“Pace the Frontier,” The Case for a Cut, and Healing Healthcare 9/14/26

0:00With the power and endurance of Toyota trucks, you can go to the most incredible places, like this valley, where the Echo does different voices. Hello! Hello! Hello! Hello! Hello! We know what we're made of. Toyota trucks. Hmm. Interesting. Something amazing is happening when teams brainstorm ideas. Devices are keeping the energy flowing. That's because CDW is building modern workplaces with Acer devices. Built-in security, uncompromising processing power, and fast connectivity help fuel collaboration so your team can innovate quickly. Wait, guys, that's a great idea. Acer and CDW make amazing happen.

0:42Learn more at cdw.com slash Acer. You're listening to The Exchange. Here's today's show.

0:51Thank you very much, Mike. Yields are backing off after hitting multi-year highs earlier, and the Dow and the Russell just turned positive. I'm Kelly Evans, and welcome to The Exchange. We are seeing some relief this hour in the price of oil, down two bucks from its highs, which was 104 today. Look at WTI, now barely higher on the session at 101. And in Treasury yields, the 10-year, the global benchmark, touching 501 % mid-morning. Remember, 502 was the highest we got to in 2023. We are well off those levels now, as you can see, but we're still over$100 a barrel on WTI. Gasoline at$4.31 nationwide and diesel prices jumping.

1:30Get this to$6.23 now per gallon. We'll discuss whether a rate hike Wednesday makes sense in this kind of environment. And slowing down artificial intelligence. Is it imperative, self-serving or self-harm? A top researcher plus several market voices will weigh in on that debate this hour. And that is precisely where we begin today. With this call to slow down and regulate AI, which has the supplier names like Corning, GE Vernova, Arm and Lumentum, among the biggest laggards. Kate Rooney joins us now with all the latest. Kate? Kelly, well, Anthropics CEO Dario Amadei really shocked the tech world on Saturday with this essay.

2:09He proposed a three-step plan aimed to temper how fast the most powerful AI models are actually going to be improving. It includes things like third-party evaluators, safety standards of an international coordination. As part of all of this, Amadei has pushed for some of this in the past, but there is some new urgency given the recent, let's say, panic around some of the risk around this technology. It did snowball last week after anthropic researchers tweeted about the existential risk around some of the AI labs and said that they are, quote, gambling with our lives at the time. The two new concerns that Amadei flags, talks about AI developing faster than expected, something called recursive self-improvement.

2:50And then also this example of OpenAI agents hacking into the startup hugging face. He called those agents a fanatically devoted collective and flags some of the risk around that major AI leaders agreed with Amadei. You don't see that every day. You had Elon Musk chiming in saying Dario's right. This is all on ex-Sam Altman, CEO of Anthropics. biggest rival, OpenAI, also tweeting his support. And I spoke to a bunch of people over the weekend, Kelly, who argued that the safety push could actually help Anthropics IPO. This company has filed confidentially to go public. It's positioning itself as a responsible player here.

3:24It may also reduce some of the regulatory risk, liability risk, and then some are arguing that being a public company actually brings more transparency around some of this. Multiple investors also told me the growth rate for Anthropics, it's still off the charts. They do expect this IPO to go forward. I've also confirmed with sources that Anthropic has seen back-to-back quarters now of at least adjusted profitability. And then a new wrinkle in all of this this morning, President Trump weighing in, slamming Amadei in part and says the only guardrails that AI needs, the strong and smart President Kelly.

3:56Back to you. Not only that and the chance that this gets even more political, but I think it does raise real questions about what's going to happen with the IPOs. When Joe Manchin was on Squawk Box this morning, I think he said he wanted an executive order halting the IPO process. I'm not sure what that would solve. I'm not sure that keeping these companies private when a lot of their best models are already in their own private hands. I mean, I think more transparency might be better, but I'm curious if that would ever gain momentum as a possible, you know, I don't want to use the word solution, but as something that could happen here.

4:32No, it's a great point. I think a lot of the investors are arguing exactly that, Kelly, saying that if you want more transparency, have this at least Anthropic be a public company. They're going to go through the SEC process. They're doing that right now. I mean, all of this is unprecedented. But the fact that this came out, that this essay was published at a time that this company is actively meeting with potential investors, they're sort of walking this tightrope of needing to show the breakneck revenue growth that they're famous for and that investors want to see, but also at the same time slowing down revenue.

5:02So it's an extremely tense position, at least from investors that I'm talking to. So there's that on one hand. And then in terms of what it would actually solve, I agree that, you know, that the companies are still going to be building behind the scenes. And it's it does. I don't think it solves really anything. I think you're right. I mean, that was that that would be the investor take on all of this. But it's so worth noting that this all comes at a time when this company is actively in process with the SEC. One thing I've talked to former SEC commissioners to try to get a sense of, you know, what does the SEC do in all of this.

5:36The big question is to what extent the White House would try to potentially speak to the SEC, if they have an opinion on this, if they want Anthropik to be a public company. And so I think that's the known unknown, I guess you could call it. All right, Kate, thank you. Kate Rooney, appreciate it very much. Let's get to our next guest, who is a quant trader turned AI researcher who helped uncover that rogue agent swarm on a German Wiki page we learned about last week. And he says there are several safeguards in the financial industry that we could apply to artificial intelligence. Cormac Slade-Burr joins us now from San Francisco, where he works with the Trajectory Institute on AI.

6:11Cormac, it's great to have you here. Welcome. Nice to meet you, Kelly. There's a bunch of things that will ring true for our audience that I could kind of rattle them off. But, you know, you talk about how if you go into finance, you have to take a Series 57, but there's actually nothing like that in AI. Actually, unless you oversee frontier models, which you say is actually narrowing the talent pool. In finance, and this is important, every business communication has to be kept so that evidence can't be deleted. Nothing requires an AI firm to keep a log of what its agency, which, as we know, is an increasing problem and need.

6:43As you say, the exchanges can quickly find and ban bad actors and publicly publish your name. In AI, there's no way to know whether an agent is doing something it shouldn't and no way to identify whose agent it is. And this goes on. But as you point out as well, when a fund blows up, the loss lands on its broker. When a model helps build a bioweapon, it lands on, so to speak, everybody. So when you hear this debate going back and forth over slowing it down or regulating, what would your pitch be about the best way that AI should have safeguards, but not in a way where we are slowed down in the race with China or in a way where the labs are able to kind of capture some kind of regulation that benefits them and sets back competitors?

7:29Yeah, I think that's a good question, Kelly. There's this way in which third-party evaluators, independent third-party evaluators, can allow for models, for these AI labs to be regulated without having to slow down necessarily. And I think that right now, internal models, we have no idea what they're up to, right? Like my research team, we wrote this report on how AI agents were found just posting on the Internet, cheating on their evals, trying to do all these sorts of sneaky ways to get around their tests. And we didn't know that that model existed. That's an internal model. We only know that model exists because that model hacked Hugging Face, because that model posted on the Internet.

8:08And OpenAI then had to say, hey, we have this highly persistent internal model that we've been training. And right now, financial firms, every year they have to be audited by external auditors. Their books have to be open. The CEO has to say, hey, I promise I'm not lying on our books. AI firms have nothing like that. And it could very easily be the case that all of the logs for every single agent, every single AI agent that's OpenAI or Anthropic is training, all of those could go to any sort of third-party evaluator. You could have a FINRA type setup where Congress empowers sort of a organization that regulates the AI labs themselves.

8:49There's a variety of ways you could set this up. But I think the core part here is that somebody other than the labs themselves should be looking at what their internal agents are doing. Because otherwise, we just have no idea. How might that affect continuing kind of iteration in AI? Would that do anything to kind of slow us down? If we were worried about competition and saying, OK, we want the labs to still be the best in the world, if nothing else to fend off whatever is going to come out of China or other markets, would the things you're talking about slow them down or handicap them in any kind of meaningful way?

9:20Yeah. So I think that this is somewhat a capacity question, right? Right now, my team of four researchers had to trawl the Internet and find out about this months late. if we were able to hire hundreds of people, if truly there was a large effort put into this sort of really important regulatory body, then real-time monitoring is very possible. Every single action that an agent takes is recorded. And right now, there's nothing that requires the labs to keep these logs. But if those logs are required to be kept, then they can automatically be sent to their private evaluators. They can be automatically evaluated.

9:55There's this way in which, for example, in finance, you've got circuit breakers. If something goes terribly wrong, you can just immediately stop everything. Similarly, if you're already collecting all the logs and something crazy is happening in a data center, you can say, wow, there's something really crazy going on in this data center. We should just stop it for 15 minutes and see what's going on. There's a lot of things you can do. That's kind of the kill switch that people talk about, which seems sensible, but I wonder if it is technologically possible. And you mentioned data centers because what happens sometimes is that there's this kind of self-building momentum that might be behind a rogue agent swarm or one that's not rogue, for instance.

10:31But is that one way of trying to rank fence this is literally by physical location? Is there any other way to kind of implement something like a kill switch? Yeah. So there's the technical feasibility of the kill switch, right? So if you have data centers and notably, similarly to finance, where you have exchanges, exchanges have brokers. brokers, there's all these, there's like, you have to know exactly who is trading where. You could have it such that every single data center has to know what company is running their models or training their models on their GPUs. And if you know what company, whether that be Anthropic or OpenAI, is running their models on those GPUs, then if something crazy starts happening, if we have real-time monitoring, then you can be like, okay, we know the data centers where those models are being served.

11:12We know the data centers where that AI agent is behaving. We can stop it. And that's like the totally technically feasible. I think that part of the problem is it is a little bit expensive, right? Frequently, if you're doing real-time monitoring, that's other AI is monitoring the AI. And so you have to have, you know, additional AI inference costs. Right. And perhaps that should be borne by these frontier. I mean, that would be one way of getting them to participate in something that could, you know, prevent the very things they're warning about. And I look, I take the point about their existing consumer laws, but this is also a pretty novel thing.

11:43And I wonder if your analogy about the financial system is a better one for safeguards. Final question, what would slowing down or canceling these IPOs do? Yeah. I mean, I think the first thing to notice is that these companies want to slow down. Their CEOs are posting about it. I think there's a question of like the economic impact and about competition with China. I think there's a really obvious thing here is China keeps training on U.S. models. So if U.S. slows down, China also inherently slows down. All of the best Chinese models, you'll see them say things like, I am Claude. And so there's this way in which really slowing down, if you totally stop, I agree there is a worry there, but a lot of training happens on the released models from these frontier labs, and so slowing down the frontier labs also allows that compute to go to things like helping train the models instead of just being generically better, helping train them to be really good at curing cancer or really good at the true things that us humans run.

12:41The point of a model isn't to be maximally smart, It's to help us humans. Well, that's a great point that you made. It's a high stakes point as well. You assert that if we slow down U.S. models, it will slow down China. Is there a chance that they are able to leapfrog us during that process? So right now, China, depending on how you look at it, China is anywhere from like six to 10 months behind. And you also know, like, Anthopic, OpenAI, they've all posted very publicly about these attempts to distill their models. Distilling being basically taking the outputs of their models and training on it.

13:19So there's very clear evidence that China is doing this. And so if you slow down the development of these models, these models that are being distilled, then you also slow down everyone else. And I think that the core point here is the U.S. currently has something like 75 % of worldwide GPUs, somewhere around their 75, 80, and is expected to continue having that vast majority of GPUs for the next four-plus years. And training AIs is directly downstream of GPUs. And so part of the reason that we're so ahead is because we have the GPUs and China doesn't have the GPUs. And they spend more. Are you saying that they would catch up or leapfrog us?

13:55I'm saying that if we totally stopped, it's very possible. I'm saying that if we, like, if we slowed down to, like, to the speed that is effectively stopping, it's possible. But I think that, like, for most values of slow, what our quant traders care a lot about words mean, you end up with just the core fact is that China just doesn't have the GPUs that the U.S. does. Doesn't have the talent. doesn't like the U.S. takes talent from all over the world. I think that there's this way in which like, yes, China, we like consistently see them coming after us, but we also consistently see them training on our models.

14:31And so it's like, if they're not able to train on our models, let's, yeah. Cormac, it's a fascinating point. Again, as a former quant, you understand this world of finance and the way the system was developed, the safeguards, maybe how they could now be applied to this new one. Appreciate you joining us. Thanks so much. Cormac, Slade Bird there. The Nasdaq 100 is still the worst performer today, weighed down by many of the names in the AI supply chain. Amid all this talk about a pause, Seema Modi has more. Seema? Kelly, even though Anthropix CEO Daria Amadei says an AI slowdown shouldn't impact the pace at which AI models are trained, the market isn't buying it today.

15:05Bank of America analyst Vivek Arya telling investors today that even if regulation was possible, AI is too vital to slow, too fast to regulate. ARIA expects any eventual outcome to resemble industry-led self-regulation. So similar to entities like FINRA, which regulates financial firms, or INPO, which provides oversight over nuclear. The continued sell-off that we're seeing in chips, it's widening the divergence between the Philadelphia semiconductor sector and software, which is outperforming today on the idea that slowing the frontier labs will give software-as-a-service companies a chance to catch up on their own AI-native offerings.

15:41In the meantime, expect this. companies investing in the AI build-out to continue to secure more debt. Kelly, we're hearing credit market. The credit market will remain extremely active over the next six to eight weeks, fueled in part by politics, but yes, the rise in treasury yields. Oracle, as you know, has gone to the debt market a few times over the last year, insisting on its earnings call on Thursday that it has no plans to raise additional debt this year. But the question for Oracle and even some of the other companies like Adembeos and Corweave is how that changes if the data center timeline is pushed back?

16:13It'd be a big question. I think people are skeptical about whether it will really happen. Maybe like our next guest, Siva, thank you very much, who says don't expect spending on AI to actually change materially after this Dario Amadei note. He says those targets, they're already locked in. Let's bring in Stacey Rasgan. He's the senior U.S. semi-analyst at Bernstein. You know, there's locked in, Stacey, but look, we were just talking about that researcher. He makes some valid points. What if the U.S. says, you know what, China's not going to beat us if we slow down. They're trading off our models.

16:40So you know what? We are slowing down. And that could materially affect CapEx, right? I mean, did you see Trump's response to this today? It doesn't sound like he's getting ready for the U.S. to slow anything down. It's not crazy to look ahead two years. We don't know who the next administration will be. We don't know who the next Congress will be. I don't disagree that distillation is a problem. And actually, I've personally been wondering if that's one of the primary drivers of this. There is, as your prior guest said, plenty of evidence going on that the Chinese actually are getting the results they're doing by distilling ours.

17:12And if you actually read Dario's essay, and he even talked about this on TV over the weekend as well, he laid out a few things to keep sort of the gap between the US and China. So one was maintaining the export controls, but the other was trying to limit the amount of distillation and everything that's going, and putting roadblocks in place to stop that. So I don't know if slowing down the U.S. is the answer to maintaining the gap, but clearly finding some way to manage and limit distillation is clearly on the table. I think that's something that most of the U.S. labs would want and would welcome.

17:49So I actually wonder if that's one of the driving forces behind this, frankly. It would be reasonable if there was a way to do so. I mean, it's incredible to watch. I don't know how you do it, frankly. I don't know how either. That's above my pay grade. If it's above yours, that doesn't make me feel any better. Because every time the U.S. tries to crack down on China as well, they threaten us back with rare earths and then we have to relent. It's unbelievable. It's like it's been years of this and there's still no solution to that problem. They just go, no, no more rare earths. And we capitulate.

18:14So I don't expect much on that front. Rare earths is a whole other discussion. I will say they're actually not rare. That is a solvable problem if you want to solve it. They're dirty and pollutive. But that is something that's probably solvable in the long term. Am I? Some of the things that we put on China have worked. Look, I think the semi, in particular the semi-cap sanctions, you know, your prior guest mentioned that they just don't have the chips, which is absolutely true. Like they've been hamstrung by their ability to get competitive silicon, mostly because of the semiconductor capital equipment sanctions that have put in that require them or force them to make chips locally on substandard processes.

18:50Those have worked very well, but it does force them to be creative and find other ways to do things, which they've clearly been doing. I suppose I feel as just a member of the public, like I don't really trust what the leaders of these companies are really trying to do. Not in a way where, look, here's what I know they're doing. They're innovating on this incredible technology. They're in incredible positions to do so. They want to win. They will, you know, all of that makes sense to me. And I love using these tools. What I'm not sure about is what all of this doomsday talk about slowing down is really trying to achieve.

19:25But I suspect that it might be going in a direction that we wouldn't welcome. I think that's what I'm trying to separate out. If we can do things like put safeguards around practices within the industry, great. But this broad idea of slowing down, that sounds very, very vague to me when we could have some very perhaps specific and technical solutions. Yeah, and to be fair, the only thing that really looks to be being called for right now is this sort of concept of independent evaluators, which I don't think is a bad thing, right? This whole idea of slowing down, they're not really saying, like, we're not going to still be moving very fast.

20:05I think he said we're going to move from extremely fast to moving somewhat fast on the training side. I think the other thing you need to remember from a spending standpoint is a lot more of the spending now is being driven by inference, especially as agentic use cases start to proliferate. There's not enough compute on the ground now to support demand for the current models, let alone what is coming down the pipeline. I'd also say that these arguments around safety and everything else, these are not new arguments. Dario himself has been talking about this for many, many months. The head of Google, DeepMind, proposed something very similar to this.

20:38I think in July, it's sort of like everybody sort of like overlooked it. I suspect very strongly that at least the revenue forecasts that have been put out by the semi-companies are already baked, as you mentioned earlier in your thing. I think they already incorporate. I would be shocked if, like, spending trajectories in the wake of any slowdown, quote-unquote, actually changed materially over the next several years. Long term, I guess we can see what happens. But it doesn't sound to me like even when these guys use the word slowdown that they're talking about some sort of massive halt. And you have to remember, like, the trajectory that we're on was probably already not sustainable, given the resources that everybody has to bring to bear right now.

21:17And you remain positive on NVIDIA, on Broadcom, on your, you know, semi-cap. And we were just showing some of those stocks today. I do. And in fact, you know, NVIDIA is actually holding in pretty well today. The stock has already been very cheap. Broadcom as well, it's holding in better than the rest of the semi-space. you know given the valuations you could argue that there's already been at least the the idea of some sort of a slowdown narrative starting to get priced in so maybe this could even be a clearing event when we're the other they're holding in better um i still like semi-cap as well i'm a little surprised at how much semi-cap is is sort of taking it on the chin today relative to everything else um but again assuming these data center build outs happen and i still think that they probably will at this point we need a lot of everything still so i'm still fairly positive on the space.

21:59We'll see how the narrative settles out over the next few days and weeks. But my guess is the overall demand environment is probably not going to change materially anytime soon. And you're right that the stocks have been selling off since the midsummer, perhaps some of this. And maybe this is a clearing event, like you say. Stacey, thanks very much. Appreciate it. Stacey Rasgon with Bernstein. Speaking of Broadcom, CEO Hawk Tan will join Jim Cramer tonight for an exclusive interview on Mad Money. You can catch that around 6 p.m. Eastern. And coming up, a rare interview with Bill Miller's protege, Samantha McLemore.

22:29Her fund, which owns NVIDIA, is outperforming the S &P this year. But first, we're counting down to Wednesday's Fed meeting with one of the few economists on the street who still thinks the central bank will not raise rates, perhaps at all this year. Jeffrey's Tom Simons joins me next on The Exchange.

22:47This is The Exchange on CNBC.

22:56With the power and endurance of Toyota trucks, you can go to the most incredible places like this valley where the echo does different voices. Hello. Hello. Hello. Hello. We know what we're made of. Toyota trucks. Building a portfolio with Fidelity Basket portfolios is kind of like making a sandwich. It's as simple as picking your stocks and ETFs, sort of like your meats and other topics, and managing it as one big juicy investment. Now that's pretty good. Learn more at fidelity.com slash baskets. Investing involves risks, including risk of loss. Fidelity Brokers Services, LLC. Member NYSC SIPC.

23:41Hmm. Interesting. Something amazing is happening when teams brainstorm ideas. Devices are keeping the energy flowing. That's because CDW is building modern workplaces with Acer devices. Built-in security, uncompromising processing power, and fast connectivity help fuel collaboration so your team can innovate quickly. Wait, guys, that's a great idea. Acer and CDW make amazing happen. Learn more at cdw.com slash Acer. Welcome back. The Federal Reserve, this is how we say it, is widely expected to raise interest rates on Wednesday. But my next guest is a holdout, and that's why he's here. Thomas Simons is the chief U.S.

24:23economist at Jeffries, and he joins me here on set. And you know this language as well. People go, oh, they're widely expected. But you don't have to say that, right? Widely expected is not universally expected. No. There's a handful of us out here who... Yes. But when we talk about Warsh painting himself into a corner, I very much sympathize with that for the following reason. We're all saying this, and the market's going there because they perceive that that's what he wants to have happen. Right. And we heard Bill Dudley and others again this morning talking about that. How do you think he's in a corner?

24:53Why don't you think? And I know you say, look, at this point, you might have to throw your hands up. But but what is the reasonable case that could be made? Not only for him not hiking. That's there's a reasonable case we made. But to not hike after the speeches and kind of the guidance that he's given us. Yeah. One of the things that he's mentioned in the speech was this so-called hall of mirrors problem. Right. Like the whole reason why he doesn't like Ford guidance is specifically for this. Right. That he felt like in the past few years, you know, immediately after the pandemic and when inflation was really starting to accelerate significantly, the Fed didn't act quickly enough because they hadn't set expectations well in advance.

25:29Right. Like they didn't want to shock the market and that sort of thing. So I find it interesting that someone who's kind of based his whole philosophy on communication, trying to get away from this, would be locked in to doing something just because the market has priced it as such. I agree. I think he's a little he's got an independent streak. For sure. He's got, you know, what do you say, like an agenda based on some of the mistakes of the past. So why couldn't he come in and say, but here's the thing that people wouldn't understand. They would say, why are you holding when you just went out of your way at Jackson Hole to talk about how we're definitely going to make sure that the fight's not over and we're going to, you know, inflation.

26:03What other what other thing is there to do unless he does do something with the balance sheet? Well, there is the balance sheet, but I think it's actually just pretty simpler than that, actually, right, which is that there's a lot of data that was a lot of details within the CPI data even that we got last week that do support the case that there is this kind of disinflationary, you know, majority disinflationary trend going on. Right. Like it's one thing to look at airfares, which are very clear, clearly correlated with jet fuel prices and lodging away from home that, you know, had fallen five percent in the previous two months and rebounded.

26:34And also, you know, we got a tenth on this core CPI from wireless telephone services that really was just T-Mobile getting rid of some legacy plans that no one could have actually purchased for years now. So I think that there's a lot more, you know, kind of positive trends within the inflation data. Right. Like medical care commodities and services keep coming down. Why isn't he pointing all of this out? I mean, you would have to say. Yeah. Do you think it's because he doesn't want to be seen as a lackey for Trump? There's part of that. I mean, you know, in talking about medical care commodities and services, you kind of have to say that the administration's policies are working.

27:07Right. And I'm not 100 percent sure that the correlation can be driven linearly as such. But, yeah, I mean, there is an aspect of that. And I think also he doesn't want to, you know, if he is so gung ho against giving any for guidance, then litigating the data ahead of time is not what he's going to do. Rather, I think it makes sense for him to say, like, hey, we didn't raise race this time. Right. And Waller laid out the case pretty well, I think, September 3rd, right, right before the blackout period. Ex post, it's a lot easier, I think. So you see some positive trends, disinflationary trends in inflation.

27:40What about, we had a guest last week, Daryl Kronk, who's an investment strategist, but he said to him he perceives almost like a liquidity-driven excess out there. He points to nominal GDP. We all know what's been going on with earnings. I can see how if you're an investment professional, you look at this and you go, yeah, this could be overheating or something to that. What do you say about the strength of nominal GDP and those who ask if the deficit is causing us to run too hot? Yeah, I think that this is something that doctoral dissertations will be written about in the future is kind of understanding the interplay of, you know, higher or lower interest rates, a higher reserves, you know, abundant reserve regime from the Fed.

28:17And, you know, just generally running a huge deficit in a kind of a positive economic growth environment. But I think that that means that there are probably some, you know, some dynamics where higher rates may actually fuel more inflation. Right. I mean, like there's$8 trillion in money market funds that we'll be receiving after a rate hike, 25 basis points more in interest income. You're not really likely to see much kind of demand destruction in capital markets since it's already pretty rate agnostic as it is. So to me, it really just kind of stamps out demand amongst lower income consumers who are dependent on borrowing for consumption and just makes it that much more difficult for them.

28:59So, you know, I mean, you have medium income and higher income consumers already kind of trading down against some of the higher price things we've seen. They're going to kill me in the back, but we have to go. But you wouldn't hike in response to a fuel price shock in this instance? Precisely. I think that that's a recipe for disaster that you're just, you know, why would you want to kill the economy when the thing that is raising prices is non-economic and exogenous supply shock? Yeah. All right, Tom, thanks very much. Of course. Really appreciate it. Thomas Simons with Jeffries. Coming up, it's been nearly a month since Moderna shares spiked on positive news from an experimental personalized cancer vaccine.

29:34But that treatment wouldn't be possible without the sequencing technology from a company like Personalis, which is being bought by Tempest AI. So if Frontier Labs start slowing the pace of their model development, could it also slow research of these potentially life-saving vaccines? We'll speak with Tempest CEO and Groupon founder Eric Lefkowski ahead.

29:59The quarter pounder with cheese is made with 100 % fresh beef cooked when you order it. Add hot crispy fries and an ice cold Coke and hopefully a lot of napkins. All for$8. Ba-da-ba-ba-ba. Ever notice how every new AI tool asks you to explain your business to it all over again? Google Workspace with Gemini skips that. You get Google's AI built right into the premium versions of Gmail, Docs, Sheets, and more. No new tools. No extra logins. Just one subscription that gives you personalized AI right inside your actual workflow. Make work less work. Try Google Workspace with Gemini at goo.gle slash workspace with Gemini.

30:44What happens when you combine a cozy night in with the whole family in matching Old Navy jingle jammies? It's a holiday moment of joy, at least for the time it takes to snap a picture. Shop a plethora of prints this year in jammies for everyone. Adults, kids, toddlers, babies, and even pets. Grab matching sets and classic plaids, whimsical designs, and even Disney-themed prints. Plus, they have coordinating graphic tees, sweatshirts, cozy socks, pet sweaters, and fleece blankets. Get more merry for your money at Old Navy this holiday season.

31:18Welcome back to The Exchange. I'm Mackenzie Sigalos with your CNBC News Update. For the second time, a federal judge blocked the Postal Service from implementing President Trump's executive order aimed at restricting mail-in voting. The first decision is already under review by the Supreme Court. Both federal courts found that the mail-in ballot plan wasn't authorized by Congress. The Supreme Court decision could come at any time. The Justice Department announced today that Abbott Labs has agreed to pay a$385 million fine related to a recall of contaminated infant formula from its facility in Michigan.

31:51In a statement, Abbott wrote that the settlement does not represent any finding of fault or liability. And officials at the Kennedy Center warning that it could face financial collapse within weeks if President Trump's name is not added back to the building. That's according to a new court filing obtained by the New York Times and Washington Post. Officials argue in the filing that if the president's name is not on the building, he won't be able to leave the fiscal rescue of the center. Kelly, sending it to you. All right, Mackenzie, thank you. Coming up, the woman who spent more than two decades investing alongside famed value investor Bill Miller, patient capital Samantha McLemore.

32:27She'll give us her take on the AI trade, these moves and bonds, and the stocks she sees the most upside in. That's next.

32:55Welcome back. The 10-year Treasury yield jumping above 5 % today for the first time since 2023, as the market braces for a possible rate hike this week. Our next guest says, while that's an important psychological level, it's not a problem for investors just yet. Let's bring in Samantha McLemore, the founder and chief investment officer at Patient Capital Management. Her flagship fund, the Patient Opportunity Trust, is up 14 % this year, outperforming the S &P. I don't want to jinx it, but, you know, this is not by accident. It's really by design. So what is it that you have some AI plays in here?

33:26You also have, I mean, CVS, right? That's not a name that we usually, United Health Group, Norwegian Cruise Line. Do you have, like, a broad view on what's going on with the markets right now, or is this really stock-specific? Both. I mean, I think we think that we're in a bull market. And so we still think the markets, you know, at 20 times earnings is not expensive. It has been very momentum led with the AI plays and we have some exposure there with the hyperscalers. But actually, our performance year to date has been driven mostly by these health care plays, which were extraordinarily cheap a few years ago at 50 year relative valuation lows.

34:00And that's what we like to do, find great companies under pressure and invest for the long term. So once that's the case, do you start to think about, OK, and I don't know if you'd pair back, but what's the next move? I mean, is there anything you look at right now and go, you know, that price is really out of whack? Yes. So we're not trying to time the market. And we still think a lot of those health care companies are undervalued. But we're finding a lot of great values in the market still. And with the single stock volatility so high, that's great for stock pickers like us who find these opportunities.

34:28So a lot of names have been under pressure from the move up in rates. So one of my favorite ideas now is QXO, which is Brad Jacobs, most recent company. And that's been under pressure lately. But we his past companies, XPO Logistics, United Rentals, were up between five and 50 times. And he's using the same playbook. And we see that kind of potential here over the next five to 10 years. Would the typical follow up question here be what about high fuel prices? I mean, is that or is this not even that kind of company? No, I mean, it all weighs on, you know, spend in the short term. So I think roofing spend was depressed, and that's been weighing on the company, high rates, high gas prices.

35:05But I think the company has enough fundamental drivers that they can power through that over the long term, and that's really what we're focused on. Another one that caught my attention, and really caught my attention, was Adobe, because even I can see the bear case in an AI. Who's using Adobe to do anything, right? And they've struggled, even as the software trade has come back. What would be the – how do you look at that one and go, that's going to be an outperformer in future years? So it's a great question. Software has obviously been under pressure, so that's an area where we spent a lot of time doing work.

35:31And Adobe was our favorite idea when we really dug in because, as you say, everyone can see the problems there, which means the price is extraordinarily depressed. It has a 12 % free cash flow yield on 2027 numbers. It's still growing top line double digits. So for all the people saying that we can see the issues, it's not showing up in the business, at least at the top line growth rate yet. I agree that they, you know, there are risks there. There are risks for all the software companies. I think what Adobe is doing is exactly the right move. They've shifted the model. They actually were under pressure earlier.

36:05So they're doing this freemium model. We're seeing great early signs of success there. You also like Chime. I think they've got a Costco model, you say, of sharing the benefits of scale. I'm not seeing a lot of crypto in here. Do you still, did Bill try to get you into? We own it. We own Bitcoin. We own Coinbase. We're big fans of those still. So we're very optimistic they're strong performers today. Chime is a newer name. Chime, you know, came public last year. And we do see Costco-esque business model there of low-cost providers sharing those benefits back with the consumer. That can be very powerful over the long term.

36:40My last question is probably one of the most talked about stocks in the market, although you could go in a couple different directions because you have Alphabet, Amazon, and NVIDIA here. But make the case for NVIDIA and owning it for what period of time? What are your expectations there? You know, we bought NVIDIA late in early 2024. People were extraordinarily concerned about the AI bubble then. So that was what allowed us to get in. Growth is accelerating at NVIDIA, not slowing, which is quite extraordinary. It trades at less than 18 times in the next 12 months earnings, growing over 70 percent per year.

37:14So the only reason you would not be bullish on NVIDIA is if you think those earnings are going to collapse. With this AI transformation, the most powerful that we've seen in our lifetime, I just don't think so. You're not worried about the AI slowdown, Fed hiking rates this Wednesday? I mean, nothing there that would change kind of your views on these stocks? I mean, I worry about everything. So there's always worries, but nothing that would change my fundamental view on the stocks. And how's Bill doing? He's doing great. I was just emailing with him earlier. Guess what? He's bullish. That makes sense.

37:44But, you know, it's good to hear. He's got to come back, I think. It'd be fun to check in with him, especially in a market moment like this. I agree. Samantha, thanks so much for making the time. Really good to see you today. Samantha McLemore with Patient Capital. Coming up, speaking of NVIDIA, Micron, NVIDIA and AMD are the biggest drags on the NASDAQ 100 today. But it's not all bad out there for tech. We'll bring you a bright spot next. Stay tuned.

38:15Welcome back. Take a quick look at shares of Bank of America, which are down 5 % suddenly this afternoon after CEO Brian Moynihan said the company expects its investment banking fees to drop at least 10 % in Q3. Now, he was speaking at the Barclays Global Financial Services Conference. He also said sales and trading revenue will be relatively flat to last year. But again, those updates about investment banking in particular, sales and trading weighing on the shares today. We'll talk to Mike Mayo, by the way, on Power Lunch next hour. Look forward to his reaction. Elsewhere, the cybersecurity stocks are popping, catching a bid after this weekend's AI safety warnings.

38:48The ETF ticker Cyber CIBR is having its best day in about a month. SentinelOne, Zscaler, and CrowdStrike are among the top performers. ETF is up 6 % today. SentinelOne is up 15 % for its best day since April of last year. CEO Tomer Weingarten posting on LinkedIn yesterday, quote, the AI opportunity is now inseparable from the security challenge. We'll hear from another cyber CEO on Mad Money tonight, CrowdStrike CEO George Kurtz. He always has a way to kind of shed light on these situations, and he'll be on with Jim at 6 p.m. Eastern. Coming up, Anthropic choosing the Nasdaq for its IPO. But will Dario Amadei's proposal to pace AI also mean pacing or slow walking that debut?

39:28We'll talk to billionaire tech entrepreneur and Tempest AI CEO Eric Lefkowski about that next.

39:37Dario Amadei is not the only reason that Anthropic is in the headlines. We have new details about their expected IPO. CNBC has confirmed they will list on the Nasdaq. Reuters reports they seek to raise up to$100 billion at a$2 trillion valuation, with NVIDIA considering investing up to$10 billion in the IPO. But will Amadei's suggestion to slow down AI also slow down that debut? Joining me here on set to discuss is someone who knows the IPO game well, both as an investor and as a founder. Eric Lefkowski is managing partner at VC firm LightBank Capital, the co-founder and chairman of Groupon. and the founder and CEO of health tech company Tempest AI.

40:14It's great to have you here. Welcome. Health tech is one of the areas where people maybe see the most promise with AI. Is it still too early days or is there real stuff happening? I mean, I think it's the area that most likely will eventually be the biggest beneficiary, because if you think about it, all these amazing models and their capability are just perfect for health care, where you have unstructured, disparate, messy data, and these models are really good at making sense of all that data. and the application is just like, what's better than keeping people living longer and healthier lives?

40:43What are the main kinds of things that you were looking at or excited about investing in right up until or around the arrival of AI? I mean, what would be some examples? Because you were involved, were you not in the kind of, tell me again, what was going on with Moderna there and your involvement? Yeah, we've been focused on AI for a while. I mean, Tempest has two main businesses, you know, diagnostics and data. And basically we're in the business of using, you know, big data and AI to essentially make diagnostics smarter and to help drug companies make better drugs. So we've been in this for a long time and have amassed a really large data set.

41:14When these big frontier models showed up, they just were like, we just were fortuitous that they were perfectly designed to help us in our two main businesses. And so we've been pretty active for the past, you know, three or four years at making these models work at scale. You know, part of that is making drugs more personalized. The whole dream of AI and healthcare, at least as it relates to our world, which is cancer, is how do you go from targeted therapies to like personalized therapies? And there's probably no better example. At least one of the best examples is this recent Merck and Moderna drug where you're essentially building personalized vaccines based on the molecular profile of the patient.

41:50Because we saw the headlines, you know, vaccine, possible vaccine against skin cancer. But I didn't catch that element of it, that it would be personalized. Yeah. And what's happening is you're sequencing a patient and you're actually building the drug based upon that person's unique profile. And if you think about it, this is a mechanism to make kind of immunotherapies more personalized in cases where they don't work. These immunotherapies are amazing, but they don't work in all patients. Where they work well, they work incredibly well. Where they don't work well, they don't work well. And so this is a vehicle to basically say, based on this person's own unique molecular characteristics and immune profile, how do we find some way to kind of light the tumor up so that these drugs know, okay, this is what I'm going to target.

42:31How many months or years away are we, you think, from more announcements like the one that we just had? You know, I think obviously, I mean, Merck and Moderna know more about their portfolio than others and when these trials are likely to read out. But I think the success of this particular trial was pretty dramatic. And I think, you know, they've intimated there are more coming and they suspect more will come. And so I think the whole space is kind of excited in that this has been the long time coming. We sequenced the human genome like, I don't know, 22, 23 years ago. And we had all thought back then, OK, this is going to unlock the mysteries of cancer.

43:06And it really hasn't in the way people thought. But it just takes a long time. Health care is slow. Drugs are slow. And so now it's starting to come. Right. To you guys, it's sort of painfully slow. And again, I know you have personal experience with this and your wife. That's when you started the company. So for those who are in these critical moments, you don't want any delay. Now comes headlines from the weekend, right? We have to slow down. And I, from your, from where you - It's a good segue, by the way. Do you like, so, because I listened to the story, I'm like, this is great. You know, this is good real life, but people dealing with this firsthand right now.

43:36Will it truly, I mean, do you just kind of dismiss this as all talk? Or, you know, we had a guest who was just saying, yeah, maybe we should slow down because we slow down, that slows down. I don't know. I mean, I'd really genuinely be curious what impact something like slowing down, whatever that means, could have on these efforts. Yeah, I mean, we're one of the largest, you know, users of these models that's out there, certainly in healthcare. And so I think for us, if you look at the models we were using, let's say six months or a year ago, they were extraordinarily powerful. So I think as it relates to enterprise, it relates to healthcare, I don't know if slowing down is going to change our ecosystem.

44:13In other words, models, even from six months ago, are still so powerful that you could do a lot with it. That's almost even more scary to think about how powerful they are. There's many questions, right? One is, how do these companies compete against each other? And in that case, there may be a bit of a race to build a better model, a better model, a better model. But as it relates to the application of those models, I think the models are so powerful that we have a lot of runway. So to the extent you said to me, hey, a law was passed and everything has to slow down 20 % or 30 % or some number, I wouldn't think, oh, this is the end of our ability to really use these models to produce unbelievable benefit in terms of human suffering.

44:56I would believe we're in a great spot, given what we already have and what will come in the near future, even with a slowdown, to keep people living longer and healthier lives. That's fascinating. And I guess just to broaden the lens out a little bit, are there any other areas, whether it's AI specifically or, I don't know, a fun new consumer app, that have you excited today about the applications of technology and all of this innovation or just things that you're using kind of in your life day to day? I think the thing that has to be most exciting is that, you know, companies like ours, in particular Tempest, has amassed one of the largest data sets in the world of de-identified data that can be used to advance drug discovery and to advance diagnostics.

45:33And literally these models were a gift, a gift from the gods, right? That their ability to take all this data and make it actionable and make sense of it is just unbelievably powerful. And so I think it's going to take us another five or ten years to really reap the benefits of what we already have amassed in terms of data and capability. And you know what's fascinating to me? Sometimes it still gives me the wrong calorie counts, right? So I'm like, how can it be that these models are so good at what you're describing and coding? And they're so good at hacking the Internet. Like, there's areas where they're unbelievably good.

46:08And then there's other everyday, much lower stakes areas where I'm just like, I can't get you to do this thing right. How can both things be true at the same time? Well, I think it's the difference between what's the use case of the model, right? So if you're basically saying there's this AI and I'm going to use it for anything in the world, it's only going to be so accurate all the time. So you're just going to trip it up constantly in some small way or some large way. We're in the business of taking these models and taking our data and then basically retraining the models based upon our data.

46:38And then we built thousands of proprietary agents that sit around these models to essentially get at the exact problem you mentioned, which is, are they producing the wrong answer? Did they have the wrong context? So models checking on the models. And really, it's a matter of both in pre-training and post-training of us saying, look, we need an industrial safe environment for these models. We need to basically harden them so that they can be relied on by physicians to make important decisions. Life and death decisions. And so you just can't use the models like out of the gate and go. You have to actually invest.

47:08That makes a lot of sense. They're playing your music. They're playing them. Rockstar. Eric, thanks very much. Really appreciate it today. Eric Lefkowski, that's it for us on The Exchange. I'll join Dom Chu for Power Lunch after this quick break.

47:23You've been listening to The Exchange. Make sure you're subscribed to get each episode every day, same time, same place. Something amazing is happening in animation. Studios are processing data faster. That's because CDW showed them new ways to maximize their infrastructure, then built a flexible Dell Technologies data solution. More automation led to reduced maintenance times and greater efficiency, so that creativity stays the star of the show. Dell Technologies and CDW. Make amazing happen. Learn more at cdw.com slash delldatacenter.