Thoughts on the Market

Our U.S. Consumer Finance Analyst Jay Bacow and our Co-Head of Securitized Product Research Jay Bacow explain why AI can transform the way Americans shop for, manage and refinance their mortgages.
Read more insights from Morgan Stanley.

----- Transcript -----

Jeff Adelson: Welcome to Thoughts on the Market. I'm Jeff Adelson, Morgan Stanley's U.S. Consumer Finance Analyst.
Jay Bacow: And I'm Jay Bacow, Co-Head of Securitized Products Research, also working at Morgan Stanley.
Jeff Adelson: Today, how AI could change the way Americans shop for, manage, and refinance their mortgages.
It's Monday, August 10th at 10am in New York.
The U.S. mortgage market is worth more than $14 trillion, and its performance ultimately depends on the choices millions of homeowners make. Today, refinancing still means shopping around, comparing offers, and working through a lot of paperwork. AI could make that process much easier, especially when rates begin to fall.
Jay, you led this work on our AI mortgage blue paper. What's the main way AI could change the mortgage market, and why does the borrower matter so much?
Jay Bacow: So we think the biggest change would be borrower adoption of using AI agents to manage their personal finance. An agent on your phone could just monitor mortgage rates, compare lenders, reduce the paperwork, and make homeowners more likely to refinance when the economics work.
Let's think about what that could be. Historically, only about 30 percent of borrowers that had the ability to lower their mortgage rate by a 100 basis points did so in a given year. When a borrower went to get a mortgage quote, less than half of them asked more than one lender for a quote.
That agent could go reach out to 30 lenders, ask for a variety of different mortgages, could upload all the documents, could do this all effectively instantaneously, present the homeowner with the best option. Allow the homeowner to effectively click a button and refinance. I think this could be pretty transformative for the mortgage market.
Jeff Adelson: Now, as we think about this transformation, Jay, mortgage investors still rely heavily on past refinancing behavior trends. If AI makes borrowers more likely to refi[nance] when rates fall, how could that change the way these investors value mortgage-backed securities?
Jay Bacow: Well, we all know that past performance is not indicative of future performance, and those models are likely to understate future prepayments. If you get a faster response, it's going to make mortgages more negatively convex.
That's going to make the durations shorten. It's likely to widen mortgage spreads by about 10 basis points in our base case. And now, if that base case were to happen and we get, let's call it 100 basis point rally in the future, we think that that could cause something like a 40 percent pickup in refinance volumes versus our current expectations of what refinance volumes would look like in that 100 basis point rally.
Jeff, you cover a lot of the largest mortgage lenders. What does this mean for their business model?
Jeff Adelson: So, it's pretty straightforward. More borrowers refinancing means more loans for the industry to originate. Today, we're still sitting below what I would describe as normalized levels of originations.
We're sitting at about $2 trillion of mortgage originations per year. As we think about normalized, we think that's somewhere in the order [of] around $2.5 trillion. So just that $600 billion alone could get us straight there.
We tend to think about this more in our bull case, where we could see something in the order of $3 trillion of originations or more, still below what we saw during the peak COVID years of about $4 trillion or more. But still pretty meaningful and material for the industry.
Now, for the scaled lenders, that can create meaningful operating leverage. Mortgage companies have historically had to hire aggressively when volumes rise, and then they've had to reduce headcount when the cycle turns. AI could allow them to process more loans with the same employee base, making their cost structures more flexible and reducing the need to rebuild capacity during every single refi[nance] wave.
But the earnings benefit we don't think will necessarily match the dollar benefit from volumes. If AI makes it easier for borrowers to compare offers and allows every lender to process more loans, then competition could intensify and pressure gain on sale margins. So the opportunity is a larger market and better productivity.
The key question for individual lenders is: how much of that volume can they capture without giving too much back through pricing?
Now, as we think about automation, Jay, it could bring in more loans, but could also intensify competition and reduce the profit lenders can earn when they originate and sell a mortgage. So, how should investors in your space weigh those two effects?
Jay Bacow: So, the mortgage investors are short the option to the mortgage homeowner of when they can refinance.
And if the mortgage homeowner is going to be more efficient about refinancing, the mortgage investor is going to need to get paid more for that. They're going to demand wider spreads, and they're particularly going to demand wider spreads where that option that they're shorting is worth more. That's generally how it's going to play out, but there's also other aspects as well.
That duration shortening, because the borrower's more likely to refinance, means that the investors that own that duration will need to buy some more duration against that. You're also going to see more demand for duration as rates rally. So it's going to be a bid for the low strike receivers, as our options experts will pay close attention to.
And then if we get a further rally, you also get a more of an impact across the consumer writ large. You can imagine a world where mortgage rates are substantially lower than they are right now. An agent could sit there and say, "Why don't you consolidate your debt between your credit card, your auto loan payments, maybe your student loan payments and your mortgage?" Allowing consumers to save more and then maybe spend that in the economy.
Jeff Adelson: If we maybe take it a step beyond refinancing, how could AI affect home sales, homeownership, and access to home equity?
Jay Bacow: So let's just go back to thinking about this agent that's on your phone that's looking at all the opportunities.
Traditionally, right now, most people are only calling up one lender, they're getting one quote. If your agent is looking at lots of different lenders and lots of different options, you're probably going to get more ability to take out a mortgage. So you're going to get an expansion of the homeownership rate.
That's going to create more demand for housing. As rates rally, you're going to get home sale activity picks up more than it used to, and people are also going to be more able to take advantage of the equity they have in their house. So, you're going to get more usage of second liens and HELOCs and cash-out refinance activity.
Once again, we think this is mostly going to happen three to five years down the road, but we're not really sure exactly how this is going to play out.
So Jeff, what would be some of the signs that people could look at to see if it's playing out in the three to five-year timeline that we're expecting – or slower, maybe even faster?
Jeff Adelson: Sure. So yeah, I mean, I think it's going to be similar to what we've already observed as consumers ourselves and what we're seeing with all the LLMs and AI tools we're adopting today. You should see some rapid advances in the ease of use and the adoption of these technologies from a forward-facing, client-facing perspective. What we all see in the websites, what we all see in the apps.
It should become easier for us to engage with the mortgage process, compare rates to actually step into the process. Whereas today, you still need to maybe speak with a bank officer, a loan officer, or a mortgage broker to get deeper into the process and actually better understand what your rate means today.
So that would be the first step. The second step would be closing speeds. The average originator today still takes about 40 to 45 days to close a mortgage. The biggest and largest originators that have invested the most in technology and AI today are closing at about, call it, 12 to 20 days. So, half the industry level. So, that should come down over time and make it much easier to actually apply and finish a mortgage.
And then quite frankly, the most obvious answer would just be at the given level of rates that are outstanding today, we should see a step up in the level of refi[nance] volumes. That would be the most obvious one. But that'll be the outcome of everything else we've talked about rather than the actual cause.
Jay Bacow: That makes sense. So faster refinancing, it's likely to make the mortgage market more responsive when rates fall and effects that are going to reach well beyond the borrower.
Jeff Adelson: That could mean higher volumes for lenders, quicker prepayments for investors, and wider swings across housing and rates markets.
Jay Bacow: Jeff, thanks for taking the time to talk.
Jeff Adelson: Great speaking with you, Jay.
Jay Bacow: And thank you all 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.

Our Head of U.S. Public Policy Research Ariana Salvatore explains how U.S.-China tensions, export controls and domestic regulation are reshaping where AI is built, who controls it and what investors should watch.
Read more insights from Morgan Stanley.

----- Transcript -----

Ariana Salvatore: Welcome to Thoughts on the Market. I'm Ariana Salvatore, Head of U.S. Public Policy Research at Morgan Stanley.
Today, a look at how government is increasingly determining the future of AI in the U.S. – from where it's built to which technologies US companies and consumers can use.
It's Friday, August 7th at 10am in New York.
AI is rapidly reshaping the economy and society, so this is a pivotal moment for government to consider the rules governing that development. The first area to watch is technology restrictions, particularly in the context of U.S.-China competition.
Now, for much of the past decade, the government's approach has been to restrict a relatively narrow group of technologies with clear national security implications while maintaining broader commercial ties. But as export controls spread across more sectors of the economy and AI moves from software into physical infrastructure, the definition of what qualifies as national security has become broader.
The Department of Commerce could, for example, expand the entity list. That would require US cloud providers, software companies, and model marketplaces to remove or stop supporting models tied to designated Chinese developers.
Congress could then make those restrictions more durable through things like the annual defense bill or other policy vehicles. We're keeping an eye on several legislative proposals, like the AI Overwatch Act, which would tighten controls and give congressional oversight around exports of the most advanced AI chips; and the MATCH Act, which would extend restrictions further upstream to semiconductor manufacturing equipment and seek closer alignment with allied producers.
These measures wouldn't directly ban Americans from using a Chinese model, but they could constrain China's ability to train future frontier systems.
But it's not just the US that could impose a set of restrictions. China has a parallel set of tools focused more on integration and market access. Regulators could block four models or APIs. They could require locally controlled deployment. They could impose Chinese data and content standards or use cybersecurity and entity list authorities to promote domestic substitutes.
The likely result is an increasingly distinct pair of AI ecosystems. That's our two worlds thesis in practice. Over time, we think that means a bifurcated global AI market into separate technology ecosystems.
That looks like the U.S. relying on export controls, allied supply chains, and largely closed frontier model platforms, while China emphasizes domestic hardware, open-weight models, subsidized compute, and localization. Over time, that bifurcation could produce different chips, models, standards, data rules, and distribution channels, while third countries navigate between the competing stacks.
The second area to watch is domestic regulation. Today, the landscape is pretty fragmented. States are moving first on certain specific issues, including automated decision-making and child safety. Now, at the same time, Congress is confronting competing objectives from industry, consumer groups, and national security officials.
So far, we think the evidence suggests that the administration's preference is for a light-touch approach, a largely voluntary national framework rather than a broad new licensing regime. But it's also moving toward more direct oversight of the most advanced models. That includes the possibility to play a more active role prior to model release to ensure that certain protections like cybersecurity and intellectual property are met.
Publicly outlined priorities from industry seem to broadly overlap with that approach: a consistent federal framework, clearer liability standards, access to data, compute, and power, and copyright rules that don't materially limit model training.
But of course, the industry isn't monolithic. There are some important nuances between frontier developers and other players.
So, what does all this mean for investors? The government's reaction function will be critical to the way AI is developed and diffused throughout our society in two key ways.
First, we see regulation altering not only the pace, but also the geography of AI infrastructure.
At the same time, we think these constraints could strengthen the investment case for bottleneck solutions like on-site power generation, fuel cells, storage, and more.
Second, greater technology bifurcation supports investment in parallel supply chains.
The key takeaway here is that the government is no longer simply regulating the industry from the sidelines. It's helping to determine how fast AI develops through domestic rules, where it develops through infrastructure, permitting, and sovereign AI policy, and which technologies are accessible through export controls and market access restrictions.
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.

America’s biggest banks are opening more local branches. Our Head of U.S. Large Cap and Mid Cap Banks Research Manan Gosalia looks at the merits of physical locations.
Read more insights from Morgan Stanley.

----- Transcript -----

Manan Gosalia: Welcome to Thoughts on the Market. I'm Manan Gosalia, Morgan Stanley's Head of US Large Cap and Mid Cap Banks Research.
Today: why the bank branch you pass on your commute may matter more than you think.
It's Thursday, August 6th at 10am in New York.
When was the last time you went to your bank? You probably do most of your daily banking online and maybe go to the local branch for occasional transactions, like getting a certified check or talking to a financial advisor.
So, you might think that bank branches are fading into the background. But America's biggest banks are actually accelerating their investments in physical locations.
That shift could reshape the competition for your deposits. In our research, we looked at where 12 large U.S. banks are expanding their footprints, and we identified 57 target markets. 34 of those markets are being pursued by multiple banks, and nine of those markets are being pursued by five or more banks.
Since mid 2025, about 80 percent of these banks' new branches have opened in those markets. Most of the expansion is happening in the Southeast and Texas, with additional activity in the Midwest and several major metropolitan areas. 95 percent of the target markets have either above median projected population growth or they have ranked in the top 10 percent for deposit growth.
That helps explain why Nashville and Atlanta are each targeted by seven of the banks, while Miami, Dallas, and Denver are targeted by six. These are places where households and businesses are growing and where banks see an opportunity to build relationships that could last for decades.
The central question is whether physical branches still attract deposits. The evidence suggests that they do. From 2022 to 2025, 90 percent of the time when a large bank increased their branch share in the market, their deposit share also increased. But to become a real contender, a few scattered branches are not enough.
Banks generally need at least a mid-single-digit share of local branches to compete effectively. At 10 percent or more branch share, deposit share exceeds branch share by a median 3.5 percentage points. So, density, not just presence, is what matters.
Most large banks that we looked at have not reached that level. 60 percent of their positions in expansion markets remain below 5 percent market share. And so, this build-out looks like the beginning of a long competitive cycle.
Even then, the pressure is already visible in what banks are paying for deposits now. The highest offered retail certificate of deposit rates are higher in the South compared to the Northeast. Higher rates do make deposits more expensive for banks to fund.
In fact, evidence from the recent earnings reports suggest that this may already be happening. And we expect higher funding and branch costs to pressure bank margins and lift expenses into 2027. This means the cost of gathering core deposits could move structurally higher. And the lesson is surprisingly old school.
You can do almost everything on an app, but a branch on the corner still carries weight.
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.

From short-term interest rates to long-term bond yields, the Fed's credibility is being tested. Global Head of Fixed Income Research Andrew Sheets discussed inflation, Federal Reserve Chair Kevin Warsh's outlook, and the options ahead.
Read more insights from Morgan Stanley.

----- Transcript -----

Andrew Sheets: Welcome to Thoughts on the Market. I'm Andrew Sheets, Global Head of Fixed Income Research at Morgan Stanley.
Today: Can the Fed hold the line?
It's Wednesday, August 5th at 2pm in London.
The Federal Reserve has a difficult job.
The U.S. economy is a complex and varied ecosystem that covers everything from brain surgery to your burger order. The Fed is asked to keep prices stable and people employed using, for the most part, just one simple tool. A short-term interest rate, and without any control over what government policy or global events might bring.
Currently, the Fed probably feels pretty good about its success with one half of this – in the job market, given that the unemployment rate is near historical lows. But it probably feels less successful about price stability. Over the last five years, overall prices in the U.S. economy have risen over 20 percent based on the Fed's preferred inflation measure. That's roughly double the increase that a goal of 2 percent annual inflation would otherwise bring.
Into this complexity steps a new Fed chair, Kevin Warsh.
He has emphasized two changes for his tenure. First, that inflation is too high and needs to come down. And second, that the Fed has historically communicated too much with the market, which Chair Warshkeep thinks has helped contribute to investors potentially taking too much risk while also restricting the Fed's options to act.
What markets are now processing is a potential tension between these two goals.
After all, high inflation is an immediate issue. In a world where the Fed is hoping to keep price increases at about 2 percent per year, their preferred measure, PCE inflation, is rising more than 3 percent on an annualized basis over the last three, six, and 12 months. In the latest ISM Manufacturing Survey, [the] measure of price increases among manufacturers is well above normal.
In the face of that, one option for the Fed to combat this inflation would have been to raise interest rates. It didn't do that. Another would be to suggest that it was very close to taking action and likely to move soon. It didn't do that either.
Indeed, our economists think that the market took Chair Warsh's lack of guidance and action at the most recent Fed's meeting to suggest a pretty high bar for rate hikes; and even the potential to redefine the Fed's 2 percent inflation target in favor of something more general and unspecified.
The result was a market reaction that would suggest less focus on inflation. The prospects for rate hikes were reduced, the yield curve steepened, led by a sell-off of long-end yields, measures of expected inflation rose, and the U.S. dollar weakened.
In the days since, markets have settled a bit. But the result is going to be a market that is now going to be much more sensitive to incoming inflation data.
If that inflation data moderates in the second half of this year, as we at Morgan Stanley expect, then the Fed's approach could look justified – as the data suggests that neither action nor more communication about what they're going to do is necessary.
But if inflation doesn't cooperate, the challenge becomes immediate. Christopher Waller, another member of the Fed, recently said that "Sternly staring at inflation until it melts before our withering gaze is not an option."
The market will expect action and expect a framework explaining that action. Until that point, our rate strategists think that yield curves will continue to steepen.
Thank you, as always, for your time. If you find Thoughts on the Market useful, let us know by leaving a review wherever you listen, and also tell a friend or colleague about us today.

Head of US Public Policy Strategy Ariana Salvatore and US Thematic Strategist Michelle Weaver, alongside Senior Economist and Strategist in Morgan Stanley’s Private Wealth Management Sarah Wolfe, examine the economics of the AI datacentre boom, the pushback and the policy implications.
Read more insights from Morgan Stanley.

----- Transcript -----

Ariana Salvatore: Welcome to Thoughts on the Market. I'm Ariana Salvatore, Head of Public Policy Research at Morgan Stanley.
Michelle Weaver: I'm Michelle Weaver, U.S. Thematic and Equity Strategist.
Sarah Wolfe: And I'm Sarah Wolfe, Senior Economist and Strategist with Morgan Stanley Wealth Management.
Ariana Salvatore: Today: the politics, economics, and market implications of America's AI data center build-out.
It's Tuesday, August 4th at 10am in New York.
AI infrastructure spending is becoming a major force in the U.S. investment cycle. But as you've heard on this podcast in recent weeks, local resistance to data centers is growing, and projects worth hundreds of billions of dollars are being canceled or delayed.
More than 300 local moratoria have passed since 2023, and restrictions now touch 40 states. Now, most are temporary pauses, not outright bans, but the community opposition is tangible.
For investors, the key question is how these local pressures shape the broader build-out. So, I wanted to talk to you both because, Sarah, you've looked at this on the local level, and Michelle, you've been leading some of our thematic work on this topic.
So, Sarah, maybe we'll start with what happens when a data center comes to town. How does a large project ripple through a local economy, especially when so much of the expensive hardware is imported?
Sarah Wolfe: I think we need to look at the data center build-out from two lenses. First, at the national level, and then what's really happening at the local level, county by county.
So, at the national level, the headline investment can actually overstate the contribution to GDP because a lot of the components that go into data centers – think chips, servers, networking equipment – most of that is imported. So, it's actually an offset in the GDP accounting.
But when we analyze the AI build-out at a local level, we see that the town experiences the project very differently. A data center still needs a physical shell, concrete, steel, electricians, construction workers, and then the restaurants that feed the construction workers.
So, the local multiplier depends on how much of that spending around the data center stays nearby. Workers are going to get paid, local suppliers win contracts, and nearby businesses will see more demand. And then importantly, governments may collect more property and business tax revenue. When we look at county-level research on the AI data center build-out, we do see positive effects on employment, business formation, wages, income, and tax returns.
So, these data centers are significant. They do have significant multipliers. But we need to dig a little bit deeper and look at how it affects different counties.
Ariana Salvatore: So, it sounds like there are some local economic benefits. How durable do you think those are?
Sarah Wolfe: Some of the effects are durable and some aren't. The largest and most important effects come through employment in the near term. If we look at the construction phase of these projects, let's look at a data center that's 250,000 square foot, in Virginia. That supports more than 1,500 workers during construction.
But then, if we look at what happens after construction is done, there's only about 50 full-time workers once it's operating. And I will say I think that's a high-end estimate. If you look at how many workers these data centers employ state by state, some numbers are 10, some numbers are 20, and some are 30 employees. So, 50 is maybe on the higher end.
So, the bottom line is that the labor market multiplier actually fades after the facility comes online. What does persist, are the smaller share of data center processing jobs, ongoing supplier and service activity, and then importantly, of course, the property tax base.
But even that fiscal benefit depends on how the incentive package is designed. If a locality, for example, grants a very large, long-lived sales or property tax exemption, it may give away much of the revenue that made the project attractive in the first place.
So, the job story is real, but it's much more front-loaded. And then the tax revenue story is real too, but it really matters on how the locality negotiated the incentive package.
Ariana Salvatore: So, it sounds like there are some benefits and some potential drawbacks. How do you think communities should judge whether a trade-off like that is worth it?
Sarah Wolfe: I think communities should be asking this question of how much spending and tax revenue actually stays local after all the incentives? How many jobs remain after construction? Who pays for new generation transmission, water system, and roads? And who bears the spillovers through utility bills, housing costs, or land use?
The evidence does suggest that data center growth can lift incomes and expand the tax base. But it also raises home prices. And as we know, it raises electricity prices as well. A typical AI data center may use as much electricity as 100,000 homes, so cost allocation is critical. The strongest agreements make benefits durable and costs explicit through transparent reporting, sunset dates or claw backs on incentives, infrastructure cost-sharing, and protections that keep the household from subsidizing this build-out.
The test is really whether the community captures enough lasting value to justify the demands on land, power, water, housing, and public finances.
Ariana Salvatore: Michelle, I want to bring you in here. The local picture that Sarah describes helped explain why the politics can be so uneven.
How are moratoria and other local restrictions changing the pace and the location of the build-out, maybe on a national scale?
Michelle Weaver: I think you have to think about just the different type of moratoria themselves even. So, we're not seeing them uniform across different states in what's been proposed.
However, the majority of moratoria are a pause, not a[n] outright ban on construction. So, they might say, "Okay, we want one year," or "We want three years to do local impact studies and, and think about the way these data centers are going to impact communities."
So, the primary risk is really to the pace of the build-out, and as more and more of these moratoria pop up, you have to start to think about how that could shift the geography and the location of where these data centers will ultimately be built.
We are seeing a shift towards more data centers being placed in rural locations. This also has implications for the international data center build-out. You're seeing more and more of these data centers go up in Canada and in Australia to serve U.S. needs.
Ariana Salvatore: The polling data show us that voters are increasingly skeptical of AI. Specifically, they're worried about electricity prices and local costs. How should investors read that concern?
Michelle Weaver: Well, there's a couple things we have to unpack here. First is really around perception. So, in certain areas where you have both high data center activity as well as unregulated utility markets; yes, it's true, there is some of this raised cost ending up on consumer power bills from data center activity.
But in other areas with unregulated utility markets and lower data center activity, you don't see the same link between consumer power bills and what's going on with data center electricity consumption. But perception is what really drives politics and given that this perception is becoming spread across different states with both regulated and unregulated utility markets, politicians are reacting to it.
And the second thing this gets at is affordability. Consumers have been stressed by inflation for years now and elevated prices. And given that they think that data center costs are now winding up on their power bills, it's not surprising that you're seeing this big reaction, and that anything having to do with affordability has become a huge issue for voters.
Ariana Salvatore: Translating that into how we think things evolve from here, what industry and financing trends do you think matter most going forward?
Michelle Weaver: We recently identified the three main bottlenecks for the data center build-out as power, people, and politics. This whole episode has been about that third P, politics, but let's unpack power and people. On power, we still think there's a potential shortfall of around 38 gigawatts needed through 2028.
So, power is going to remain a huge bottleneck, and as the politics layer gets placed on top of the power layer, you're seeing more and more of an issue there. And so, what that really argues for is for data centers to be off grid. That way they can say, "Okay, there's no way we can potentially impact consumer power bills if we're not even connected to the grid."
The second P, people, is another big bottleneck, and we're seeing a very tough time for data centers to get skilled laborers. It's very hard to find electricians right now and other skilled laborers needed to set up these data centers.
Ariana, that brings us to the policy debate. Why is data center opposition moving from town halls into state houses and Congress? And what does this mean for a conditional build-out?
Ariana Salvatore: Yes, I think the points that you both touched on really explain why we're seeing this sort of pushback evolve, right?
Local communities are concerned about their electricity prices. Again, we see that on more a regional than a national basis. They're concerned about quality-of-life concerns. They're concerned about the environmental impacts. And so, all of that has caused these efforts to sort of cross state lines. We see it in both Democrat-held state legislatures as well as Republican-held.
So, it's definitely resonating with voters, and this is an issue that we think is going to be a key wedge issue into the midterm elections. It started to move into Congress rhetorically, but we still think something like a federal ban or a federal moratorium is very unlikely. And that's because we see a different incentive structure for lawmakers in Congress from the state and local level.
Principally, I'm talking about the U.S.-China relationship. So, when you look at the geopolitical backdrop to this debate, there are certain things that you can't ignore. And one of those things is that the U.S. and China are locked in this race for AI supremacy at the moment. And I think federal lawmakers have more of an incentive to respond to those policy demands and those policy needs, meaning they want to keep facilitating the build-out.
So that's why you're seeing the national level still relatively supportive of this build-out. We're seeing permitting reform. We're seeing Defense Production Act being leveraged by the president. We're seeing still an overall very favorable environment trying to unlock, sort of, that power bottleneck, for example.
So that's kind of what brings us to this conditional build-out.
Now, what does that mean? We think that the hyperscalers in these companies are going to have to offer some concession to local communities to facilitate the build-out. And that could be a number of things. I think it depends on the state's concern or the local community's concern principally, but we see a few different options.
One of those things is behind-the-meter power generation. So on-site power is one of the clear kind of offsets to this debate. Another thing would be improving utilization rates. So, our sustainability analysts found that the capacity utilization rates are actually quite low at some of these data centers in the range of 30 to 40 percent.
So maybe that can be increased. We've got some potential new regulations or transparency requirements around water usage. So, the short of it is, there's not going to be a one-size-fits-all solution here. But we think there's enough on the policy side that these companies can do or offer essentially to local communities. So that the entire build-out doesn't get delayed or doesn't get stopped.
And, and that's kind of why we still expect elevated AI CapEx, not just this year, but next year as well. We think that the risks are skewed to the upside for those numbers.
Sarah Wolfe: Ariana, I want to touch back to the comment you made on low odds of a nationwide ban on AI data center build-out – and tie it to this broader competition between the U.S. and China, with global supremacy in AI.
Can you talk a little bit more about how competition with China is going to prevent a nationwide ban and some of the national security concerns around that?
Ariana Salvatore: This ties into the theme of sovereign AI, which is something that we've been focused on recently, especially with all these discussions of more tech restrictions and controls between the U.S. and China.
And specifically, it's one of the reasons that we think the geographical build-out will be constrained to either just the U.S. domestically or countries that we are closely aligned with. And really the point I want to make here is that there's three geopolitical realities that are going to form, we think, the incentive structure for federal lawmakers and that are slightly different from the things that state and local policymakers tend to focus on.
The first is that we're seeing China leverage its supply chain position to pressure the physical inputs required for AI infrastructure, right? So, we're seeing that tit-for-tat escalation in the context of a broader strategic détente, but there's still a competitive aspect there.
The second is that we know China's accelerating its own physical AI build-out. Reporting indicates they're spending something like $300 billion over five years on its own domestic network, so very much full steam ahead in terms of its own domestic potential.
And the third is that research has identified China-linked influence operations that use an American frontier model to generate social media posts, comments, and political cartoons linking the data center construction to rising energy prices.
So, there's still a little bit of uncertainty as to whether or not those campaigns actually influence public opinion at scale. But in our view, it really underscores the linkage between national security and geopolitics and the AI data center build-out.
All these developments together we think underscore the physical component of the AI race and make something like a national data center ban or federal legislation toward those ends really difficult to reconcile with the growing bipartisan strategic imperative around AI, which is something that we think persists past the midterms as well.
But at the end of the day, the pace of the AI build-out will depend not just on demand, but on how well projects address the concerns of the communities hosting them. That's why we think this conditional build-out is probably the right base case for now.
Michelle and Sarah, thanks so much for taking the time to talk.
Michelle Weaver: Great speaking with you both.
Sarah Wolfe: Thank you, Ariana.
Ariana Salvatore: And 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.
*****
Sarah Wolfe is a member of Morgan Stanley's Wealth Management Division and is not a member of Morgan Stanley’s Research Department. Unless otherwise indicated, her views are her own and may differ from the views of the Morgan Stanley Research Department and from the views of others within Morgan Stanley.