Why Marsh May Let Markets Tough It Out

Carolyn Campbell: Welcome to Thoughts on the Market. I'm Carolyn Campbell, Morgan Stanley's Asset-Backed Securities Strategist.

Vishwas Patkar: And I'm Vishwas Patkar, Morgan Stanley's Head of U.S. Corporate Credit Strategy.

Carolyn Campbell: Today, how fixed income markets are helping fund the AI build-out.

It's Thursday, June 18th, at 10am in New York.

Let's get right into it, Vishwas. We've both come on this podcast before to talk about how credit markets are financing the AI build-out. And over the last ten months, I think it's fair to say that things are faster, broader, deeper than we perhaps expected initially.

This investment now spans investment-grade corporate bonds, high yield loans, and a range of securitized products. From your seat in corporate credit, why does AI infrastructure matter so much, to investors right now?

Vishwas Patkar: This is a big talking point in our client discussions. it's also telling that less than a year ago, we wrote about this topic for the first time, identifying a $1.5 trillion financing gap that credit markets could help bridge. At that time, data center debt was not something that investors were really focused on. Yet less than 12 months forward, this, I think, is the number one theme dominating both your and my market.

And why it's important, I would say, is across, three key vectors. First, just the scale. So, if you look at overall AI-related debt issuance so far this year, we're close to $250 billion. For the balance of the year, we expect that number to double, so about $500 billion of total AI debt financing for 2026.

Increasingly the second vector, I think, is around the complexity of deals. So initially, while AI financing was dominated by vanilla investment-grade corporate bond deals, we are now seeing that broaden out into project finance style deals in the high-yield market. We have seen an uptick in chip financing across the different credit silos.

And that's important for investors, as identifying value across these different options does require deep credit expertise. And third, as this investment cycle rolls along, it's also important to be cognizant of risks that are building. Not just from a very broad top-down sense around the demand for compute. But also, what are some of the nuances in these different structures – whether it is in data center construction or is in chip financing that investors will need to monitor.

So, it's across these three themes that we think data center debt financing is gaining importance.

Carolyn Campbell: Now, the underlying demand for AI infrastructure is very strong. That doesn't necessarily mean that every bond tied to this theme is automatically going to be attractive. And as you mentioned, [$]500 billion of supply for the year; a large amount of complexity between those structures.

How should credit investors think about the various risks within these different structures?

Vishwas Patkar: So, in investment grade, the story is a bit simpler. So, we have had unsecured hyperscaler bond issuance. We have had issuance from semiconductor names. And then we've had some, what we call, private style data center deals.

But the vast majority still comes from hyperscaler investment grade rated bonds. For this market, our focus is less on fundamentals because fundamentals are very strong. And then hyperscaler are some of the more most creditworthy companies that we've seen in the history of the market. Our emphasis more is on just the quantum of supply.

So, year to date, we have had north of [$]100 billion of hyperscaler debt in the dollar market. We've had north of [$]50 billion being issued in other currencies. If you look at the overall investment grade market, supply is up almost 25 percent versus last year. That's consistent with our call for a year of record issuance this year.

And increasingly, if you look forward and then map these issuance numbers to our CapEx estimates, where we could very much be on track for another record to be hit next year. So, the issue of the investment grade market is not around the fundamentals of the companies or these deals. It's more about the quantum of supply, which we think eventually will test the demand capacity of this market.

And our base case for the investment grade space is similar to 1997-1998, where credit was starting to finance the business cycle, spreads widened modestly, and IG could underperform other risk assets. But over a longer time horizon, spreads still look historically very low.

Carolyn Campbell: Now, what about further down the credit spectrum into the non-investment grade portion? What about that part of the issuance spectrum for AI?

Vishwas Patkar: Yeah. So, what we're seeing in the sub-investment grade space, especially in high yield, is very different. There, the growth in data center financing has happened around project finance deals for data center construction. In many cases, these have come from crypto miner companies that effectively provide what we call speed to power solutions.

We've also had some unsecured issuance from neo clouds, although that's relatively small. But this sector has expanded from effectively zero billion around the fall of last year to about [$]40 billion this year. We expect to see another [$]20 billion of issuance by the end of 2026.

And the way they fit into this whole ecosystem is – these project finance deals we think are interesting diversifiers for regular credit investors. They do come with construction risks, especially initially for the first two to three years till the data center is up and running.

But on the flip side, you do get a lot of structural enhancements and creditor protections, which is something you don't see in the vast majority of the high yield market. So, I think a key shift in the framework that investors have to do for these deals is focus on asset-level risk, which is again, I think a big divergence from how the vast majority of the credit market trades, which is largely unsecured corporate-level risk that investors have been used to.

Carolyn Campbell: All right. You just brought up construction risks. Do you think that's the biggest risk facing the high-yield investors today?

Vishwas Patkar: Yes. I think for the high-yield deals in particular, construction risk is the dominant vector that investors are focused on. Because it's important to remember a lot of the debt issuers are first-time borrowers. And they have a limited track record of construction in the past. So, you could see potential delays and things like cost overruns that can affect sentiment on the sector. Or at least on specific bond deals.

And this will be especially important to monitor going into the second half of the year, as we have some of the first delivery dates coming up for the deals in the sector that were announced last year. That being said, you know, even though some of the tenants have termination rights, if delays go beyond 180 days, our view is that given the structural power constraints, these termination rights are unlikely to be exercised.

So, while construction milestones can affect sentiment and short-term valuations, we would look at any blips as buying opportunities in the space.

Alright. So Carolyn, let me throw this back to you. So, construction risk clearly very important for the corporate credit market, especially for high yield investors. Is that something ABS investors or commercial mortgage-backed investors care about? And in what other ways are these asset classes different from corporate credit?

Carolyn Campbell: Okay. So first and foremost, the biggest difference is that in securitized products, the assets are stabilized, they're cash flowing, they're online. We don't have that first vector of construction risk in our space.

The second biggest difference is while in high yield and IG we've mostly seen – or we've entirely seen single campus, single tenant data centers; in securitization issuance, it's mostly multi-tenant, multi-asset, multi-regional, deals that have come to market.

And so, it's a very different risk profile. And as a consequence, investors are focused not just on who is behind this one single lease and what are the termination rates, but what does the landscape look like in general for compute? How does that affect vacancy and churn rates?

And then lastly, the issuers themselves are different. You talked about the crypto companies. You get a little bit more of the data center, data center construction. Whereas in securitized products, these are companies that have been around for 5, 10, 20 years. They're accustomed to managing a fleet of assets, dozens if not hundreds of tenants. They've got a little bit more of a track record for the most part, than the types of issuers we're seeing in the credit market.

Vishwas Patkar: Your market post-construction, more leverage to the thematic of demand for compute – and how the AI investment cycle is playing out. Versus the corporate credit market, which is largely exposed to construction risks as the data centers get built out. So that's a very important difference.

That being said, one theme that ties both our markets are just healthy fundamentals, but at the same time heavy supply. So, I talked about how we see that affecting our view on investment grade. How is that same tension showing up in securitized products?

Carolyn Campbell: So exactly as you said, the fundamental story is very strong. We don't see deterioration in performance of the assets either that has happened yet or that we expect to come in the near term. So, it really is a technically driven story. Supply in this space, we're forecasting at around [$]30 billion for year, so smaller in magnitude, but relatively large for the market. That has very elevated supply expectations, and so as a consequence, we've seen spreads back up across the space.

We do think that some of the cross-asset comparisons will help keep spreads contained from here. And so, we do see value in securitized credit across the stack for the rest of the year.

Vishwas Patkar: All right. So, you brought up the cross-asset comparison. And so, we've discussed the fundamental differences in our market, how much issuance we expect. But, you know, just to end on a commercial note – if we are advising investors on where is the best relative value and what's the framework for comparing opportunities, how do you think about that? Where do we see value across the ecosystem?

Carolyn Campbell: I mean, I think this is probably the biggest question that investors that are looking at this space are facing today. And there's... If we're thinking just about the data center backed assets, I think there are two main things.

One is the asset itself, where we're focused on things like the geography, the tenant, the interconnectivity, the flexibility of this asset for multiple uses. And then the second is on the structure of the deal itself. How much leverage is being raised against the asset? How cash flowing is it?

And then of course, the duration as well. But it's a great question. And because of the complexity of this space, it can be really hard to compare one to the other.

Vishwas Patkar: Yeah. And, at the risk of providing a non-answer, I very much think investors are in the process of coming up with a framework because these deals have come very quickly. This is a new sector for most credit investors to analyze. But I think what we can say with a high degree of certainty is this is blurring the lines between corporate credit and securitized credit.

So, you know, this opens up more avenues for us to collaborate on this topic going forward.

Carolyn Campbell: All right. That's a great place for us to leave it today with that nice cross-collaboration. Vishwas, thank you so much for taking the time to talk.

Vishwas Patkar: Great speaking with you, Caroline.

Carolyn Campbell: Thanks for listening. If you enjoy Thoughts on the Market, please leave us a review wherever you listen and share the podcast with a friend or colleague today.

Ariana Salvatore: Welcome to Thoughts on the Market. I'm Ariana Salvatore, Morgan Stanley's U.S. Public Policy Strategist. 

Today, I'll be talking about chipflation and what policy tools can or can't be used to address the memory bottleneck.

It's Wednesday, June 17th, at 10am in New York.

Last week, you heard my colleague Shawn Kim talk about chipflation and the surging cost of memory. Today, I'll get into what policymakers can and can't do about it.

As listeners will know, memory chips are becoming an increasingly strategic resource because AI infrastructure depends on them. And when a resource becomes strategic, governments tend to get involved. The challenge is that policy can help at the margin but probably can't solve the problem quickly.

There are three reasons for that. First, many U.S. policy tools all take time. Direct subsidies, tax credits, procurement guarantees, and faster permitting are all things that can support new fabrication plants, packaging facilities, and testing capacity. But memory supply is not going to appear overnight. This new capacity has to be built, equipped, qualified, and ramped – and that process can take years.

Second, China may be able to add some supply in conventional memory markets, but not enough to close the broader gap created by AI demand. That's especially true for high bandwidth memory, the more strategic type of memory for frontier AI systems. Supply there still remains highly concentrated, technically complex, and difficult to scale.

Third, our base case is that U.S. policy remains more restrictive, not less. We don't expect a broad loosening of export controls given the strategic imperative of this technology. Instead, we think policymakers are likely to continue to prioritize supply chain resilience, trusted capacity, and geopolitical de-risking over the near-term price relief.

Now, from a policy perspective, we think it's important to split memory into two categories. The first is AI strategic memory, high bandwidth and advanced DRAM. That's the memory that enables the most advanced AI systems. And for that reason, we think policy here is likely to focus on protecting strategic capability, limiting geopolitical vulnerability, and expanding trusted supply across the U.S. and its allied countries.

The second category is commodity or legacy memory. That's the memory that you can think of as being used in autos, industrial systems, consumer electronics, and other non-frontier applications. Now here, we think policymakers could consider more flexible options, like differentiated licensing or targeted support for critical sectors. But even then, the limits are practical: permitting, workforce, tools, qualification cycles, and production lead times.

China is the other major variable. Chinese producers are expanding in conventional DRAM and NAND. In some consumer-grade applications, that supply could act as a relief valve for buyers that have been crowded out by AI-related demand.

But still, there are limits. Chinese producers face yield and technology gaps, even if policy is supportive. And China alone will not solve the high-bandwidth memory bottleneck. The regulatory backdrop reinforces that point.

Some Chinese memory producers remain subject to U.S. restrictions or even heightened scrutiny. Access to the most advanced lithography tools also remains a hard ceiling. Without that access, scaling leading-edge memory becomes much more difficult.

So, the bottom line is this: policy can mitigate chipflation, but it's unlikely to end it in the near term. For AI strategic memory, policymakers are more likely to defend access, deepen allied coordination, and encourage trusted capacity than to loosen restrictions. For commodity memory, there may be room for some targeted flexibility.

But of course, geopolitics and timing still matter.

Thanks for listening. If you enjoy the show, please leave us a review wherever you listen and share Thoughts on the Market with a friend or colleague today.