Thoughts on the Market

Cheaper obesity medicines could unlock broader demand, while supply-chain bottlenecks and premium-drug innovation may also shape how the market evolves. Our analysts Terence Flynn and Thibault Boutherin break down the investor implications.
Read more insights from Morgan Stanley.

----- Transcript -----

Terence Flynn: Welcome to Thoughts on the Market. I'm Terence Flynn, Morgan Stanley's U.S. Pharma and Biotech Analyst.
Thibault Boutherin: And I'm Thibault Boutherin, Morgan Stanley's Europe Pharmaceuticals Analyst.
Terence Flynn: Today, how cheaper GLP-1 obesity medicines could reshape access, pricing, and supply chains; and what the first generic markets may signal for Europe and the U.S.
It's Monday, July 13th at 10am in New York.
Thibault Boutherin: And it's 3 pm in London.
Terence Flynn: Around one billion people live with obesity worldwide, including over a 100 million in the U.S. Right now, the introduction of the first lower cost generics of semaglutide, a GLP-1 medicine, in some international markets, could have consequences on affordability and demand.
Thibault, what are the first countries seeing the introduction of sema generics? What are the current dynamics, and why should global investors pay attention?
Thibault Boutherin: Sure. So, so far generics are being introduced this year in three countries: in India, Canada and Brazil. And if we look at India, this is the first market where the generics are being introduced. The patent for semaglutide expired in March 2026, and 13 companies have launched 26 generics across different formulations: autoinjectors, vials, and pills, which price is lower than the branded drug.
And because the India market was quite under-penetrated for GLP-1, we are seeing affordability driving volume expansion. In Canada, two generics have been launched so far. Four other generics are waiting for approval, and more are being filed. And finally, in Brazil, one generic was approved last month, and we are expecting these generics to be launched in Brazil in July. And 17 other generics are in different stage of regulatory review in Brazil, and we would expect more to enter the market by the end of this year.
And the reason why we focus on these markets is because we believe they could provide a blueprint for what could happen later in the U.S. and in Europe; in particular for Canada, which shares some characteristics with Europe and the U.S. And the patent for semaglutide will expire in Europe in 2031 and in the U.S. from 2032.
Terence Flynn: Great. Maybe on the India front, I know that's at the leading edge. What happened with patient demand when price came down?
Thibault Boutherin: Sure. So, what we saw in India is a surge in volume when generics were launched, and the volume in April 2026 were already six times higher than the volume in February. And that expansion has been driven mostly by these generics launch, which captured 80 percent of semaglutide volume in April. And our India team expect that the GLP-1 market in India will actually expand in value from $125 million in [20]25 to more than $1 billion by 2030, despite lower prices as we see better, you know, greater volume and greater adoption of GLP-1s in India.
Terence Flynn: The other thing, you know, you and I have discussed is the supply chain, and one of the questions is the ability of some of the generic manufacturers to scale semaglutide. So, maybe talk to us about the current capabilities. And could we see bottlenecks in the supply chain formation here?
Thibault Boutherin: Yeah, sure. So, there are three key elements to watch on the supply chain. The first is the active pharmaceutical ingredient or API, and that's the semaglutide molecule itself. The second element is the device and the device components, and the third element is the fill and finish, which is basically putting all of these things together.
On the API side, so semaglutide molecule, we believe there will be no bottleneck in supplying for generics as we see a handful of large Chinese companies, out of China, building multi-ton capacity for semaglutide. So, we believe there will be no shortage of API to supply the generic supply chain for injectables.
On the device, these are the same device companies that are supplying the branded version of semaglutide, and other GLP-1s for the device that are also supplying the generic makers. And we are seeing meaningful investments being made, so we don't believe there will be a bottleneck here.
Where we could see a bottleneck emerging is on the fill and finish side. Fill and finish requires highly controlled clean room space to minimize contamination. It requires regulatory approval, and it takes up to three years to build fill and finish capacity. And so, that's where if there is not more investment being made over the next few years, there could potentially [be] a bottleneck emerging for the generic companies.
Terence, while semaglutide generics will definitely represent a challenge for the existing branded version of this GLP-1, there are some insights in these emerging dynamics that suggest that tirzepatide, the other GLP-1, could be less at risk. Can you touch a bit on some of these dynamics?
Terence Flynn: Absolutely. So, just to remind listeners that semaglutide targets a pathway called GLP-1. Tirzepatide actually targets two pathways. The first is GLP-1, and the second is GIP. And there are some data comparing these molecules, both in Type 2 diabetes and obesity. And tirzepatide gives not only better efficacy but also improved tolerability.
And so, what you're seeing in some of the ex-U.S. markets is segmentation, where there are some consumers that are willing to pay a premium price for tirzepatide. Our team in Brazil has done a lot of work on this front looking at this dynamic and, you know, we expect that to play out in many geographies.
So, despite the entry of lower-cost generic versions, we think you will still see segmentation of the market between differentiated brand and the lower-cost generics. And that as a result, you will continue to see branded growth.
In the U.S. right now, market share is about 60 percent in favor of tirzepatide. And so again, you're seeing a differentiation between these two molecules.
Thibault Boutherin: And beyond the introduction of generics GLP-1s, there are other dynamics in the industry that are driving this market. And the introduction of oral drugs this year has been a big topic. Terence, what are your views on the role that orals could play on the market?
Terence Flynn: Yes, as a lot of people are probably aware, the many of the existing GLP-1 medicines are injectable. And so those are delivered once a week with a needle. But there are now additional oral options of these GLP-1 medicines. They started off first for Type 2 diabetes, but they have now broadened into obesity as well, following some recent FDA approvals.
And what we're seeing is that the introduction in the U.S. so far is expanding the market. So, the majority of people that are taking the oral versions of these medicines are new users to GLP-1s. So again, you're getting market expansion.
When you think about the orals as well, one of the other questions is capacity. I know, Thibault, you were talking about the supply chain. There are similar questions for these oral medicines because not all of the oral medicines are the same. Some are easier to manufacture than others, and as a result, that's another variable to consider.
So, some of these are what's called peptide-based orals, and some of these are non-peptide-based orals. And the non-peptide-based orals are much easier to scale, for a larger global market. And so that's definitely another variable that we're monitoring and that I think investors need to consider.
Thibault Boutherin: And beyond the pill versions of these GLP-1s, we are seeing more innovation in the drug pipeline of the industry, which could be a key driver of differentiation against the competition from the generics. So, what are we seeing emerging today from diabetes and obesity pipelines, which could be exciting for the future of the category?
Terence Flynn: So, as we see time and time again in pharmaceutical markets, the key players continue to innovate to try to improve profiles of the existing medications. So, there are, you know, kind of two areas. One would be efficacy; another would be safety tolerability.
And so, there are a number of players that are working first to develop longer acting medication. So, as I mentioned, the existing injectable drugs are dosed once weekly. But there are a number of companies that are working to develop potentially monthly or less frequent injections. So, that's one area that we're monitoring closely.
And then the second, and again, this plays into what I discussed on tirzepatide, is additional pathways that are involved here in diabetes and obesity, and a number of players are working to target additional pathways beyond GLP-1 and GIP. And so, some of the leading pathways that are being studied are something called amylin and glucagon, and there are a number of medications that are in the late-stage pipeline that are coming along, which have some pretty interesting data. And so that's another area that we're watching. And again, the goal there would be to either improve efficacy and/or improve tolerability versus the existing medications.
Thibault Boutherin: Great. And maybe we can also take this opportunity to talk about some of the short-term drivers in the market that are not facing generic today, like the U.S. So, what could be, you know, the key drivers for growth of GLP-1s and the overall obesity and diabetes category over the next five years?
Terence Flynn: Yeah, obviously the key one is seeing additional uptake of these medicines. I think right now we estimate, again, obesity in particular, there's about low double-digit percent uptake. And so obviously seeing increasing uptake of these medicines.
The orals, as I mentioned, are already driving market expansion.
And then the third is access. So obviously in any market, that's very important. In the U.S., I think about 50 percent of employers cover these medications right now. We expect that to increase in the years ahead as the data continues to build.
But then this year starting very shortly, the patients in the Medicare program in the U.S., so those people over the age of 65, will be able to access these medicines for $50 per month. And so, we think that is another driver of growth – is this will broaden access to about an additional 18 million people, starting this summer.
So, the next phase of the diabesity market comes down to execution, lower cost and scaled supply in the mass market, and innovation and differentiation to compete in the premium segment. Thibault, thanks so much for taking the time to talk.
Thibault Boutherin: Great speaking with you, Terence.
Terence Flynn: 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.

The USMCA review is underway, with implications beyond tariffs. Our Head of U.S. Public Policy Research Ariana Salvatore breaks down the key issues shaping the road ahead.
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 at Morgan Stanley Research.
Today, I'll be talking about the USMCA review – what happened on July 1st, what it means for North American trade, and how investors should be thinking about the road ahead.
It's Friday, July 10th at 10am in New York.
Last week, the six-year review deadline for the USMCA came and went. And as we'd anticipated, the U.S. declined to extend the agreement for another sixteen-year term. U.S. Trade Representative Greer stated that the U.S. did not agree to renew the USMCA in its current form, pointing to shortcomings and trade deficits with both Canada and Mexico, much of which echoed his testimony in front of Congress in December of last year.
So, what happens next?
This decision triggers an annual review process that could continue until the agreement's scheduled expiration in 2036. So, that means effectively the new deadline for negotiations is now July of 2027. And if we get to that point and see a similar outcome, this procedure repeats until the deal is terminated in 2036.
Now, importantly, the agreement itself remains fully in force during this period. The current tariff regime, rules of origin, investment protections, and dispute settlement mechanisms are all unaffected for now. That's actually in line with the expectation that we laid out earlier this year. In short, we anticipated an outcome in which negotiations stall and the deal moves to annual reviews. We thought that was becoming more likely than an ambitious expansion of the agreement in its current form.
That being said, there are some important implications of this outcome.
First, we think North American trade is being reshaped by a transition from a rules-based framework – where tariff schedules and preferential access anchored trade decisions – toward a more discretionary, sector-specific approach tied to industrial policy objectives. That, of course, increases uncertainty around exemptions, sector treatment, and consequently investment decisions for corporates.
Second, we think two bilateral deals may not be off the table. While it's still our base case that the trilateral framework remains intact, reporting seems to suggest that negotiations are progressing much more substantively with Mexico than with Canada. A third round of U.S.-Mexico negotiations is scheduled for the week of July 20th, while substantive text-based negotiations between Canada and the U.S. have not yet begun.
That asymmetry could mean that bilateral issues between the U.S. and Mexico are resolved more easily, while outstanding frictions like Canada's dairy market quota system could prove to be an overhang in those bilateral talks.
Third, the structural divergence between Mexico and Canada is accelerating, which is something my colleagues have highlighted in their recent work. If we think about Canada's manufacturing export base – autos, metals, machinery, energy, and transportation equipment – that actually overlaps with the areas that the U.S. government is increasingly defining as strategic. And therefore, necessitating more government involvement through, in things like Section 232 tariffs.
Canada accounts for only a negligible share of U.S. imports across computers, semiconductors, communications equipment, and advanced electronics. Those are actually the sectors where Mexico has become deeply integrated, particularly through assembly and re-export activity linked to AI servers, electronics, and industrial hardware.
Mexico now supplies roughly 35 percent of U.S. IT hardware imports and nearly 50 percent of U.S. server imports. And the North in particular has emerged as a vital interconnection hub between Latin America and the U.S. That's been driven by nearshoring trends, AI adoption, and multi-cloud strategies, as my colleagues Nik Lippmann and Fernando Sedano highlight. That means the scope and the objectives of the bilateral talks between the U.S. and Mexico and the U.S. and Canada may diverge even more from here.
So where does that leave us?
The USMCA is still intact, but the annual review process means North American trade policy is now a recurring negotiation, not yet a settled framework. And that will likely remain the case if policymakers agree next July to punt the issue yet another year.
The primary risk, in our view, stems less from the possibility of a full USMCA collapse and more from the prolonged uncertainty around implementation details, sector-specific trade measures, and Section 232 tariffs.
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.

Robert Feldman and Michael Gapen discuss how AI could reshape growth, labor markets and productivity in the U.S. and Japan.
Read more insights from Morgan Stanley.

----- Transcript -----

Robert Feldman: Welcome to Thoughts on the Market. I'm Robert Feldman, Senior Advisor at Morgan Stanley MUFG Securities in Tokyo.
Michael Gapen: And I'm Michael Gapen, Morgan Stanley's Chief U.S. Economist.
Robert Feldman: Today, we'll discuss why the U.S. and Japanese economies may react differently to the AI productivity test.
It's Thursday, July 9th at 8 pm in Tokyo.
Michael Gapen: And 9 am in New York.
Robert Feldman: AI is the biggest theme around the world right now, but AI will play out differently in different economies. Take the cases of the U.S. and Japan. In the U.S., it's already a catalyst in investment, imports, productivity, and the labor market outlook.
But here in Japan, it's seen as a savior for an economy with an intense labor shortage, low unemployment, and very little room to raise labor force participation.
Mike, in the U.S., AI's contribution to real GDP growth will rise from about 0.05 percentage points in 2024 to an estimated 0.43 percentage points in 2027.
What does that mean for markets?
Michael Gapen: Well, Robby, I think it, it means a number of things, but, you know, I'm an economist, so the answer is always, "It depends." I think the real crux of the issue over time in the U.S., and therefore what it means for financial markets, is ultimately whether AI is labor replacing – and pushes the unemployment rate higher. Or it acts like a more traditional general-purpose technology that's labor augmenting.
So, if, that's the case, meaning it looks similar to the internet and digital era, then it would mean faster output growth, stronger productivity growth, but still an economy that's running at or near full employment. That would be very beneficial in our estimation for risk assets, equity markets, credit markets, and it would probably mean that we stay in an interest rate environment that's certainly higher than it was during the post GFC period.
But if – AI is a very different technology than we've seen in the past, and it displaces labor, and we get increases in the unemployment rate as AI diffuses through the economy. Then it could be very different for markets. Maybe returns to capital and equity markets are supported, but that might be more narrowly for technology stocks and not broader, say, consumer discretionary stocks.
So, the answer, of course, is it depends. We don't know. And I think, ultimately, we come down on the side of thinking that AI will not create dystopian outcomes in the labor markets, that employment will hold up.
So, we have a fairly constructive view, perhaps an optimistic view. And we think, ultimately it'll benefit markets greatly, similar to what we saw from the mid-90s to the early 2000’s.
Robert Feldman: Well, in your model, you have a particular variable that captures the speed of diffusion. But your baseline has AI spreading twice as fast as the internet did. But without that rise of employment. Is that really manageable? And if it's not, what economic indicators would warn us, if we're crossing into the danger zone?
Michael Gapen: This is really the tricky part as, as you know. We have a new technology. We have to model how it diffuses through the economy. And I would say I think there's an argument here that penetration rates and usage rates are very different than what economists think about diffusion, which is how the production process is reshaped because of this new technology.
And so most economists look at the internet and digital era and think it took 20-25 years to fully diffuse. Mass penetration in maybe 10 years, but full diffusion in more like 20-25 years. And so, each innovation cycle tends to happen more rapidly.
So, I do think AI will spread more rapidly. And even by saying it spreads twice as fast as the internet did still means that it'll take roughly a decade, maybe 10-12 years for this to fully diffuse. So, our argument here would be that that is enough time for a flexible economy and a flexible labor market, like we have in the U.S., to rebalance labor.
But if we're wrong, then Robby, what I think you will see is that as AI rolls through, it diffuses faster. And what we would see then is increases in rates of job separation and layoffs that would overwhelm the labor market's ability to reallocate workers.
So, I think we would see two things – or three things: scale layoffs, a rise in the unemployment rate, and probably a significant amount of underemployment. Those who get rebalanced may be rebalanced into work that's not, say, consistent with the skill of that worker. So, I think we would see a very disrupted labor market in the process.
But if it takes a decade, maybe 10-12 years, we think ultimately the U.S. economy is flexible enough to rebalance labor without large scale layoffs.
Robert Feldman: Now, people are afraid of a lot of things, but one other thing is that AI might create new kinds of jobs, new kinds of tasks, have different impacts on people's wealth, and different responses from policymakers as well.
How do these knock-on effects change the AI labor story?
Michael Gapen: Yeah. That's right. I think you make a very good point there that I think it's easy to fall into what an economist would call a partial equilibrium trap. So, for example, we look at occupations exposed to AI task replacement, and we say, "Wow, if all these tasks are replaced, we might lose 10 million workers or 20 million workers."
But that's too simplistic, in our view. Because as you note, AI may destroy some tasks or replace some tasks, but it's also going to create new ones. So, it may eliminate some types of occupations but create others.
And in addition, if people are, say, laid off because of AI, you get a loss in labor market income for the economy. But AI will likely create returns to capital, say, stronger equity performance, and that's an indirect wealth effect.
So, our model kind of, looks at, say, three wedges or three horse races in the economy then. It's about the speed of diffusion of AI against the ability of the labor market to rebalance. It's task destruction or task replacement versus new task creation. And then third, it's we might have weakness in labor market income in the short run, but there are indirect wealth effects.
So, thinking about it this way in a richer general equilibrium context, these feedback effects matter a lot. So, the combination of if the labor market's disrupted, we get easing in monetary policy, maybe a fiscal response. There are new tasks, new jobs that are created for workers to rebalance to over time. And overall demand in the economy gets held up because wealth effects can offset some lost income.
All of that is extremely important in our view that ultimately the U.S. economy can rebalance and handle the AI diffusion in a manageable way.
We could be wrong, of course, but our main point here is you have to think about this in a richer context. You can't just simply, say, stack up workers and occupations and say, "Oh, we're going to lose a lot of employment." That's not the way innovation waves have worked in the past. We don't think they're going to work that way in the future.
Robert Feldman: Mm-hmm. That's fascinating because the situation in the United States is so different from that in Japan, largely because of the demographic situation.
Here in Japan, the key element is how much AI can ease the labor shortage. In fact, in some labor-intensive jobs now, we're seeing 6 percent wage increases, and that's great. As long as productivity rises fast enough that price hikes aren't necessary.
Michael Gapen: So Robby, in your scenarios for Japan, the same 10 percent productivity gain can lead to very different outcomes. Deflation and weaker employment in one case. More inflation, higher wages, and more employment in another.
What do you think drives the difference?
Robert Feldman: Mm-hmm. Well, the crucial element really is the flexibility of goods and labor markets. With high flexibility, you get higher GDP, higher employment, and moderate inflation. With low flexibility, you may get a bit higher GDP, but employment plunges, and there's deflation of both prices and wages – more in wages.
Now, in Japan, over the last two decades, we've seen monopoly power in key markets go down. For example, agriculture and energy. Labor markets are more flexible too, but lifetime employment system still applies to about two-thirds of the economy. And that deters people from trying to find better jobs and even from acquiring the skills needed for a new job.
Michael Gapen: What conditions are needed for AI to be additive to Japan's economy?
Robert Feldman: We need more reskilling. Japan is lucky because people are healthy, and they want to work into their 70s and beyond. But acquiring the skills to remain productive is a challenge, even though Japan's workforce is well-educated and still has a strong work ethic.
So, to sum up, in the U.S., the race is between diffusion and absorption. But in Japan it's between labor scarcity and productivity. Is that fair?
Michael Gapen: It is fair, and we come down on the side of optimism. We think diffusion will happen fast, but it'll happen at a pace that the U.S. economy can handle.
So, we come down having a positive view overall. We do not lean in the direction of dystopian labor market outcomes.
Robert Feldman: Mm-hmm. I agree with that as well for Japan. So, Mike, thanks for taking the time to talk.
Michael Gapen: Great speaking with you, Robby-san.
Robert Feldman: And thanks for listening, everyone. 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 Global Head of Fixed Income Research Andrew Sheets outlines what could potentially go wrong and disrupt markets’ optimism this summer.
Read more insights from Morgan Stanley.

----- Transcript -----

Welcome to Thoughts on the Market. I'm Andrew Sheets, Global Head of Fixed Income Research at Morgan Stanley.
Today, discussing three things that could disrupt a quiet summer.
It’s Wednesday, July 8th at noon in New York.
As markets turn the page toward the second half of the year, there are lots of reasons for optimism. Global growth remains solid. Earnings growth is strong, and broadening across more companies. Capital markets remain open and deal activity is robust. We continue to think that the best analogy for current conditions is something like 1997 through 1998 or 2005 through 2006 – periods where corporate aggression was increasing, and had further to go, leading to equities outperforming credit.
Even more immediately, July also happens to be one of the best months of the year for markets. And while one should never base their entire investment strategy on how far the earth has travelled around the sun, this month has been the best month for the U.S. High Yield returns, by far, over the last 15 years. The last time the S&P 500 fell in the month of July was 2014.
So given all that, what could go wrong? Well, here are three things that are on our mind.
First, a key part of our most optimistic view is that U.S. inflation will be lower than the Federal Reserve expects in the second half of this year, leading them to leave interest rates unchanged, rather than raise rates as the market expects.
The risk is that this assumption is just wrong, perhaps soon. There is certainly an argument that, if the Fed is worried about inflation, it shouldn’t wait to act, and the market is currently placing roughly 1-in-3 chance that the Fed hikes rates on July 29th. If that happens – and again, our base case is it does not – it could drive volatility.
Second is earnings season, which kicks off next week. While the general trend of earnings is important, the bigger focus is likely to be on the results of large U.S. tech companies, and in particular, how much they plan to spend building out AI infrastructure.
Over the last several quarters, almost like clockwork, these spending estimates have been revised higher and higher. And that has helped boost confidence in AI – as the spending is a sign that the technology holds promise – as well as boosting the broader earnings outlook; since all of this spending is becoming other company’s revenue.
Our base-case remains that this AI spending cycle has further to run, with capex from the major U.S. hyperscalers rising from over $800bn of spending this year to roughly $1.2 trillion of spending next year.
But the risk would be that second quarter earnings now show more hesitation to spend, maybe because the share prices of some of these big spenders have been recent underperformers. And given how much the current growth and earnings story is linked to AI, and how popular AI exposure is with investors, that would create a risk.
Finally, there’s Iran. Our base case assumes a gradual renormalization of flows through the Strait of Hormuz, and we forecast Brent oil at about $75/bbl in 12 months time, which is pretty similar to current levels. But as of this recording there were reports of renewed hostilities, and the ceasefire may be fragile.
The U.S. has already drawn down its Strategic Petroleum Reserve to its lowest-ever levels, potentially reducing some ability to absorb shocks if the conflict re-escalates.
Historically, July tends to be strong, and markets have a number of helpful tailwinds at their back. But an unexpected rate hike, an unexpected reduction in Hyperscaler Capex, and a resumption of the Iran conflict are three factors that are not in our base-case – and could disrupt that.
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. Also tell a friend or colleague about us today.

The biggest AI stocks have had a remarkable run – but questions still remain. Our Head of Americas Specialty Sales, Thomas Wigg, speaks with Global Head of Thematic and Sustainability Research Stephen Byrd and Global Head of Public Policy Research Ariana Salvatore about the competition and durability of the investment cycle.
Read more insights from Morgan Stanley.

----- Transcript -----

Thomas Wigg: Welcome to Thoughts on the Market. I'm Tom Wigg, Morgan Stanley's Head of Americas Specialty Sales.
Stephen Byrd: I'm Stephen Byrd, Morgan Stanley's Global Head of Thematic and Sustainability Research.
Ariana Salvatore: And I'm Ariana Salvatore, Morgan Stanley's Head of Public Policy Research.
Thomas Wigg: Today, the rally in AI CapEx beneficiaries has taken a breather in recent weeks on concerns of competition from open-source models, backlash to token-maxxing, and growing political opposition to data center builds.
It's Tuesday, July 7th at 10am in New York.
Let's start with you, Stephen. There's a lot of discussion recently around a backlash at token-maxxing. Essentially, enterprises trying to curtail their high spending on AI tokens from the frontier labs, and, in many cases, shifting to cheaper open-source China models.
Can you first offer some perspective here on the value of tokens for enterprises? I know you have a popular token factory model that walks through the economics of agents.
Stephen Byrd: Yeah, Tom, we do have this model that really walks through token economics, both from the adopter side as well as the hyperscaler side. So, let's do the adopter side.
So, there's a study out that shows a whole range of enterprise use cases of AI, and the average single use case that they identify would save a company about $55 or provide that much benefit. And while we don't know exactly how many tokens it will require, we can make some educated guesses as to a typical token usage to achieve that $55 outcome.
And we know that a typical American model, though this varies a lot, you can think of as the cost per million tokens being in the range of $5 per million. Some will be lower, some will be higher. So, for a few dollars of token cost, an enterprise can generate benefit of $55.
So that doesn't make me overly concerned about token spend and concerns about token-maxxing. I know we're going to get into that, but the foundation here is really good in the sense that enterprise use cases are very much in the money.
Thomas Wigg: How do you think market share ultimately shakes out on tokens? Do the cheaper models overtake the frontier AI labs? Do tokens bifurcate based on the complexity of workloads? How do you think this plays out?
Stephen Byrd: What we continue to see is this relentless pace of innovation and cost reduction. So, the frontier keeps going out – meaning model capabilities continue to increase, and, with that, we see enterprise adoption growing quite a bit.
Long way to say there is a role for both the frontier as well as these open-source models, and we'll continue to see both flourish. What I see is a lot of tokens will be spent on open-source models. A lot of the value will be in the higher end models because that's where enterprises are going to go. Let me give you an example.
I was speaking with one of our programmers about a recent project, and he used a very high-end coding tool, an American coding tool. And for him, that incremental cost of the tokens was very much worth it. And here's a very practical example as to why it makes sense for many enterprises to use the higher end models.
If a coding tool gets one of the thousands of lines of code wrong, the cost to remediate is very, very high. In other words, that incremental cost – in this example I'm thinking of, it's a few dollars incremental cost – is so worth it because if the quality is not there, the cost to any enterprise to go back and remediate is so high.
And that's true in a lot of enterprise use cases, but not in every use case. And what we are seeing is these open-source models that are cheaper will be very good for a variety of more mundane use cases that are still very valuable. That said, what we've seen in data from places like OpenRouter is dollar-weighted, meaning valued by enterprise spend, the vast majority is still the proprietary models.
But even within proprietary models, we could have more expensive and less expensive models. You do not need to go to the frontier. Where I come out on all this is that I'm very confident that the demand for compute is going to exceed the supply. What is difficult to exactly know is who are the winners, what is the exact mix. But the fundamentals of the demand for compute look extremely strong.
Thomas Wigg: So, I think you just gave me the answer, but I do want to bring this all back to AI CapEx. Now, last year, when the market sold off on Deep Seek concerns, the concept of Jevons paradox ultimately prevailed, where the cheaper pricing led to even greater demand and CapEx went higher.
Do you think the same plays out here?
Stephen Byrd: It does look that way very much. And the Jevons paradox dynamic is what we still see today in the sense that as the models get better, what we can do with the models increase, the cost of tokens will keep dropping, the cost of compute will keep dropping.
But let's talk about what might derail that, just to make sure we're thinking about all the risks. If somehow commoditized models could perform at the same level as proprietary models in all situations, then I would feel differently. But I don't see that. What I see is that these newer models really do have capabilities that are fairly breathtaking and that are worth that extra money.
But if somehow, we hit a wall where these models aren't getting better and therefore the sort of the open models are going to catch up, then I'd feel differently about that. This is where Ariana will, will come in in terms of policy and, you know, this comes up a lot when we think about U.S. versus China. How do we think about, you know, access to different models? How do we think about the cost of different models?
What about the risk of appropriation of capabilities by the Chinese firms, for example? That comes up a lot in policy circles. But the base case that I have is this just looks more like Jevons paradox, and there's going to be continued innovation, continued reduction in the cost of producing these services from these models. That looks like more of the same.
Thomas Wigg: Let's shift to Ariana to talk about the political angle here. The cover of Barron's over the weekend was a guy wearing a no data centers T-shirt. And this does seem to be one of the few bipartisan issues of agreement heading into the midterms.
The stat that the article gave was that 75 data center projects worth $130 billion were blocked or delayed in 1Q26, which is equal to the total number for 2025. This is according to Data Center Watch.
Now, most of this is in blue states like New York, Michigan, Illinois, Minnesota considering a statewide moratorium, but you're also seeing Pennsylvania, Arizona, Ohio, parts of Texas restricting tax incentives here.
So as this gets louder into the midterms, how do you think this plays out?
Ariana Salvatore: So, this is definitely one of the big wedge issues, not just for the midterm elections, but for 2028. And to your point, it's expanding into something that's got bipartisan momentum behind it.
Our view is that as long as the Trump administration is in power, something like a federal ban is unlikely to come to fruition. That's because we think the administration is still broadly supportive of the AI data center build-out. And I think even if you were to see a Democrat in office further down the road, that position is the same. And the reason is, it's just too difficult to imagine the U.S. giving up that strategic imperative relative to China.
So, while it is true that voters are against AI, while it is true that you are seeing these sorts of local efforts pick up steam, it's also the case that China is accelerating its own AI build-out – not just domestically, but around the rest of the world too. It's also the case that they are kind of tweaking some export restrictions on inputs for some of these data centers, and those geopolitical realities, I think, are hard to ignore.
So, at the end of the day, there is a broader strategic imperative here that both Democrats and Republicans kind of recognize and get behind. Now, what does that mean in the near term for the build-out? I think it's not that you're going to see a real pushback or moratorium so much as a conditional build-out.
That means you're going to see data centers have to incorporate things like grid modernization in their contracts, agree to longer term investments, for example. Do something that benefits the communities or give it back in some way. And I think that's kind of the policy trajectory in addition to the administration continuing to lean on tech companies to basically, you know, square the circle here and find some way to make this more affordable for, you know, local constituents.
Thomas Wigg: Stephen, let me get your take on this too, because I know you live in the D.C. area, and you have a lot of political conversations like you referenced earlier. How do you think this plays out? Is it a red state versus blue state dynamic?
And if what Ariana says comes to fruition, where it's a conditional build-out in terms of either giving back to the community or ensuring certain prices or certain technologies behind the meter, in front of the meter, does that have implications for certain areas of the market?
Stephen Byrd: Yeah. First, I think Ariana's points were all spot on. I just want to, kind of, build on that and, and dive into it a little more detail.
A few things. The politics are, from my perspective, not being the expert that Ariana is, I find them a little strange – in the sense that at the federal level, we have one dynamic, and at the state and local level, we have a bit of a different dynamic. And what I mean by that is, at the federal level, I think it's becoming increasingly clear just how geopolitically important AI supremacy is.
As these models get more capable, I think it's pretty clear that the Trump administration really sees just how potent these tools are from a geopolitical point of view. So that points in the direction of wanting to support AI and wanting to ensure that the United States has a leading and dominant position in terms of AI capabilities.
Pause there, and then go to your point about, sort of, the local and state level.
Building on what Ariana said, what I see are basically two approaches to data center development. In states where the utility is vertically integrated, meaning they control everything, like Louisiana, I do see a path where – in those kinds of states where the politics are a bit more favorable – you could develop a data center connected to the grid, where the data center developer is paying full freight and then some. Meaning that they are providing back to the community, they're providing sort of net benefits, and there should be plenty of capital to make that work and really support all constituents.
That can work – in a state where the politics work – because utilities are really weather vanes from a political point of view. So, if their state supports data center development, they will more likely support a data center development.
The other approach, though, in many states, whether it's deregulated or it's in a state where the politics are a little less favorable. Which, to your point on the cover of Barron’s, it's a lot of states, what I'm increasingly seeing is that the developers are going to go off grid. And they just don't want to show any impact to the community that could be considered negative.
So, no use of water, no use of power, and hopefully have a, you know, low or zero emissions profile to show no impact at all. Even then, you want to give back to the community. But the view there is, look, we want to sidestep all of these concerns that we might be causing impacts to the grid by just not being connected.
So, I think we're going to see a whole lot of off-grid data center projects. That's mostly natural gas turbines and fuel cells, that general approach. Energy storage will be required in a big way.
That's not easy to do. So, in the context of delays there, the Bitcoin players who do have grid access today are clearly seeing a lot of demand for their products.
So, I would say politics is now a huge issue that's showing up.
The other thing I'd flag is often local communities and states are rejecting projects and using permit requests as a way to do that. So, for example, if your data center needs an air permit because your turbines are going to emit some kind of an, you know, sulfur dioxide, et cetera, into the air, you can run into trouble there. If your data center requires water and you need a water permit, you can run into trouble.
So, that's causing these developers to try to find approaches that really minimize or eliminate the need for those kinds of permits.
Thomas Wigg: Stephen and Ariana, thank you for taking the time. And to our audience, thank you for listening. If you enjoy Thoughts on the Market, please leave us a review wherever you listen to the show and share the podcast with a friend or colleague today.
*****
Tom Wigg is a member of Morgan Stanley’s Institutional Equity Division and is not a member of Morgan Stanley’s Research Department. Unless otherwise indicated, his views are his own and may differ from the views of the Morgan Stanley Research Department and from the views of others within Morgan Stanley.