Schwab’s Sonders on the Biggest Mag 7 Investing Myth and AI

Liz Ann Sonders, Chief Investment Strategist for the Schwab Center for Financial Research, pulled back the curtain on the disparity between individual stock performance and contribution in major indexes at the Schwab IMPACT 2025 Conference in November with ETF.com's Dave Nadig. Their conversation covered why the AI bubble is different from the dotcom bubble, the role that AI can play for companies, and more. The following is a full transcript of their discussion.

Transcript

Nadig: Liz Ann, thanks so much for taking some time. You gave a heck of a great opener yesterday for the conference here at Schwab Impact, I should say. You dropped this one line when we were talking about the AI bubble that Michael Burry has entered the chat. Tell us what you meant by that.

Sonders: And then my colleague Kevin followed up with, “Maybe it was entered the chat GPT”, which I thought was funny.

Nadig: That's a clever line, too.

Sonders: So, you know, for obvious reasons, having been the man behind the Big Short, it was disclosed – well, first of all, he did a couple of cryptic posts on social media.

Nadig: As he is wont to do.

Sonders: As he is wont to do, and also through industry disclosures, a significant short position in Palantir and Nvidia. And I think there was already a little bit of angst about stretched valuations and the week, last week when we got five of the Mag 7 that reported and, you know, almost 90% growth in CapEx spend. How do we continue to finance this? What's the return on those investments? All of those obvious concerns. So, I think there was already a little bit of anxiety, and then I think the Michael Burry news, at least for a day, caused some dislocation in the market.

Why This Bubble Is Different

Nadig: Became part of the story, I think, that, “Now it's a bubble because the bubble caller has said it's a bubble.” But under the hood of that, you also talked, I think very cogently about how, if we're in a bubble of spending, it's not coming from the usual sources. People aren't issuing tens of billions of dollars of corporate debt to go buy servers. They're spending the cash that market pundits have been saying, "What are we doing with all this cash?" for years. How does that change how we should think about bubble dynamics?

Sonders: That though is starting to change. So, yes, this boom so far has been largely financed by equity or out of cash flows. And these are the companies at the heart of this boom, unlike many of the companies at the heart of the dot-com boom, are big, huge companies with strong top-line growth and bottom-line growth and massive margins and strong balance sheets and ample free cash flows.

But to sort of “pick on” the Mag 7, just because it's the most popular cohort that we all talk about, free cash flow growth for that group of seven companies, six quarters ago, was more than 60% year-over-year. The last two quarters in a row, it's slightly negative. Even though that cohort represents 33% of all S&P 500 CapEx. You've had some high-profile discussions from the likes of Oracle about some margin compression. Some of the stocks that have been hit disproportionately through earnings season, some of those stories have been about little weaker than expected margins, more debt financed deals.

Now, deals now being debt financed, even in the circularity of a lot of this financing, isn't some sign of impending doom. This could be 1997. There still could be a long runway. It's just a different backdrop. So, I think there will be and should be attention on what these debt deals look like. Obviously, what the demand is. There's a lot of buyers for that debt, and so again, it's not – it was not a comment made to raise alarm bells of some impending doom, but just to describe how we're morphing to a different backdrop in terms of the financing of this spend.

The Mag 7 Paradox Some Investors Miss

Nadig: Another thing you pointed out recently, along those same lines though, has been the sort of increasing concentration at the top of the food chain. We've had the spread between say the equal weighted and the cap weighted S&P is at particularly long highs. We've had dispersion that's all over the place. You pointed out that contribution does not equal price performance, and as individual investors, we have choices. Tell us a little bit more about that.

Sonders: Yes. So, I think there's conflating that happens when people hear about, talk about, focus on cohorts like the Mag 7, they conflate contribution with performance. So, and I always use an example, and this is a real-world today example, but these numbers will change every day. So, Nvidia is the best performer among the Mag 7 year-to-date, but it's ranked 40th in the S&P for price performance year-to-date, meaning there's 39 stocks in the S&P outperforming the best of the Mag 7.

Now, Nvidia is the number one contributor to S&P returns, but that's not the same thing as saying it's the number one price performer. And in fact, in the Nasdaq, Nvidia's ranked 495th in the Nasdaq. So there’s 494 stocks in the Nasdaq performing better than the best of the Mag 7.

Nadig: Which seems very counter. I think that surprises a lot of people. Is that just a function of the fact that we consume this narrative as opposed to looking at our actual portfolios?

Sonders: I think it is. And, quite frankly, if you look at what has been performing well, if you look at a top 10 list of best performers in the S&P 500 year to date, there's a lot of tech and AI related names in there. You know, Palantir is on the list. You've got some utility names, Western Digital is on the list. I think Broadcom and AMD have been on and off that list. Robinhood actually is the best performer. That's not really an AI story. But then you've got Newmont Mining has been on the list at times. Eli Lilly, GE has been on the list.

Nadig: It's been all sorts of weird pocket. I mean, we've had GLP-1s, we've had gold, we've had all these stories.

Sonders: Exactly. So, I just think that even though only 17% of the S&P's constituents have outperformed the index itself over the past year, there are still opportunities in the market aside from – and that doesn't mean don't own names like that. What I've also talked about, you're not benchmarked as an individual investor against the S&P like institutions or fund managers are. They are indeed at the mercy of the construction of these indexes. They are doomed to underperform if they don't own the biggest names with similar concentration in their own portfolio.

That's an institutional problem. That's not an individual investor problem. So, that's not to say don't own these names, but use disciplines around rebalancing and profit taking and not putting all your eggs in either one basket or one collective basket, because there's a downside to it. You know, memories tend to be short, and in the corrective phase from mid-February to early April, that in the case of the S&P was just shy of bear market territory, but the Nasdaq well into bear market territory, same with the Russell 2000.

Those names were disproportionately hurt, and that's what dragged the indexes down. So, there's another side to their lofty concentration. In fact, when Apple was at its low point this year in year-to-date performance – it was down like high single digits, I think, I could be wrong on that – but whatever its low point, its contribution rank to the S&P was 500th. So, you didn't have an implosion in the stock. It wasn't in bear market territory, but you were multiplying weak performance by a huge cap size. So, it works in both directions. And it's also the case, by the way, that only three of the Mag 7 are outperforming the S&P now year to date. So, there's even dispersion within.

Nadig: And even earnings wise, it’s been four to seven.

Sonders: Yeah. I mean, last week you had five of the Mag 7 in very different outcomes. And not just in terms of what they reported but what the price reaction was.

Where AI Is of Value to Companies

Nadig: Well, so much of the conversation around earnings has been AI, whether it's been the CapEx spending or it's been companies using AI, or even some talking about slowing down hiring because they don't need to because of AI. Let me turn this back to you as an employee of a company. How are you thinking about AI in your processes? You're a writer, you're a researcher, you're an on-air personality, lots of places I could imagine you using AI. Where are you really?

Sonders: So, in the moment queries I use all the time. However, I don't assume that the data I'm going to get is truly trustworthy. Hallucination rates are not high, they're, you know, low single digits, but it's still a hallucination rate.

Nadig: It’s a lot for financial data.

Sonders: And I think one of the analogies that's been used to describe sort of AI in the workplace and large language models in particular – it's like hiring an intern. They do a lot of the stuff that makes your life easy and a lot of stuff you don't want to do. So, I think one of the things that we're increasingly talking about as it relates to the adoption of AI, how companies are using it, and the grave concern about displacement of… So, yes, there will be creative destruction, there will be job displacement, no question about it. But I also think other jobs that haven't existed will be created. That's what happens when you go through a period of innovation.

But I also think that this is not universal, but AI is more about replacing tasks than it is replacing entire occupations. And what you don't get via AI, especially large language models, is a lot of C's: context, creativity, community, compassion, common sense. So, I think that, I actually think that companies that don't figure out how to bring it into their workplace are more at risk of failure and losing jobs than the ones that do embrace it.

Nadig: Let me throw this idea back to you. A few pundits in the market have said their position on AI is, “Yes, it's going to be an incredible transformative technology, but it's not going to be distributed the way we think it is. It's actually going to benefit small and medium businesses more because they have more leverage to use it and fewer constraints in corporate culture and process.” Do you think that's true?

Sonders: Yes, I think that’s valid.

Real World AI Applications

Nadig: How does an organization like Schwab, which, I hate to say, is not a small business anymore.

Sonders: No, no, not at all.

Nadig: You've got to be wrestling with that like face-to-face.

Sonders: On the one application that we've been working aggressively on that is most relevant to the world at Schwab in which I live, Schwab Center for Financial Research, is what we're calling Schwab Research Assistant. So, it's a LLM that eventually clients will be able to ask questions. And it scrapes our content as opposed to just scraping the entire internet. It's scraping our content.

Nadig: Which makes the hallucination zero because it's just your content.

Sonders: Right, because it’s our content. And we're working on keeping that content robust and up-to-date and gearing it toward the types of questions. So, we're in pilot phase. I've been playing around with it a lot.

Nadig: That’s exciting.

Sonders: And it is, but it is productive, it's fun. I will sometimes throw in a report that I've written and it's been more experimental at this point and say, "Write a seven sentence conclusion." And it's pretty good at that. I wouldn't just put a bunch of bullets in and say, "Make me a report for next Monday," but I've played around a little bit with that. You know, taking something that is a bunch of bullet points and just say, "Make me a report for next Monday," but I've played around a little bit with that. You know, taking something that is a bunch of bullet points and just say, "Hey, write a couple of paragraphs on this," even if I'm just using it for myself or internally, and it's helpful. It's also kind of fun.

Nadig: Arguing with it. “Tell me what I’m wrong about.” I always find that helpful.

Sonders: I remember when I first played around with ChatGPT, I thought, "Well, let me ask it a question that I definitively know the answer to," so I could check it. And I thought, "Well, let me ask a question though tied to something that's out in the public domain." So, I said, "Write a bio on Liz Ann Sonders." Not that I think I'm the most famous person in the world, but my bio is out there in the public domain because when I speak at events, my bio gets published. It's on social media. So, I thought, “Yeah, ChatGPT is going to nail this.”

And I tried seven times, and each one got something different wrong. And the thing it got wrong every time was my age, and it made me older every time. So there was a period there where I was like, “I’m out.”

Nadig: Liz, it’s been an absolute pleasure. Thanks so much.

Sonders: My pleasure. Thank you for having me.

Discover the news, data, and voices shaping the ETF community. Follow along here.