast a listen. Hello, everyone.
Welcome back to the credit sites podcast. No more risk better. This is when he sees our global head of strategy at credit sites. And today I have assembled a true dream team to talk about all things AI, data centers, and maybe a little bit about the sauna, you know, live from Eric Vegasana.
We have a really excellent team assembled today to talk about technology and infrastructure and all of those headlines that you're seeing lately. First, we have Andy De Vries. He has our head of investment grade and head of utilities and really becoming quite the data center energy electricity experts. Jordan Shelfin, our head of technology, Andy Lee, a senior analyst on our tech team, and Eric Vega, senior analyst for telecom and infrastructure.
And we really are going to be making our way around the world of AI, data centers, hyperscalers, chips, you name it, we are going to discuss it. And I wanted to start with a speed round opener because we have some true experts on the topic of AI. And I think that one of the big things that comes up is how are people using AI? What's the tangible benefit so far?
So I wanted to go to the primary sources who are on top of the AI world and ask them, how are you using AI? What's your favorite use of AI tools so far? Andy De Vries, let's start with you. So we use the reasoning models in AI to track how much electricity AI is going to use in 2030 and 35.
So the reasoning models are not the quick chat back and forth. They scrape the web, they go by deep research, they go by everything. Grock is hands down the best. It pulls in more third party surveys than everyone else does.
Everyone else just pulls the same old McKinsey ones. Grock finds new ones from academia that we don't hear from. And we also tell the models to give us the answer in gigawatts or terawatt hours. It actually does that.
Some of the other ones don't. And that includes open AI's new 5.5. That's great. And also a very effective work and productivity use.
Jordan, how about you? What has been your favorite use of AI tools so far? Sure. So I'll give you two.
So personally, I really like uploading pictures into chat to BT and ask some questions. I find the answers to be shockingly very accurate and very helpful personally. And then professionally, I think improved search functionality has been a game changer for me. And that's both like Google AI overviews as well as called pilot within Edge browser.
The AI generated responses aren't frankly good enough to publish and they are prone to hallucination, but it has drastically expedited my research process. So I'm finding information and answers to questions much faster. Yeah, that's great. I find it definitely helps with kind of getting the brainstorming process going in a big way.
Eric, how about you? Yes, I'm really enjoying using the chatbots, you know, as a great supercharge search engine, not just the scan, you know, the web or info, but potentially helping find primary sources of information also works pretty well as a search agent inside the document, which is pretty helpful for pinpointing topics across different areas of interest and saving good deal of time. But as I've been ramping up my Latin American telecommunications coverage, it's also a great translator. And I'm fluent in Spanish, but I don't use it every day, so it can also help me save some time in translating some of those financial statements from Spanish back to English.
Yeah, how do you say you've been done in Spanish? You've been done. Perfect. All right, Andy Lee, what is your favorite use of AI so far?
Yeah, I really like it just for idea generation. I like to think of AI as sort of a consensus engine. So it's useful as a starting point when you're trying to come up with trade ideas that maybe are not consensus, right? I also think very good at steel manning most positions that you ask it for.
So it can be a good debate opponent to help you refine your thesis. Overall, I like to use it more as an iterative tool rather than something that's fired and for death. But that's great. I'm going to have to use that for when my husband and I are getting in deep philosophical debate.
He's going to love that. So I recently consulted chat GPT one dinner time when my five year old who's 26 today, happy birthday Joe, decided to squirt a bunch of ketchup in his eye and immediately went down because I guess salt and vinegar in your eye does not feel good. So I immediately turned bestie shot GPT to ask how does one remove ketchup from my five year old's eye and is this permanently going to leave him blind? And it was wildly successful in triaging the situation, walking me through steps for flushing his eye out, giving me tips from pediatric nurse on how to do it without having him fight back.
And it was fortunately a 15 minutes of chaos rather than something that could have lasted much longer. So if you need some quick medical advice, I'm not a doctor, but chat GPT apparently plays one in my phone, which is super useful. All right, with that behind us, I guess we have to get into the more work focused stuff and Jordan, I think that it probably makes the most sense to start with the big guys, the hyperscalers and they're hyper spending. You've written a lot of reports lately, especially post earnings season.
The AR arms race is in no danger of slowing down anytime soon. We continue to get growing cat back skydons. We've all seen headlines about these iPod popping signing bonuses and salaries being offered to top talents. Where do you think we are in the spending cycle?
Are there any signs of slowing? Yeah, sure. So on the capital spending front, there are no signs of slowing just to provide some some contacts behind us. The top four hyperscalers spend $150 billion in catbacks in 23 that increased to $250 billion in just 24.
And they're going to spend $380 to $390 billion in 2025. This doesn't even include Oracle, which is seeing the most significant growth, a little bit from a smaller base. And it doesn't include some of the smaller like neo cloud companies either. The hyperscalers continue to raise their catbacks guidance and the commentary for 2026 indicates that this growth will continue next year.
We're still supply constraint in AI infrastructure. So any new capacity that comes online is immediately absorbed by the market and translated into accelerated revenue growth for hyperscalers. If there is a sign that this could break, I would say it's the capital intensity or the catbacks as a percent of sales. This was nearly 60% of sales for Oracle in the most recent quarter and well above 30% from Med and Microsoft.
However, it's still very much an arms race for technological supremacy. These companies view AI as a once in a lifetime opportunity and they found ways to finance this, whether it's free cash flow generation from other businesses, data showings, leasing arrangements, joint ventures, or simply running capacity from other players. Med by the way is doing all of those things, but yet it's still very much up until the right. Yeah, up until the right for sure.
And clearly there's a lot of development and spend coming from big tech, but they're not the only players in town, right? Like who else is building out this data infrastructure data centers? And then how do we factor this into the kind of broading compute power equation? Yeah, sure.
Definitely new entrants here. And that's pretty common. Anytime there's a platform change in tech, you know, we've seen that historically in different eras, whether it's mainframes or PC's and servers or cloud or mobile, we're definitely seeing with AI, the major public cloud vendors will continue to be relevant. I would say Oracle is increasing its relevance.
They have that Stargate project, which alone will generate over 30 billion in revenue in fiscal 28. There's new entrants across the entire stack in tech, right? So that's data center operators, model builders, GPU as a service providers and software developers as well. There's a plethora of these like Neo cloud tech companies.
These are pure plays in the GPU as a service space. Core weaves is the largest and most well-known one player, but there's like dozens of smaller players as well. IBM in the video also provide GPU as a service. I think an interesting player is Crusoe.
They're in the data center space. They're building the AI data center campus in Abilene, Texas, which is being leased by Oracle and then used to power open AI. And Crusoe also has a GPU as a service business as well. I don't think there's this room for so many players.
There's economies of scale in this business. There's capital and technology requirements. I would expect there to be a shake up in the next few years. All right.
Well, I look forward to seeing that shake up because we all know that volatility creates some opportunity. And then I think we also all know by now that data centers require just a tremendous amount of electricity and the degrees that you have been at the front of AI coverage from this utility electric demand perspective. You have been very focused on really what is a massively wide range of estimates for demand and have also pointed out that estimates have stopped going up since deep-seak came on the scene back in January despite consistent spend. Can you give us an update on where you think data center electric demand is headed over the next five years and the capacity that utilities are actually planning to add?
Sure. So we're smart enough to know that we don't know the answer for 2030 or 35 because even the experts don't know how much electricity data centers consumed at 2024, which was almost nine months ago here. But to put it in perspective, you're looking at around 30 gigawatts of data center consumption last year. That's capacity.
A terawatt hour basis, 200 terawatt hour. So the consensus estimates are we're going to get 85 gigawatts for 2030. Over a wide range on that, summer down as low as 55, summer size 100. And then for 2035, there's only a few consultants out there that have estimates, but you're around 175 gigawatts.
To put this all in perspective, US to man for electricity on the hottest day in the summer across the whole country is around 760 gigawatts. So some of these estimates are you're going to be essentially using a third of the grid for just AI data centers. That seems like a stretch to me. But we're tracking those estimates.
And as you mentioned earlier, since deep-seak and efficiency games were announced in January, no one has raised the rest of it for 2030 higher, nobody. So despite the slow growth in AI, new models out there, no one's raised the rest of it. And the only estimates for 2030 have actually gone down since then. So it's pretty interesting with all these growth rates on the power side, you're just not seeing those estimates increase anymore.
So I think that's very important. Yeah, that does seem massively important. And I just cannot even fathom a third of our grid being used to fuel data centers. Do you think there are some disconnects that we should be keeping in mind?
What are the hidden risks there? We hear about the positives of investment in data, AI infrastructure, all of these things, but there must be some risk. So yes, we wrote a report highlighting data center disconnects. That was the title of the report a couple weeks ago.
So most of the report that week, and when we looked at these third-party estimates, so staying on 2030 and 2035 consensus call at 85 gigawatts and then 2035, 175 gigawatts, two thirds of the utilities that track data center spending, they already say 120 gigawatt backlog now that's contracted and planned and another 600 potential beyond that. The ability is mitt, there's some double and triple counting of data centers there. And that's 600. You know, even a third-party estimates double.
And again, they stop going up since easy. Even if they double, utilities are planning a building twice as many. They're tracking twice as many. So there's a real riskier.
You overbuild these data centers. And that creates risk in the utility and power sector where you've got 700 billion of investment from grade debt, another 100 billion or so of how you'll get that. Because these data centers get built, the utilities spend the money to put that respect to PPL, rural Pennsylvania, says $100 million connecting a new data center to the grid. So if you guys elect or twice that, you spend that money, the data centers have sounded three, four years, and all of a sudden you've got to recover that spending plus the utilities cap X on top of it, usually 10% RWE, then that cost is going to get shifted to residential ratepayers.
And that creates all sorts of political risk and then potentially credit risk. So we're tracking that very closely. Luckily, most of the utilities have put in place new rates to protect the residential customers. That's early exit fees, that's money upfront, that's minimum volume commitments, things like that.
But there are some states like Texas that are just behind the curve and getting these new rates in place. Yeah, I mean, that is absolutely fascinating. And also Andy, like, does that sound a little inflationary if we're getting new rates in place? Am I, is my utility bill going up?
Well, you live in the Carolinas and there's a lot of customer growth there. So your utility bill is going up a little bit, whereas where I live in Connecticut, and there's no customer growth, my bill is going up a lot, but that's a whole separate issue. Yeah. You know, my husband works for the utility as well.
So I figure that when I pay it, I'm just paying for his salary and pretend like that is okay. All right. So Andy, I'd like to bring you into the conversation and just trying to put some context around we have increased power demand and utility build and that from a revenue perspective, you know, that is how close are tech utilities data centers to capturing a solid return on investment? Are you worried about the over build angle from trying to really monetize some of AI?
Yeah, that's a great question. So first to maybe set some context in video is on track to do about 200 billion of revenue this year. And we estimate that about seven gigawatts worth of GBU shipments and that's globally. And if you add up all the consensus sales estimates through 2030, which are by the way, all up into the right, that's the only account for maybe a list called a 60 gigawatts total.
So yeah, perhaps some caution on the potential over build is warranted. And I think more generally it's clear that AI is transformative technology and often that requires a front-loaded capital investment cycle, right? And that's what we've seen for the last two years. There are currently signs that inference amends is starting to accelerate, which of course translates to revenue dollars for the hyperscalers, but in terms of the pace of monetization relative to the size of investment that's already taking place and continuing, I'd say we're still in very early stages of those two reconciling, right?
So ultimate ROI is still very much an open question. I do tend to worry about over build just because I'm a firm believer that technology is this secular disinflationary force. And we've seen cat-back movements and bus in the past, right? We've seen the fiber build out in the late 90s, the railroads in the latter half of the 19th century.
Those were also transformative with respect to communications, logistics, but those industries had to endure years and decades of underutilization after the initial boom phase. I think what's especially tricky about AI, which is really about information or knowledge generation, is that it's not as clearly tied to the physical world as maybe railroads or fiber. So it's just harder to model, right? It's harder to account for how algorithmic improvements might drastically reduce the physical inputs required compared to what we're seeing today.
And it's also harder to predict what sort of broad productivity uplift it could bring to the economy in the longer term. Yeah. I mean, you bring up so many interesting points that cycle of creative destruction, which is really why a lot of American capitalism has been built on is really powerful in a lot of ways, but it does lead to some setbacks and pain, especially when you see overbuild of capacity. So Eric, on the data center side, you really cover that data center infrastructure angle really closely and you've seen a lot of companies on your coverage, not surprising significantly increasing demand forecasts.
Can you give an overview of what you're seeing from demand growth perspectives? Do you think that the aggressive increase in demand expectations is realistic or are you too a little bit concerned about overbuild? Yeah. I think that's an important question.
And I think if you look back a few years ago to the onset of the pandemic, I think demand for data center capacity began to experience a pretty sizable boost, important drivers being the explosive growth in cloud-based applications, then magnified by the adoption of more remote working arrangements. And then the digitization trends continue internet traffic has grown at a K-GIRF over 30% from 2020 to 2024, and now surpassing six petavits per second. So all of these have been important drivers of capacity growth before we even really get into AI. And all of these trends are pushing companies to ramp up their infrastructure plans quite aggressively.
And now AI has added fuel to the fire. It's resulted in unprecedented search in the demand forecast across the industry, like we just mentioned. So to put it into perspective, how companies under my coverage have reacted, you have visual realty, which is one of the largest data center reads globally, essentially doubling the amount of megawatt capacity and reconstruction from the end of 2023 to 2024. And now under the leadership of a new CEO, one of Equinix's flagship strategic initiatives is called Build Bolder, which aims to increase its total IT low capacity from roughly two jigawatts to four and five years by 2029.
And so this is being done in order to capture future demand expectations around AI infrastructure, where Equinix forecasts, the total address will market size to grow by about two and a half times to nearly $100 billion by 2029, of which 70% of the expects will be weighted towards AI inference. Now, is this aggressive increase in demand expectations realistic? I think there's very real substance behind the forecasts, the competition requirements for AI are orders of magnitude greater than traditional workloads. But that said, the risk of overbuild is also real.
But the extent of it, it's very difficult to predict right now. I'll quote Microsoft's CEO earlier this year. He said that he expects her to be an overbuild. And they would expect to take advantage of that to at least cheaper capacity and 2027, 2028.
And so the amount of capital deployed right now, I think, is unprecedented. It's essentially aiming to build as much IT look capacity in five years as they built in previous 20 years. And so there's billions of dollars being committed to new facilities before customers have even signed contracts in some cases. So I do fear that build didn't enable that they will come approach, which has historically been risky across the communications landscape, thinking back to the overbuild of fiber networks at the turn of the century, like Andy mentioned, which led to a severe capacity coinciding with the dot com bubble.
But what's different this time is, I think, the scale of the power requirements for AI workloads. So securing power capacity has become the primary constraint for data center capacity growth rather than capital or customer demand, at least for now. So ultimately, I would personally bifurcate the overbuild risks between high concentration markets and newer emerging markets that lack a robust data center ecosystem. I think that the emerging markets located farther away from compute demand are at greater risk of overcapacity, but down the line.
And I think ultimately, some lessons could be learned from China's whole core press to rapidly grow their data center capacity. They had this state-led infrastructure development plan that they launched in 2022 called Eastern Data Western Computing. And it was predicated on building up data center capacity in the less populated regions in the West that have a better access to cheap energy. By this year, vacancies have remained high.
I think Reuters reported that utilization rates in many of these facilities stood at just 20 or 30%, which is because users still prefer to take up capacity closer to population centers where the compute demand remains concentrated to this day. Yeah, I mean, I feel like we've heard a lot about China infrastructure overbuild and ghost cities previously. And this is another kind of example, and hopefully a lesson to be learned from that. A lot of lessons also happening, I think, in the chips side of things lately.
So Andy Lee on chips and semis, we've seen so many headlines around. We have Tara Sun-Semi, the US government taking a stake in Intel. What are you watching on the chip side of things and what should we be really focused on? Yeah, sure.
So the chip sector has become very geobligally relevant in recent years, and a lot of that has to do with the transformative potential of AI. So I think the concern is mainly on two fronts. One is the US leads and AI chip design, thanks to NVIDIA mostly and to a lesser extent, the rest of the fabulous industry. But chip manufacturing has been largely outsourced to Taiwan over the last 10 to 15 years.
And what the current administration is trying to do through some pretty unconventional policies is two things. Number one is to restrict the outflow of AI enabling technologies to computing powers, right? So that's why we've seen tighter export controls overall. We've seen a 15% export tax flat on to NVIDIA and AMD's prior generation chips going to China.
And number two, what they're trying to do is to subsidize the build out of a more or less self-contained advanced manufacturing ecosystem in the US. So that's what the Intel Equity stake is about. It's also what the deal with TSMC for an expanded Arizona project. That's what that was about.
And ultimately, I'm kind of a realist about all of this. I think the increased government involvement in the space is going to continue. It's somewhat unavoidable. One can argue that pretty much all the Asian players, including Taiwan, have heavily subsidized their domestic chip industries.
And maybe at this point, it's just a competitive necessity. As far as what to look out for next in the space, there's two things that are on my mind. Number one is how heavy-handed is the US government going to be in encouraging fabulous customers, making video like Apple to shift away from volumes away from Taiwan and to US labs as they come online? And is there going to be an element of favoritism toward Intel, which, you know, until now has really struggled to get external customers relative to TSMC Arizona?
I think it would be less disruptive for most of the fabulous vendors to preferentially use TSMC Arizona for their US manufacturing. So we'll get to see if that's good enough in the eyes of the administration. And number two, we are looking for a bigger strategic shakeup at Intel sooner rather than later now. You know, one of the more under-discussed aspects of the new Intel deal is that it waves a bunch of Cloud-backed provisions from the original chip-backed subsidies that made it more difficult for Intel to engage in asset sales and kind of lock them into a fab-build-out schedule regardless of whether there was limited, regardless of whether there was customer interest or not.
So we actually think now because Intel receives this cash infusion upfront and there's less stipulations on how they can use the money, it gives them more room to explore things like maybe technology licensing with other Foundry partners, perhaps they sell a stake in Foundry, maybe that leads to a sendout eventually, maybe they can look at some AI related that previously was not on the table because of capital constraints. So definitely we're looking for the headlines from the Intel side to accelerate rather than taper off. Yeah, I mean, it's been quite a story with headlines so far and Andy, I know that you will continue to cover it with a lot of good speed. Eric, on the data center side, you mentioned Equinix earlier, that's one of the largest data center reads.
And this is a question actually that Davis Abert, one of the podcast favorites, TDAF. We have so much hype around AI and data centers. And in recent quarters, we've not seen super robust revenue growth from Equinix, about 4% in Tukyu. That doesn't seem like a lot when we look at other AI aligned company top line.
How should we think about this dynamic? Is Equinix just not benefiting from AI? Is it not showing up yet? What is happening there?
Yeah, it's a great observation. Equinix revenue is growing just not as much as it is another data center providers alone. And you know, Equinix has been a common target of speculation and public long or short bets, which have intensified interest in the name. But looking back, you had Jim Chainos and Hindenburg research, both of which publicized their short bets on this company in the past, where we historically tended to take the other side and be more constructive based on our analysis.
Most recently, Elliot has reportedly taken a big long position in his company. And so we think there's a potential activist angle that comes with that. So we're more cautious today with respect to Equinix following their expansive new CapEx program, some of the results that you mentioned in, you know, this CapEx build that it's actually a break from tradition for them, moving to $4 to $5 billion in annual CapEx to be financed with that in order to capture incremental demand. A lot of which is not expected to materialize until the later years.
And so that's going to push up net leverage by about a turn of view. So looking at the results today, I do agree they appear underwhelming compared to the explosive growth that we're seeing from other companies building the picks and shovels for this gold rush, like Nvidia. Ultimately, I think there's a timing disconnect between the winners and losers regarding the benefits from AI with Equinix. They operate primarily as a retail co-location provider focused on interconnection and so typical deployment sizes are smaller.
Think like single cabinets and cages rather than entire buildings and their strongest value prop has historically been their rich ecosystem of potential interconnection partners. So in this early ending of AI demand, I think a lot of focus has been more on the training of the large models, which require wholesale data center capacity, large halls filled with high density compute, and these deployments are often measured megawatts rather than cabinets or kilowatts. So companies like Digital Realty or Cyrus One have traditionally been better positioned for this type of large scale deployment. But over time, I expect Equinix will benefit from companies building out their AI infrastructure, even if the primary compute workload isn't hosted at Equinix.
Organizations need low latency connections between their AI systems and their different data sources or public cloud providers and ultimately the end user. So I think data center companies stand to benefit from incremental AI related demand by the varying points in time. You have other providers like Digital Realty, which have been seeing early stages of AI demand in its bookings and for wholesale capacity, which have been robust in recent quarters. Earlier this year, actually their CEO said that demand for large AI oriented capacity blocks remained pretty strong with about half of their bookings, they're weighted to AI.
But then you have Iron Mountain, which is seeing some near-term weakness as their facilities are better optimized to host more normal run of the middle cloud workloads and in their most recent earnings call management actually blamed cloud companies prioritizing the procurement of large capacity blocks to support AI training as the reason for consecutive downward revisions to their data center leasing forecasts last year and this year. There's a lot of great stuff in there, but I did hear you mention capital spending and some plans there. Is this all coming to the bond market? Should we be preparing for a deluge of new issue related to these aggressive capital spending plans for our companies pursuing creative financing?
Yeah, I think you name it. It's really a mix of that equity, free cash flow generated by other business lines, at least for companies under my coverage. So it really depends. That, however, remains an important source of funding for some more than for others.
For instance, Equinix plans to borrow $8 billion of incremental debt to fund their development plans for the next five years. They already have about $8 billion of bonds maturing over the same period. So we expect to stay combined, total of $16 billion of new issuance there for Equinix for the next five years compared to a total debt balance of $20 billion as of the second quarter. So that's pretty substantial.
Wow. But they also employ equity in their hyperscale developments to join ventures, where their equity stake amounts to just 20% to 25%. This is a much smaller business for them, expected to only amount to up to 6% of total consolidated revenues once it's fully built out. Digital Realty has been pretty good about using a mix of equity in debt.
Capital intensity there has been quite elevated, bringing near 50 to 60% of revenues since 2020. So the company has been aggressive in diversifying their sources of capital, using a mix of bond issuances, common equity, partner equity capital. And most recently, they set up a new US hyperscale fund that received $3 billion of LP commitments, which will enable them to fund some of their developments through that as well. And then other reads that we cover with data and exposure, like American Tower and Iron Mountain, has a repurpose a lot of the free cash flow generated by other highly cash generative business lengths to then invest in their data center capacity expansions.
All right. So a good mix, but definitely some bonds coming the way of our bond buyers in the corporate universe. Now, this would not be a podcast discussing credit if we didn't mention private credit. And that has definitely been a topic related to the AI data center electric generation value chain.
There have been articles lately highlighting recent meta bankers having private credit firms compete for a $30 billion financing opportunity for a data center in Louisiana. Are there risks or things to consider when we are thinking about private credit financing for data centers and AI and electric generation? Yes, absolutely. So we mentioned earlier these rates.
So you build a new data center, the utility spends hundreds of millions of dollars connecting it. And then you have volume ramp up, some minimum volume commitments, ex-fees, if you're not running in order to protect those residential customers. Well, if your counterparty is Oracle or Facebook, meta, Microsoft, Sterling, counter parties, but you saw in that recent one in Louisiana that Blue Owl came in for $3 billion of equity, which indicates it's a sale lease back. So Facebook wants it off its balance sheet.
So you cut a question. We don't know the answer because the talks weren't filed. But from everything we got from the Bloomberg story, it looks like it was a sale lease back. So if I'm entered Louisiana and that data center stops running or runs drastically less than expected, my claims no longer against meta.
My claims against some Blue Owl LLC, which needless to say is a significant lower credit quality than Facebook. So we're very, very tuned for keeping an eye off of that. There's another vantage data centers of a similar one. So I'm not sure it should be reflected in current bonds, but it's at least utility investors should be aware that your counterparty is not the Sterling big tech name.
Using AI to see who's building all these new data centers, it says that private equity is building over half of them. And on this podcast, the presenters and listeners listen to a lot of podcasts. Every single private credit firm out there is chasing the digital infrastructure wave. So you got to imagine they're all eager to lend and you think the issuers, the borrowers are getting some pretty spectacular terms there.
So utility investors should be aware of the counterparty risk. That's for sure. Yeah, definitely. Yeah.
What do you make of it all? Yeah, I'll try them in here too, just real quickly. For meta, it's a slight credit positive, right? I mean, they're offloading some of the risks.
They're moving some of that debt off the balance sheet and drastically reducing the issuance needs to the $29 billion cap of the raise. However, I would point out that, as Andy mentioned, it's a sell-least back. Specifically, it starts in four years and it goes for 20 years as per the news reports. And those lease liabilities will be on the balance sheet and we will treat that as debt.
And so it's a credit positive, but there's a pretty major offset there as well. Right. So to be continued and stay tuned to how this works. Hey, Jordan, I'm curious.
At this point, have you quantified how much debt is leveraged to the success of AI, maybe for hyperscalers, but also more broadly, have we tried to quantify it across sectors? Yeah, sure. And I know there's debt that touches everyone's sectors here, which specifically kind of within technology. I would say we're on our way to having hundreds of billions of dollars of debt issuance specifically for AI infrastructure.
And I'll give you a few examples here, right? So the meta deal we talked about extensively is $29 billion raise. $26 billion of that is debt provided by Pemco, which is expected to be syndicated out. So that's a meta of $26 billion plus they're probably going to come with a jumbo deal in the near future.
It could be around $10 billion, just regular course of issuance and will help satisfy AI infrastructure build out or coal. We recently published three or forecasts on the name. They recently came with a sizable deal, but we're forecasting over the next three years that their debt will increase by 25 billion. But it's a much more staggering staggering amount if you include lease liabilities.
We think that their debt plus leases could increase by 75 billion over the next three years. You have Corweave, which is a pure plan of space, right? So that's all AI infrastructure. They have about 15 billion of debt, including three or four billion of lease liabilities.
And in XAI, I think another good example, they raised $5 billion of debt in June. And the following month, they said that they're looking to raise another $12 billion of debt. And so you add all these up, you're kind of way past $100 billion and there's plenty of other debt to support AI infrastructure as well outside of these examples. And then like I mentioned, in the broader space, whether it's kind of utilities or chips or the Eric's coverage with the data center operators, the figure goes significantly higher from there.
Yeah, the limit does not exist apparently. Eric, on the data center side, when we are building data centers, I know that from a construction perspective, there's jobs there. But once a data center is up and running, is that a job creator? I think it's definitely a boost to the labor market.
The construction is very big forecast to data center developments that we have here at the US. While it's being built, you have all of these construction jobs that need to be taken, also putting pressure from other industries like in manufacturing. But then once that data center is built, you still need trade labor, HVAC, electricians, to help maintain those data centers. But then you also need engineers in highly skilled labor.
And so you have all of these data centers coming up in different regions of the US, these primary markets that are seeing the biggest portion or allocation of data center developments are likely going to see a significant boost to the demand for labor that can help maintain those facilities. Yeah, it'll be interesting to see how the labor market evolves with AI and data centers, because I know that we had a lot of companies actually reporting on earnings, some AI-related impacts on hiring, saying that we maybe don't need as many kind of entry-level jobs. I know that the software side of things, there's been a big focus on AI kind of unseeding entry-level software developers. I'd be curious to do any of you have kind of a high conviction view on and outlook for AI-related impacts on the labor market?
Are we there? Are we seeing the effect already? Or is this going to be a slowly melting ice cube that just kind of takes a long time to play through? I know the utilities are saying it's tough to find this sort of skilled labor, and it's just impossible to find the electrician to work on these data centers in the deep south.
Kind of reminds me of a couple years ago with these LNG terminals. You're hiring 10,000 employees at a time. There's just not enough people that have been to that trade school to do that stuff. So I think it's a big boost.
Apollo has the widely cited stat that data centers and data center construction had one percentage point to GDP growth in the first quarter. So certainly some good tailwinds in the economy in the near term. I would say when I say high conviction, but I'm a concern about the labor market. I think you have AI automating a lot of jobs.
We certainly see that, like you mentioned, with software developers, Coding Agents was one of the first primary use cases. There is, you mentioned earlier, like hundreds of millions of dollars of signing bonuses that met its paying right for top talent. I think that's still there, but that doesn't really provide a boost for the labor market and I'm concerned whether it's content creation, cost center agents, software developers. There's lots of these sort of white collar type jobs that are at risk of being automated.
And I think what ultimately happens is enterprises will spend more money and continue to spend more on global IT and they're going to cut back in other areas, specifically in labor, unfortunately. I'm glad that I didn't hear you say credit strategists on that list of white collar jobs that are going to be quickly be unseated because I do live in constant state of fear around that watch. Hopefully we can stay at the crowd. The head of global strategy will be safe for a very long time.
I am glad to hear that. And I guess given that I live in the south, I will be encouraging my children to pursue careers in the electrical trade to meet some of the data center. And AI field demand. My last kind of AI related question is what's the weak link?
If we're thinking about the AI value chain, it has truly touched so many parts of the US and global economy and there's so many stats around, you know, AI value chain is really the only thing that's growing right now. Everything else is a little bit sluggish to contractionary territory. So if we're kind of trying to pinpoint a weak link, are there places that you guys get nervous about? I'll sort of jump in with one.
I don't know if necessarily nervous, but I think the weakest link so far has been on enterprise software. The consumer adoption has really taken off. And we've seen sort of that like general purpose use cases for the enterprise, but you know, for enterprise software providers specifically, including the cloud type names like a Salesforce or even like an Adobe, we haven't really seen the revenue acceleration yet. We haven't seen meaningful AI revenue.
We've seen some of the strategy shifts right from co-pilots to agents. So we're at the really kind of early stages of agent AI. And I think that's going to be really needed to keep the cycle going, but promises to kind of bring more revenue in and kind of get that ROI equation going in the right direction to justify all the training costs. You know, I don't think we're going to kind of get there just on subscriptions and like API calls.
I think I think we really need the agents to take off and provide a value add for customers. Right. Taking the LOM, having that customer specific information and providing value add service or completing tasks autonomously. And so I think that's sort of been the weakest link so far.
It's also one of the biggest opportunities. So we're kind of, I think it'll be very telling over the next like six to 12 months, you know, how that revenue ramps and how successful that part of the businesses. Yeah. I think that's helpful.
Anyone else have a weak link you want to throw out before we wrap it up? Am I the weak link? I just wanted to highlight something that we've written about and we've talked about in the past. This is their new entrance because there's so much capital chasing this AI trend and some of my discussions at industry conferences, you know, one thing that people there have in the back of the minds is you have these new data center operators with very little industry experience going after speculative developments, maybe on aggressive terms, perhaps on leases.
So I think if they were to achieve a critical mass in terms of market share, that could probably weigh on leasing prices, perhaps vacancies. I think in some cases, they're waiting out on some mandates and already taking business for more established data center companies. And so more aggressive terms can have more wide reaching effects throughout the industry. So that's something to watch in my opinion.
Yeah, that's super helpful. Okay. Last question from me, another speed round. I know that we're all thinking big picture, you know, what's happening in 2030, 2035, but let's think about the next 12 to 18 months.
Where do you think AI will have the biggest economic impact over that kind of shorter term timeframe? Andy, I'll start with you. There's a lot of obvious answers for that sort of back office of Wall Street banks and college grads, whatever. I think the biggest impacts can be on identity theft because AI is just going to be able to rip off all five of our faces and all five of our voices and recreate us.
And I think there's going to be a whole booming industry for the life locks of the world. And I think that's the one that no one really talks about is identity theft and protecting against it. Wow, really starting it on a nefarious turn. I love that for us.
Jordan, how about you? I just want to point out that life lock is owned by Gen Digital, which is a WP high yield credit that we cover FYI. I actually want to execute on that on Andy's trade idea. I think that I think risk the labor market as well.
And I also think the use case that is ripe for disruption is probably the call center agents. The experience is so incredibly frustrating being on hold and trying to get real the problems and just imagine it how much better it will be once you have a powerful AI model that can understand and speak in natural language and also have all of your data available as well and just provide you with like a bespoke, contextually relevant response. And so I think that we've seen a lot of innovation in other areas, you know, like chat bots and coding agents and proof search, recommendation algorithms, content creation. I really think that call centers is we're going to see a lot of innovation there and a lot of much better customer experience, but also unfortunately less calls than our agent jobs needed as well.
Well, shout out to my mom, who's an avid listener to the podcast and hates the call center experience. You'll be really happy to hear that, Jordan. Eric, what do you think? Where are we going to see the most AI transformation in the near term?
Well, I think first I go with Jordan was saying the home security company, which is one of the credits that I cover, they now handle, I think over 90% of their customer service chats with AI agents and they're introducing AI voice agents to handle the remaining voice calls as well. So I think that's a very important thing to do with the customer experience. But in terms of where I think I will have the biggest economic impact over the 12th week next 12th, 18 months, maybe I'm just hyper focused in my sector, but I think in the infrastructure bill that itself we're already seeing unprecedented capital deployment for data centers, my conductors and networking equipment. I think this is creating immediate economic activity through construction jobs that we talked about, equipment manufacturing and just overall supply chain investments.
Great. And Andy, Lee, I'll wrap it up with you. Yeah, just echoing what everybody else has already said, I think even in the near term, we'll see some incremental down refresher on labor demand, especially for knowledge workers, close to entry level, a couple of years out of college, et cetera. I'm also sympathetic to the view that that'll probably be offset by some strengths in the trades and things like that until they roll out the robots.
But I think that's a, you know, some time away, hopefully. Robots are a little scary to me. So I'm hoping I'm hoping we have some time so that I can adjust that real quick. Robots is a logical extension of AI.
And for robots, you're going to need a whole new wave of batteries to power these things. And there's argument that Lithium Eyead won't get you there. So you might need to look at some of these new ones, like the Silicon batteries and there's some plays on that, but we'll save that for another podcast. I love that the preview for another podcast.
And I think that is a good place to wrap up. Thank you, Dream Team of Andy, Andy, Jordan and Eric for coming on talking about all things AI data centers, electricity generation, chips, you name it. I think we probably covered at least a little bit of it here. If any listeners have follow up questions for the team here, you can always find them on CreditSights.com using that Ask an Analyst feature and reaching out to your credit sites sales representative.
If you're watching on YouTube, like, share, subscribe. I think that's what the kids say, probably not, but I'm trying to promote us. All right. Thank you everyone.
And with that, we'll wrap it up.