[00:00.4]
Steve Lee: Okay, well, good morning, everybody, and welcome to our webinar this morning. We’re glad you’ve taken time out of your day to join us. I’m Steve Lee, a partner emeritus at JFS, and I’ll be facilitating things this morning. Shortly, we’ll hear from two speakers: Stephanie Aliaga
[00:19.9]
of J.P. Morgan and Manish Upadhyay, our own JFS Chief Investment Officer. Stephanie is going to discuss all things artificial intelligence. A topic that has sucked most of the air out of the room around forecasts and headlines in recent months, and then Manish will step back with a slightly larger viewpoint of the economy and the markets.
[00:42.2]
Should be a great hour. First, a couple of brief housekeeping items. We do have some questions that have come in already, but as the webinar proceeds, please submit any questions you have in the box at the bottom of your screen. There’s a Q&A prompt.
[00:58.4]
And you can submit questions. Your audio is going to be muted throughout the webinar, but we will have time for the Q&A after each speaker, and we’ll look to answer the questions time permitting. If we don’t get to a question, we’ll follow up with an answer if you’ll let us know who you are when you submit the question.
[01:18.4]
Little bit of context as we enter the fourth quarter of 2026, literally today, October 1st. Financial markets are bifurcated year to date, with both the U.S. and international stocks sitting on solid gains, while bonds, facing stubborn inflationary pressures and climbing interest rates, are down somewhat.
[01:40.4]
The U.S. economy is growing. Yesterday, GDP data for the second quarter was revised upwards to an annual rate of 2.2 percent. And employment data and manufacturing data have generally surprised to the upside.
[01:58.1]
More recently, however, worries around unconstrained AI have grown, and calls for AI curtailment and halts to new data centers have increased in volume. The Iran war and the broader conflict in the Middle East have led to bond yields jumping, with the 10-year Treasury breaking through the psychologically important 5% level.
[02:21.1]
And that is raising volatility for stocks around the globe, and leading to the path ahead feeling a bit rockier and less certain. And as always, there are conflicting storylines for where we go from here. It’s my pleasure to first introduce Stephanie Aliaga, a global market strategist for
[02:40.2]
J.P. Morgan in New York. Stephanie’s research has focused on artificial intelligence and how it is reshaping the economy, productivity, and markets. She’s a key contributor to J.P. Morgan’s publication, The Guide to the Markets, as well as the On the Minds of Investors blog, and she serves on the committee that produces the Long-Term Capital Market Assumptions,
[03:03.2]
underpinning J.P. Morgan’s strategic asset allocation process. Stephanie’s a frequent guest on CNBC, Bloomberg, and other financial news outlets, and is often quoted in the financial press. She is a fellow of the Economic Club of New York and leads the AI Roundtable for the National Association of Business Economics.
[03:24.9]
She graduated from the Wharton School of Business at the University of Pennsylvania with a Bachelor of Science degree in economics. So with that, thank you very much, Stephanie, for joining us this morning, and I’m going to hand it to you.
[03:39.9]
Stephanie Aliaga: Wonderful, thank you so much, Steve, and hello, everyone.
[03:43.3]
Really excited to be here speaking with you all today. As mentioned, I am a global market strategist from J.P. Morgan Asset Management, and a lot of my focus, a lot of my research over the last three to four years at this point, has really centered around AI.
[03:58.6]
How AI is impacting the economy, productivity, where we are in the adoption cycle of actually using AI, and of course, as the title suggests, how do we invest in a market that is undergoing this rapid, tumultuous AI transformation.
[04:18.2]
Well, over the course of the next 20-25 minutes with you all today, hoping to get through three key parts to this conversation. First, where are we in the technological frontier around AI, the speed and the advancement of models, and what all of that has meant for tokens and the capex that we need to use this technology.
[04:43.4]
That gets me to two, how fast are we moving and actually using this? And then three, how the tides have continued to change around leadership in the AI space, and the continued broadening out of the AI value chain. So first, to cover where the models are today, the good and the bad.
[05:06.7]
First, it is clear that AI is getting exponentially better at completing real tasks that skilled professionals do. So this slide shows two different leading benchmarks around AI model progress.
[05:23.4]
And it’s like the good and the bad, right? So the good is the length of tasks that AI agents can autonomously complete has been experiencing seemingly exponential growth. So the latest model, or the current frontier model, is the Mythos model from Anthropic.
[05:42.3]
It was so powerful that they weren’t able to actually release it to the public. You may remember this discourse earlier this summer. They eventually did come out with a more guarded version of the model named Fable, and that model can achieve automation of tasks that take skilled humans over three hours to do.
[06:02.9]
Now, if you were to go backwards about a year and a half from then, well, the frontier models at the time could only do about 10 minutes worth of tasks that skilled professionals could do. So this huge leap in capability really owes to the increased degree of inference that these models are doing.
[06:23.0]
So the amount of thinking that they’re doing, that you no longer are really just interacting with a model in your browser in a chat interface, but you’re increasingly able to hand a goal to these models and let them execute that goal however they can find a way to do so.
[06:41.8]
So that can involve reading all sorts of research, spawning different AI models to tackle different pieces of that, iterating, calling different tools, and that requires a lot more tokens, which is the unit of data for these models.
[06:58.6]
But it also means that these models are increasingly capable, right? And so this has really sparked this year this increased focus and urgency around agentic AI. AI models that you set off to execute tasks, to execute goals.
[07:15.3]
Now, the challenge has been on the right-hand side. These models don’t always get things right, right? Even on the left-hand side, this is at an 80% success rate. The challenge is when they get things wrong, it can be very difficult to spot.
[07:31.3]
So, this shows the hallucination rate for these AI models, and what that means is the share of all the times that these models answer wrong, how often they just assert an answer confidently, right? They give you a wrong answer without admitting they don’t know or declining to answer.
[07:50.1]
And these hallucination rates can range at quite elevated levels for even very smart models. So if you look at Claude Opus 5 or Claude Fable 5, right, those models that I was saying are some of the most capable models, they are incredibly capable.
[08:06.3]
But because of that, they really don’t want to ever tell you they can’t do something, and that can be a real challenge for the user, for the enterprise actually using this model in its day-to-day operations: how do you mitigate and control these hallucinations?
[08:23.1]
They can be very difficult to spot. So, this has been the challenge, but clearly a huge opportunity in incorporating this highly intelligent technology. Now, a lot of focus has been around the token explosion around these models.
[08:42.3]
So, as I mentioned, all of that great intelligence comes at a price, right? And across Google’s platform, which is just one model lab or hyperscaler, right? They have said that across their platforms, they are now processing monthly over 3 quadrillion tokens.
[09:01.9]
So that’s like a new unit of measurement. We’re not really accustomed to speaking in numbers so grand, right? But that is how big things have gotten. Like, talk about big data, right? We remember big data from like the pandemic era. Well, things have just gotten a lot, lot bigger in the world of AI.
[09:20.8]
Now, this has also highlighted the challenge for investors in paying for AI, because if you’re using really premium models and you’re tasking AI agents to use them, well, you can find that an agent working overnight, running all sorts of math problems and pulling all sorts of context and checking all sorts of things, can run a very large token bill.
[09:45.6]
So, what we’ve seen just this year, which is what this chart shows, is this changing nature, this wave of price discovery in the world of AI usage. So when this line moves higher, that means that the average price of inference, the average price of using this model, is getting higher.
[10:05.0]
And when it moves lower, that means that average prices are moving down. So why do I show this? Well, there was this big trend earlier this year around token maxing. Let’s just use as much agentic AI as we can because that’s going to show how innovative we are,
[10:22.0]
how serious our engineers are in actually maximizing the potential of this technology. And companies were finding out that they were really running out of their AI budgets in like March and April of this year.
[10:37.3]
Now you’ve had Chinese open weight models also enter the fray, and they’ve shown that, hey, they can do what Claude Code is doing at a fraction of the price, and so that’s been a whole upset in the marketplace as well in terms of offering more cost-competitive solutions. And what we’ve seen most recently, really since this summer, as this line has moved pretty consistently lower, is that not only are Chinese models gaining popularity, but U.S. models are responding to this.
[11:07.6]
And if you pay attention to the latest model releases from companies like OpenAI and Anthropic, they’re actually not trying to push the raw intelligence frontier that we were just looking at. They’re trying to offer exceptional intelligence at very competitive prices.
[11:25.1]
They’re trying to compete on that price frontier so that they can make sure they keep that wallet share from enterprises and they keep that competitive standing for U.S. companies. We think this whole price discovery is what is necessary to get to a point of sustainable adoption so that we can continue
[11:46.8]
to bring these tools into our day-to-day lives and into our companies. Now, I also really like to keep a pulse on where the entrepreneurs are when it comes to AI, because the people in Silicon Valley, the smartest tech-savvy people out there, and even those that are maybe not as tech savvy but have suddenly found incredible ways of empowering their own toolkit with AI,
[12:15.6]
these are early signals of what is to come, and this slide shows two measures to that end. One is vibe coding, which is a term many of you have probably heard at some point this year. It is essentially just that you don’t actually need to know anything about coding languages to code these days.
[12:34.5]
You can code via vibes or natural language, and that is probably behind the surge in mobile apps that you’ve seen released on the App Store in the past year, up 160 percent. I myself am not a great coder. I avoided all the Python classes that they tried to throw at me.
[12:51.3]
I have found it so empowering and fun, honestly, to experiment with these tools, to figure out how to vibe code a slide like this. I’ve actually done a lot of my charts through code now without touching one Excel, without touching any PowerPoints.
[13:09.3]
Just plugging into the right data platforms and feeding it all of the data that it needs to know on how I like things to look, and that’s been really fun because now it takes out the grunt work of chart creation and iteration, and it helps me brainstorm to a much faster degree than I would have been able to otherwise.
[13:28.9]
I don’t have a research analyst working directly under me just yet, but I now have plenty of AI models working for me, and it’s been really exciting to see how it’s expanded just the surface area of research that I can cover. On the right-hand side, you also see this in the industry, and what gets me so excited as an economist is to see a greater degree of business creation, of entrepreneurship.
[13:57.4]
The number of new business applications filed at the IRS is at an all-time high, and I don’t think it’s a coincidence that this is happening in an era of AI. Many of my peers have gone on to start businesses in this era of AI, have gone on to work for startups, and I think this is like a completely new boom of business creation that we’re not really used to seeing.
[14:22.3]
And we’ve yet to fully see this in the hard data of the economy, but that is still to come. And it also tells us a number of things that we may expect when it comes to adoption. So let’s take things up a bit more and let’s think about the macro, economy-wide level of progress that we’ve seen around adopting AI.
[14:45.2]
Well, relative to any other technology that we’ve had before, AI is moving faster than any of them. If we look at this page, we show the years that it’s taken for all of these different technologies to reach mass adoption in our economy.
[15:02.9]
And I always get some laughs on this page, because who would have thought in the 1800s it took 75 years to reach mass adoption of the flush toilet. It was a really gross time back then for our predecessors.
[15:20.9]
Fortunately, every other technology has been faster and faster. Air travel, 55. Home air conditioning, 45. Smartphone, 10 years. Generative AI probably reached that point at like three years-ish, and that owes to a lot of things.
[15:39.2]
On the one hand, the actual hardware that you need to use AI is something that we as the digital ecosystem, a digital economy, already have. Beyond that, the barrier to entry is very low. You just need to know natural English, just plain language, not even perfect English,
[15:58.8]
to use these tools, so the barrier to entry has been really, really low. And I think, fast forward to today, where you have this new wave of personal AI assistants, a lot of people may not feel comfortable with giving these personal AI bots access to their personal accounts and financial accounts and card info.
[16:25.9]
But you might be surprised that people will get further along that comfort curve a lot quicker than they did for online shopping way back then. Like, it took so long for customers to get comfortable with buying something online. And how long is it going to take for us to get comfortable with sharing this data with our own personal AI assistant, when maybe that personal AI assistant could make our lives enormously easier?
[16:52.5]
Yes, there’s a risk, but I don’t know. How do we size that risk against the upside that we may feel every day, all day? I will say I’ve been using both Muse and Instinct in the last few weeks. So I just got a text from Instinct right now, and they are so handy in just the day-to-day, like errands.
[17:12.7]
Like, help me research this. Remind me to call this person. Can you go post this listing for me? Call Marriott and change my reservation for something. This technology is going to change so much of the way that we interact and the way that we consume things.
[17:30.9]
And we’re in the very early innings of that. But I say that just to point out that consumer adoption of this technology is unparalleled. The next question is, what is our willingness to pay for that technology? Most people here probably don’t pay much for their AI subscriptions.
[17:48.3]
Muse right now is free. Instinct is free, and we don’t know how long it’s going to last, but these services are so far offered for free. I will say I pay like $200 a month out of pocket across a range of different AI services, and that is beyond what my employer pays for my own internal AI software.
[18:07.9]
I’m trying to be a heavy user of this. It’s made me enormously more productive, but I didn’t intend to pay this much every month. I started in January thinking, I’m just gonna pay one month, test it out, and that way it can help me with my research and my conversations.
[18:24.1]
And now I can’t live without it. And I think many others will probably get to that point, that when you actually taste and appreciate the value that you’re getting from these technologies, you’ll find yourself very willing to pay that. How fast we move along that is a different story.
[18:41.8]
Now, for markets, a lot of focus not only on the consumer adoption story, but the business adoption story, because a lot of the ROI that these model labs, that hyperscalers are expecting to get from AI, is coming from our employers, is coming from Fortune 500 companies, and how much they’re going to be willing to spend on AI.
[19:04.1]
And the race is on. Every company is talking about AI. They’re figuring out what AI services they should buy. They’re giving their coders access to these AI agents, and they’re having these conversations. But the challenge is, while the technological frontier looks exponential, the actual progress that corporate America is making, it’s far more pedestrian.
[19:30.1]
We’re moving, we’re making progress, but if you just look at that blue line on the left-hand side, 22% of firms say they are formally using AI in a business function. That is not as high as some of you may expect it to be, just given how many companies are talking about AI.
[19:48.8]
And clearly, 56% of companies are paying for some form of AI subscription. So there’s a big gap there between experimentation and formal adoption. The other challenge is progress is very concentrated. This is some of the most striking data I’ve seen, on the bottom right-hand side.
[20:06.6]
The top 10% of adopters spend $650 per employee per month on AI, but the median firm just spends $12 a month. There’s a massive gap there. Now, part of that’s because you’ve got some companies there, like my own J.P. Morgan.
[20:24.3]
We’ve got a massive balance sheet and a massive tech budget every year, right? And there’s a lot of smaller companies that don’t. But even beyond that, you will have some small companies spending $200 a month per employee because they’re giving their employees Claude Pro Plus subscriptions, versus many other companies that probably aren’t doing that.
[20:45.0]
And there is a big opportunity here to be the company that not only just pays the most but extracts the most value out of premium AI, versus those that are maybe just waiting, maybe more cautious, maybe for reasons beyond their own ambitions. Maybe it’s because they’re regulated, maybe it’s because they just simply don’t know what to buy.
[21:07.1]
Everything is changing so quickly, and these are all of the challenges that the industry is really focused on fixing. But the reality is, the progress there, it’s not going to be defined by whatever the model labs have cooking. It’s going to be defined by human progress, by software innovation, by data readiness, by all of us getting on board and using these tools and bringing them into work every day.
[21:28.7]
And that’s a very different equation. So a more pedestrian pace, still a promising one. Long term, we see a lot of opportunity for productivity growth. It’s something we don’t think has been realized in the actual economic data just yet. With that said, the data doesn’t capture everything, and there is a likelihood that GDP is really not capturing this AI boom in a great sense, for many reasons.
[21:55.0]
Our government accounts were built for a manufacturing-based economy, and they fail to capture a lot of the nuances of a digital economy, let alone an AI economy. But beyond that, from a variety of different signals, we do seem to be in the midst of a productivity upturn.
[22:11.8]
But there is a lot of doubt around how fast it’s moving, how much more productive we are. And that’s a challenge for people like the Fed that are trying to figure out in real time how fast or how not fast the economy is running, right?
[22:29.2]
Now there’s going to be a lot of evolution in the ways that we work, and the transformation that we see in our day-to-day jobs and in the labor market overall. This looks at some estimates around what that potential exposure is across all the tasks, across all occupations to AI, versus what is currently being touched by AI.
[22:48.7]
That doesn’t mean that it’s currently fully adopted and scaled, but currently touched by LLM models. There is a big gap there, one that should narrow, but one that will likely never be fully realized either, for a variety of reasons. And I’m going to spend a couple minutes on this next page.
[23:04.8]
I think this is just so important. I get this question all the time. Is AI going to wipe out all the jobs in the economy and so forth? And look, I’ll leave the existential questions for another time. But for now, what we’re seeing in the real data of how these tools are being used is far different from the kind of jobpocalypse fears that were so popularized.
[23:28.4]
So, what does this chart show? It is really great research from Harvard and the NBER that follows what happens to software developers, and they’re like the poster child right now for AI automation because we fully automated coding. It’s like we fully reached automation with coding, right?
[23:44.8]
Well, turns out not so much. If you think about the job of a software engineer, and I think this is exactly like most knowledge workers, it is a decide, execute, deliver sandwich. Decide: what are you going to build? What is the product change, the product release that your customers really want right now?
[24:04.4]
What is the market telling you? What is your own business prioritizing? That decision process, it’s still very human. The models don’t actually have all of that industry-wide, business-wide context right now. Nor do we think they will for a little bit of time. Beyond that, execute, right?
[24:22.4]
Executing code, that is where AI is enormously helpful. And you can see it. The lines of code changed by these models versus developers that weren’t using AI, it’s up almost a thousand percent. So now we don’t have to spend time writing code, which is great.
[24:38.6]
However, what is the ultimate productivity gain when it comes to the release that you’re shipping out to users? That is only up 20 percent. And they’ve actually updated this study for more recent coding tools, and they found that even with the latest fully agentic models, you’ll see lines of code up 2,000 to 3,000 percent, and yet releases are still only up 20 percent.
[25:05.5]
How is that the case? Well, who is delivering this code? That human has ownership. They have accountability over what they’re releasing, what they’re putting forward, and that is a huge bottleneck. But also, it’s where a lot of the hard work gets done in figuring out what is actually valuable, making that judgment, making that decision, and ultimately delivering.
[25:29.3]
And that is still very human. So I will say, I have friends that are software engineers. I have a lot of peers that work in Silicon Valley. No one’s like just twiddling their thumbs and taking summer Fridays. People are working around the clock even more as they bring AI into their workplaces, increasingly because of all of this new work when it comes to orchestration.
[25:53.2]
So I think at one point a lot of our jobs are going to be orchestrating AI output, and that does point to a shifting nature of the tasks and the skills that are going to be demanded in the new economy. Now, shifting gears a little bit, let’s talk about how we even get to all of that adoption.
[26:13.5]
And that is where the infrastructure wave has just been in really great focus here. The hyperscalers have really accelerated the amount of capex they’re putting towards AI. This year, they’re expected to spend $800 billion, and now we’re looking at
[26:30.9]
the next few years of over a trillion, maybe even getting to one and a half trillion dollars in annual AI capex. We are already experiencing the biggest industrial revolution that this economy has had, even in inflation-adjusted dollars.
[26:48.7]
What is so remarkable about this boom so far is a few things. One, how it’s been financed. These companies are some of the most cash-generative companies that have ever existed, and while they are now raising money in debt and equity markets, they have so far funded most of this with their own cash flows.
[27:08.1]
And going forward, it will still be predominantly cash, but they are leaning more on external financing too. And the other thing that is remarkable is, we are not building a pipe dream. We are not building years before the demand is actually there, and that is where we got into a big old mess
[27:27.1]
in the internet era, because we built way ahead of actual demand. Today, these companies are highly, highly capacity constrained. Demand is well above what they can reasonably service today, let alone what they think they can service in the next few years.
[27:43.9]
And while they’re spending $800 billion this year, the combined backlog of customer commitments, customer-signed commitments for future revenue in the coming years, is now at $1.7 trillion across these companies.
[28:00.2]
So demand is there. It’s real. It is massive, and it is why they feel so confident in continuing to up the ammo here on capex. That doesn’t mean that investors are singing their praises as much as they used to, and that’s been another really important theme as we close here on these last few slides on the AI investment landscape.
[28:22.6]
The first few years of this AI wave, we were talking about the Mag 7, the hyperscalers and Nvidia, and what has changed is a lot of things. One, the hyperscalers have continued to invest and increasingly invest.
[28:39.1]
So that’s funneled more capital towards these picks and shovels on the bottom here: chips and data centers and power. The kinds of chips we need have broadened. We don’t only need Nvidia GPUs, we also need CPUs, which are more traditional chips, because that’s what’s going to help power agentic AI.
[28:56.2]
You need memory, because what’s going to make a personal AI assistant so useful? One that’s actually personal to you, that remembers everything about you. We need a lot of memory for that, and we need all of the picks and shovels that come together to produce this highly sophisticated technological hardware that AI relies on.
[29:14.6]
In data centers, we not only need to construct more of them, we need them to be more efficient. We need more efficient data center cooling. We need better networking so that we can connect data centers to each other, because we’re going to try to be building these bigger and bigger clusters of data centers. And then we need power. We need power wherever we can get it.
[29:31.4]
It’s all hands on deck right now. And that is creating a lot of opportunities not only in public utilities but also private infrastructure, some of these longer-term financing deals. And then finally, AI tools, AI applications. This has been one area that
[29:47.9]
could become the most lucrative area, but is also one that is fast evolving. The model labs are competing with one another. Take this example of Meta’s Muse, for instance. Meta has not been very good at creating models. Their model suite, Llama, has been like forgotten, right?
[30:04.8]
Like, not many people use it. But they realized, okay, we might not build frontier intelligence, but we’re really good at building something that consumers actually want, something that’s actually consumer friendly. So why don’t we repackage what Claude has cooked up? But instead of forcing people to build their own Mac minis over the weekends and try to figure out how to get things to run in the cloud, we’re going to take care of all of that, and we’re going to give them some really cute emoticon that they can name.
[30:31.9]
Mine is named Remy, that they can use in their daytime, and it’s going to be really easy and consumer friendly. And so that’s been the latest surprise in this race, right? Just a testament to like, we really have not seen how these new AI services are going to mature and who’s going to be the winning horse in all of this, which is why for clients and investors, we emphasize the need for a few things.
[31:00.8]
And I’m going to end on this slide. We emphasize the need for diversification. You do not want all of your money on any one horse. But also the need for an active approach to technology investing, because things are changing quickly.
[31:18.5]
You don’t want to be leaning into memory chips in June of this year when Micron was surging every single day, but our portfolio managers saw that things were getting a little over their skis in terms of momentum, right? What are some of those underappreciated opportunities that are in this ecosystem that may have a lower hurdle to reach
[31:39.1]
and also be an attractive winner over the medium to long term? Those are some of the big questions. You also want to make sure your portfolio is not all AI, right? Right now, it’s a really good time to think about insurance. You don’t want to be buying insurance when your house is on fire, right? We’ve been talking a lot about fixed income lately, given that you can actually get paid for waiting in insurance, given current yields, and that is something that should help you if things suddenly go wrong in this whole AI space, or even if something macro happens that pulls the rug out from underneath.
[32:12.0]
But I think this AI theme is still the most attractive to us over the medium to long term. It keeps us very constructive on the outlook for U.S. equities and the outlook for earnings. This market is like a market of nine lives, and it has weathered so many different headwinds: geopolitical shocks, wars,
[32:30.6]
rising inflation, a Fed hiking cycle now, the 10-year up over 100 basis points in just the last year. All of these different headwinds, and yet markets seem to be doing just fine. And how is that the case? It is remarkable earnings growth. It is record profit margins, healthy corporate balance sheets, and the promise of AI productivity, not only benefiting the Mag 7, but a much broader swath of companies across this page, across asset classes, and so forth.
[32:59.8]
So, I’ll end on that positive note. Would love to open things up for any questions in these last few minutes. We didn’t get through all of the slides here today, but all of my research is on our AI hub, which is linked here. And you can also follow me on LinkedIn if you’re active on the platform.
[33:17.8]
Always sharing my nuggets on AI on there as well. But with that, Steve, I’ll turn things to you.
[33:24.7]
Steve Lee: Okay, Stephanie, that was great. Thank you for the insights and the comments. We do have a couple of questions that we’ll pose to you.
[33:36.9]
One is, and you mentioned this, how to recognize hallucinations for those of us that aren’t necessarily versed in AI, wading into it every day.
[33:52.4]
Stephanie Aliaga: Yes, that is a great question, and it is the most challenging one.
[33:56.1]
I think it depends what you’re using it for. Most hallucinations, I’ve found, if you know something about what you’re talking about, if you’re using AI for something you already have familiarity with, you can sniff this stuff out quite easily.
[34:12.9]
Beyond that, what I would say, and what I’ve asked my models to do, is I ask them to give me the sources. I click through the links. Like if I’m actually using it for something research-wise and I’m going to go print out this number, I’m going to go cite this stat on TV, I double check. Double check the sources.
[34:29.1]
Sometimes I’ll just ask it. I’ve given it a skill, for instance: verify all of this. You can have it verify itself and do that rubric for you already, and they’re going to catch their own mistakes. These models, they get better with iteration. So I will say, use it to iterate. Use it as a first step, have it verify itself, and then go in and do some of the checking on your own too.
[34:50.5]
The other thing I would say, and this is slightly different than hallucinations, is the more that you use these tools, the better you get at catching this. And the better you get at catching things like AI slop. If you ever use AI for writing, there are some very easy telltale signs of AI slop or AI writing habits, and people don’t like it.
[35:08.7]
People are also really good at catching it. So the better versed that you are in it, the better you will be at editing all of that away, or even training your AI models to not give you that kind of stuff, which is what I’ve been trying to do. But the other thing I would say is, you want to make sure you are in the driver’s seat.
[35:26.9]
You are not just asking AI and offloading the thinking process of what you’re doing to them. Use them to help you structure your thinking. Use them to aggregate research for you or aggregate ideas. Brainstorm.
[35:42.3]
Use them as that co-thought partner. But when you start giving away decision making to them, that’s when things get very dangerous, because of these hallucinations and because they don’t know all of the great things that you know, right? So that’s what I would say.
[35:58.1]
That, and just verifying the data, the stats, and always approaching things with this healthy dose of skepticism will all be to your advantage.
[36:09.6]
Steve Lee: So we can channel Ronald Reagan with “trust but verify,” I guess.
[36:14.3]
So that’s good. Thank you. We do have a question that just came in. What is the industry’s response to all the backlash that is now coming around AI, and how does that affect the trajectory for AI growth?
[36:32.5]
Is it going to slow it down? Is it more press than reality?
[36:39.9]
Stephanie Aliaga: So, really great question. So let me speak specifically to what I think has really dominated headlines in recent weeks, which has been
[36:49.6]
Dario Amodei, who’s the CEO of Anthropic, writing this letter that we need to pace the frontier, like we need to slow things down. And this came after a few weeks of really strong public discourse around the fears of AI, the 10% chance of humanity ending because of AI, all of that, and obviously this growing tide of discomfort and backlash against data centers.
[37:14.4]
What is this going to result in? I think it’s already resulted in a few things. One, the model labs are taking this more seriously. Now, fears of human threat and all of this, this didn’t come out of nowhere. Many researchers from these labs and ex-researchers from these labs have been sounding these alarms for quite some time.
[37:35.6]
The urgency has come from the fact that in these training environments this summer, incidents have happened where they’ve lost control over these models, and they’ve realized, oh wow, the time to slow down is now, because we realize we don’t actually have the control that we thought we had, even at the current state of these models’ intelligence.
[37:53.8]
So that’s part of it. And what they’re doing now, obviously, is asking for regulation. I think these companies know they’re companies, and game theory is going to make them want to race to be the best, but they’re also people and they’re also humans.
[38:09.7]
And I don’t agree with the cynical takes that some have shared that they’re just trying to bolster their IPOs or whatever. I think some of this is actually genuine, and so the united front amongst these companies is really encouraging to see. And there are some actions that they can take on their own to self-regulate, but it is hard when it’s inconsistent with the administration and the federal government’s push to just continue charging ahead.
[38:37.4]
So, at the end of the day, I think we’re probably going to see some progress. It’s probably going to be focused on just these few companies, the frontier labs. It’s unlikely to stop progress, but it will probably slow things down. And I think slowing things down is not just a matter of, hey, slow down because of safety, but slowing down also makes sense for these companies, economically speaking.
[39:00.6]
They don’t get paid by training models; they get paid by selling them, by inference. And if they can slow down that pace, they can focus on inference, on optimizing their models for people and companies to use, and I think that’s increasingly what they’re doing. So there’s that.
[39:16.8]
The other thing is, what is the public backlash doing to the data center build-out overall? I will say there has been a huge increase in moratoriums around these data centers being built. I will say when you actually read the fine print of these moratoriums,
[39:34.2]
it is not so much “we’re not building this data center,” but it is rather policymakers being like, “Hey, hold up, let’s make sure the terms of these agreements actually protect our constituents.” And I think what is likely to come from this backlash is probably good.
[39:52.5]
I think for all of us as ordinary taxpayers too, it’s that the hyperscalers end up shouldering more of that cost burden than the local communities. They’re further incentivized to reinvest in communities, to focus on job creation, and these data centers are hiring a lot of people, so there is that to their end.
[40:11.1]
And it’s also shifting where data centers get built. It may result in a delay of data center capacity, and if that is the case, it could mean that AI becomes more expensive, or doesn’t become as cheap as it would otherwise, which might mean that we can’t move as fast as we’d like to in adopting it.
[40:28.5]
So a number of different ways that we can take this here, but a little early, I guess, to say for now.
[40:35.3]
Steve Lee: Okay, well, thank you, Stephanie. I think we’re going to wrap up your portion at the moment. And I’m going to turn it over to Manish in a second, but again I want to say thank you very much for your time and your insights this morning.
[40:50.8]
Wonderful to hear from you and we appreciate it. So thank you again. Manish, I’m gonna hand it to you now.
[40:57.8]
Stephanie Aliaga: Thank you.
Manish Upadhyay: Great, thanks Steve. I’ll apologize in advance. My presentation will not be as exciting and enthralling as Stephanie’s was, but hopefully you can bear with me here for the next 10 minutes as I give a quick update on where we’re at in the economy, markets, and potentially what to look forward to in 2027.
[41:21.6]
Starting with market performance, we’ll look at the U.S. Again, another robust year. The one thing to highlight here is, as Stephanie mentioned and we all have noticed, over the last five or 10 years the market’s been pretty dominated by large-cap tech stocks. Under the hood this year, and even going back a little bit to 2025, small-cap stocks have actually performed a little bit better. So you can see here the Russell 2000, which represents small-cap stocks, has outpaced the large S&P 500 so far year to date. This is again another sign, as Stephanie was mentioning too, that although large-cap growth stocks have really pushed the market, there are going to be beneficiaries across the board, and we can see a little bit of a change here in the last 12 months.
[42:10.8]
Similar story in the foreign markets. 2025 was a nice rebound for international markets versus the U.S., and that has continued so far in 2026. You can see that especially in the emerging markets category, up almost 20%. So again, I think these last two charts really emphasize the importance of diversification.
[42:30.3]
We haven’t seen it play out very well in the last decade, but over the last two years it’s been very prudent, and we’ve benefited from being disciplined in having allocations to all markets, not just the ones that we see on the news, which are the S&P 500 and Nasdaq.
[42:51.2]
Moving on to where we’re standing with valuations, it’s a good indication of where we’re buying in today. You can see on the left-hand side, which is interesting, at the beginning of the year with the blue diamond, in the U.S. market the P/E ratio, or valuation, was 22.
[43:09.0]
Even with the increase it’s had this year, it’s actually come down. So typically what we’ve seen prior to probably this year is the overexuberance of the tech stocks. The price has just been going astronomically high and earnings haven’t really kept up with it. But now we’re seeing, as Stephanie mentioned as well, that earnings are fantastic in the U.S.
[43:28.6]
So even though we’ve seen double-digit growth in the U.S. markets, earnings have kept up and actually outpaced the price, which has brought valuations down a little bit closer to historical averages. You can see the same dynamics happening in the EM market, even with the big run-up this year. As of the beginning of the year
[43:46.4]
the P/E ratios were 13, but now they’re even undervalued. So, looking prospectively going forward, the right-hand side breaks it down a little bit more in detail with the different sectors involved. And you can see, as the chart says, historically, if you see that rectangle there where the index is, the international markets typically trade at a 20% discount to the U.S.,
[44:08.9]
and currently it’s at 30%. So going forward, we’ve seen a lot of positive economic growth on the U.S. side. We’re seeing really favorable valuations on the non-U.S. side. So going forward, equities are in a pretty reasonable position to provide the returns that they have the last three or four years.
[44:30.5]
To the bond markets. So Steve mentioned in the opening that the bond markets have had a little bit of a tough time so far in 2026. You can see from this chart the year started off great. If you recall, in late 2025, the Fed made some rate cuts.
[44:46.4]
Inflation was really getting close to their target, and all signs pointed to a couple more rate cuts in 2026, and inflation would have come down and everything would have been exactly as planned. Unfortunately, there was the conflict in Iran that broke out in March, and you can see from this chart, as bond returns were close to 2%, they obviously diminished from that time forth,
[45:09.2]
a lot from then till now. Obviously oil prices have a big effect on inflation fears among investors, so yields started going up. The Fed was really hesitant to make any moves during the year. I think, as you probably all know, throughout the news there was a back and forth between potential resolution and then back to the drawing board.
[45:30.2]
So you can see the path of bond returns has been pretty whipsawed throughout 2026, and just lately in September there’s been a really high increase in yields. Obviously, the Fed did increase the Fed funds rate by a quarter point a few weeks ago. And that again is an inverse relationship.
[45:47.8]
As rates go up, price goes down. So you can see, as of September 15th or 16th, the bond market as measured by the U.S. Agg is down 1.5. I just checked today. Yields have gone even higher, and bond markets are down close to 3%.
[46:06.3]
The one gauge you can look at to get an idea of how your bond portfolio is going to return is the 10-year Treasury. Most of our JFS portfolios are invested in Treasuries and corporates, but really looking at the 10-year Treasury and the path it takes can give you an idea of the price fluctuation on your bond portfolio.
[46:27.3]
So the general relationship is, as yields rise, prices go down. So you can see from this chart, in the beginning of the year, yields were a little bit over four. They’ve crept to five as of mid-September. Now they’re five and a quarter. So within nine months, there’s been almost a 100 basis point, or 1%, increase in yield.
[46:46.8]
The way to look at how that affects your bond portfolio is duration. Duration is a measure of how rate movement will cause price fluctuation. So the measure here is that if rates go up 1%, the years of duration of your portfolio is how much price you lose.
[47:07.5]
So currently, the aggregate bond index, which is the bogey that we use for the bond market, has about a six-year duration. So what that means is, if rates go up 1%, you should expect roughly about a 6% loss in price, and we see that in the total return of the index.
[47:23.7]
As I mentioned, it was down close to 3%. So given the 1% increase in yields throughout the year so far, that means your price went down 6%. But also remember, you’re getting about 4% annually on those bonds. So we’ve offset some of that negative return of 6% with a 3% coupon, leaving us at that 2.5% to 3% negative return this year.
[47:48.2]
There’s still a quarter remaining. So all else equal, if we get no rate movement, we can expect another 1% in the fourth quarter to bring returns to about minus 2%. Although not exciting, it’s definitely a far cry from 2022.
[48:04.3]
If you recall, that was a really tough time for bonds, where the Fed had to raise rates because inflation was well over 8 or 9%. And you’re starting from zero, so they increased from zero to 5%, and that had a pretty profound effect on the bond market, and most bond portfolios lost between 10 and 15%.
[48:21.5]
So losing about 1% on your bond side, I don’t think is catastrophic compared to a potential bear market in equities. So again, as we’ll see on the next page, it’s really a good starting point as we look going forward.
[48:36.7]
So, the positive of rates rising, if you’re long-term bond investors, is you’re getting more yield. Yes, in the short term, you’re going to get some price depreciation, as we’ve seen this year, but for those of us that have a portion of our portfolio in bonds and are longer-term investors, this is a very good starting point.
[48:55.7]
So this chart just generally shows expected returns going out five years from your starting yield, and you can see the correlation is pretty tight. Obviously it doesn’t happen all the time, but a 90% R-squared says that it’s a very good indication that your starting yield will really indicate what will happen five years from now.
[49:13.9]
And given the fact that it’s at 5%, flash forwarding five years from now, there’s a high probability that your bond returns will be around that range as well. The one thing to illustrate too is that I was talking to a client the other day and they were worried about interest rates increasing, and they should be, it does decrease price. But it wasn’t too long ago, six or seven years ago, when the 10-year Treasury was offering a half a percent or a percent, and if I told you back then you could get 5%, I think
[49:45.0]
everyone would have backed up the truck and loaded up on there. So, I think right now, if we look forward, we’re in a pretty good position for the bond market for future returns for five or 10 years.
[50:00.4]
Economic update. So we’ll look here at two measures. The first one you’re looking at here is the U.S. Index of Consumer Sentiment. This is a survey conducted by the University of Michigan. It’s about 50 questions.
[50:16.8]
They interview about a thousand households every month, and it ranges across the U.S., in different regions, different economic backgrounds, to get a general pulse of how the common person is feeling about the economy and future prospects.
[50:33.4]
And you can see the trend over the last three years has gone down. So as of right now, the public is a little concerned, and rightfully so, right? We have a conflict in the Middle East with no resolution in sight, which has been going on now for about six months.
[50:50.8]
Rates are increasing. There’s obviously a lot of talk about AI, which we just talked about, and the threat of it taking jobs and what it will mean going forward. So people are a little bit unsure. And then lastly, there’s really the inflation. It was coming down and now it’s increased, and there’s a lot of uncertainty in the market.
[51:08.8]
And it’s reflected by this consumer sentiment. The other metric on this chart is the VIX. It’s a volatility index. It just really looks at the volume of activity in the market through options trading. They call it a fear index, and you can see there are spikes there in the last three years.
[51:27.3]
If you look at that one spike there in early 2025, if you recall, that was really as the administration started and the push to increase tariffs. That’s where it really spiked, and there was Liberation Day on April 1st that really sent the markets into a frenzy, realizing that tariffs are indirectly a tax and that would make the economy not well from a consumer’s perspective.
[51:54.1]
So that spiked, and then there was a pushback. Some of those tariffs went off the table and you see the VIX settled down after that. Right now we’re at 17, so it’s within the normal range. But as you can see, these things come out of nowhere and it’ll spike
[52:09.9]
as developments happen. But overall the volatility of the market seems to be okay compared to how the consumer is feeling about the economy with the Index of Consumer Sentiment. All right, if we look at economic growth, Steve mentioned earlier that projected third quarter GDP is coming in at 2%.
[52:28.7]
It’s been great. If you look at the last three years, we’ve been pretty much at or above the historical 20-year average of GDP. Nothing really shows any signs of slowness. With some of the information that Stephanie shared with us earlier, it looks like the prospects could be better, depending on how this AI adoption takes hold, how it’s going to be incorporated into productivity, and at the end of the day how it’s going to affect GDP growth. But so far, all positive from an economic standpoint, and a lot of this is highlighted by earnings, which, as we looked at on the previous slide, have kept up with the price increases, if not more.
[53:08.4]
Inflation. Obviously, this has been in the headlines for the last four to six months. We were in a really good spot last year, as you can see from this chart: 2.3, very close to the Fed’s target of 2%. Things were settling down. There were rate cuts. Everything looked really positive coming into 2026.
[53:25.5]
And then obviously, there was a little bit of a 180 with the breakout of the conflict in the Middle East, and again, that really is simply that the oil supply shrank, prices went up, and oil is a very, very big factor in how inflation is calculated.
[53:42.4]
And really, there have not been signs of any resolution. So what was shaping up as a really good, easy year potentially for the Fed, the tables have turned, unfortunately. And now there’s a question of whether the Fed increasing rates will actually bring inflation down going forward.
[54:05.1]
Labor market. So we’re looking here at the last three years, and the labor market is fairly stable. Obviously there haven’t been a lot of payroll gains; they’ve been on a decline. They just came out with a report today and job gains were about the same, slightly less, so nothing’s increased there. The unemployment rate is pretty stable at 4 percent, below the historical averages. The one thing that is getting a little bit tight is wage growth.
[54:30.0]
You can see it’s come down. It was well over 6% back in 2021, and we’re creeping close to 3.2, below historical averages. So overall, not extremely strong and not extremely weak. We’re going to see how it plays out for the remainder of the year, but all else equal, it’s in a neutral status.
[54:51.3]
All right, so looking at inflation and looking at unemployment, those are the two things the Fed looks at for what their next steps will be. This is a chart that gives a probability of what they’ll do in future meetings. We just had a quarter point increase a few weeks ago.
[55:07.4]
So right now, if you look at these probabilities, it looks like going into the remainder of this year and into January, there’s a very high likelihood of a quarter point to a half point increase over the next three meetings. As you can see, at the end of January, there’s a pretty high probability of it being four and a quarter on the lower end, which is 50 basis points higher than it is now.
[55:32.8]
Again, take it with a grain of salt. I just mentioned at the end of 2025, there were very high probabilities that there were going to be cuts, and obviously it’s completely changed, and now we’re looking at a very high probability of hikes. The Fed has been on the hot seat this year. There is a new chair.
[55:53.0]
He was fairly cryptic on forward guidance. There’s pressure from the administration. And the key, I think, the hard thing to determine, is where is this inflation coming from? Is it supply driven or demand driven? And many people think that it’s a supply issue, which has been with oil.
[56:11.1]
And by increasing rates, it’s not going to really bring inflation down, because there’s not excess demand or excess consumption. It’s just the fact that there’s lower supply, so prices go up. So, increasing rates may not be the solution. But again, they’re pretty much in a very difficult situation right now because they have to wait and see.
[56:30.8]
And the real wild card for it to come down is if and when we have a resolution in the Middle East. That will ease a lot of tensions and make things a lot easier as those supply chains open up, and theoretically inflation may come down on its own.
[56:49.4]
One last slide here to go through is U.S. government debt. I know it’s been in the headlines. We’ve had a lot of clients call in very concerned about where the U.S. debt is. This is a chart showing credit default swap spreads. Those are basically insurance: if you’re loaning someone money, you can pay for these swaps to make sure that if that issuer fails to pay the debt, you get covered.
[57:16.1]
You can see over the last three years the problem hasn’t gotten worse, based on what these insurance costs are on government debt. I’m not saying it’s a problem. It’s been a problem. It’s been a problem for a long time. It just really hasn’t suggested anything’s gotten worse, even though there have been a lot of headlines indicating that it has.
[57:36.7]
All right, to wrap up here, looking forward to 2027. Again, the economy and GDP are positioned for continued positive growth. Obviously some pockets of the economy will be a little bit different, but overall all signs point to things going well.
[57:52.9]
The labor market remains stable, and as I mentioned, it’s really going to be down to inflation. I don’t think any real direction is going to happen until there’s some resolution in the Middle East, and unfortunately right now there are no signs of the two parties coming to the table to make that happen, so that’s going to be constantly on the table for the remainder of the year, probably into 2027.
[58:17.3]
Looking at it from an equity standpoint, stocks are strong. There are some rich valuations, but nothing extreme, and earnings continue to be robust. I think you should be happy on the equity side. Cash and bonds: again, if the Fed decides, based on that probability chart, to increase rates about 25 basis points or 50 basis points, you’ll see your money market yields increase again, maybe get back to 4%.
[58:41.8]
So for those of you that have short-term liabilities that need to get paid for, holding some cash is a reasonable strategy given where rates are and potentially would go. Again, bond returns. Unfortunately, it’s been a story of these inflation shocks, supply shocks, the oil that’s caused these yields to go up.
[59:01.2]
That’s really caused the total return to be flat or slightly negative, but for those that are long-term and have some patience, it’s a really nice situation to be starting at a 5% 10-year Treasury yield going forward. So, with that, Steve, I will turn it back over to you if there’s any Q&A.
[59:23.5]
Steve Lee: Okay, well, I think we’re right at the top of the hour, Manish. Perfect timing, so I want to be sensitive to the time of everybody who has come on, and our promise to keep it to an hour. If you submitted a question and we did not get to it, we will get to it if we have attribution, and there’s at least one that I know of that we’re going to circle back on.
[59:48.2]
But I want to thank everybody for joining us again. Thanks to Stephanie, although she’s not on. Manish, thank you for your slides and your outlook. And if you have any questions, any concerns, always reach out to your JFS team, and they will help you get answers to those questions.
[60:07.3]
Okay, thank you again, everybody. This will conclude our webinar. Have a great rest of your day.
[60:12.7]
Manish Upadhyay: Thanks, Steve.
