Tempus AI, Inc. (TEM) Earnings Call Transcript & Summary

Okay. I think we can get started. Kallum Titchmarsh from the Life Sciences team here at Morgan Stanley. Really pleased today to be joined by the team from Tempus. We have Eric Lefkofsky, Founder and CEO; and Jim Rogers, the CFO. Before we get started, I'd -- I'm required to read you some disclosure. So for important disclosures, please see the Morgan Stanley research disclosure website at www.morganstanley.com/researchdisclosures. So again, thanks, guys, for being here. Maybe we can just hit on Q2 first. Talk us through what drove the strength there? But then also zooming out just over a year being a public company, what are you most proud of? And maybe any of the key challenges you would probably flag thus far?

Do you want to start with Q2, and I'll take the most proud?

Yes. So Q2 was a great quarter for us. Genomics business kind of reaccelerated our growth rate. We grew 20% year-over-year in Q1. That accelerated to 26% unit growth in Q2. And a lot of that was driven by sales efficiencies and really just the adoption of the product. So Genomics revenue was kind of north of 30% given some reimbursement tailwinds. And then on the data side, really just continuing to execute on a lot of the agreements that we've signed over the last couple of years. We announced a big partnership with AstraZeneca and Pathos. That project got underway in Q2 and so started to contribute revenue. So overall, a great quarter, continued our improvement from an adjusted EBITDA standpoint, about $10 million of improvement quarter-over-quarter, so on track to flip positive in 2025.

And then I think in terms of what it's like being public and we're most proud of, I mean, the business is performing just incredibly well on really all cylinders. So it's nice that it's less about being public or not public. It's more getting to the scale at 10 years where we're getting close to $1.3 billion of revenue, and you still have these 2 main businesses growing at roughly 30%, which is only compounded by our acquisition of Ambry, which accelerates our growth rate. It's just nice that the business is this solid, this strong really across all the major growth levers. And so if you would have said to us 10 years ago, where do we want to be, we would have said to you right here.

Fantastic. So let's kick off with the Genomics. You, obviously, have one of the broadest oncology portfolios out there. You give physicians that offer to have a kind of one-stop shop. So how do you weigh up the pros and cons as you now think about expanding the portfolio from here? And how does kind of a single vendor value prop resonate versus a more broad spread provider?

Yes. I mean, there's -- give or take, just under 15,000 oncologist in the United States. And so it's -- there's kind of no single group of physicians that control this much spend relative to the U.S. health care system. So these folks are incredibly busy. They have a lot going on. And so I suspect over time, this group will want to work with fewer vendors who can solve their problems in a more holistic manner. And so we have long felt like in order to win this space, you have to be in hereditary risk, you have to be in treatment selection, both in terms of solid tumor profiling and liquid biopsy and you have to be in MRD and monitoring. And I suspect if we go forward 5 or 10 years, the largest players in the space for MRD will be the largest players in the space for treatment selection and vice versa. So we've long thought that that's how this would evolve, no different than Amazon didn't win books in e-commerce. It won e-commerce. So I think the biggest issue for us is not how do we expand our portfolio within oncology. We already have a very broad portfolio. It's really how does the portfolio expand outside of oncology. For example, Ambry is going to be doubling down. Our acquisition, we get to double down in rare, especially pediatrics. So you have a kind of a brand-new large category that we'll be doubling down in. And then I suspect more categories like rare will also start to get positive reimbursement. And so that will be additional categories we get to go into.

I'm curious what the feedback has been from physicians on the MRD side covering both tumor-naive and tumor-informed? Again, it's early days, but how has that been thus far?

I think the -- it depends on the category, right? So in certain subtypes, tumor-informed, like, for example, in colorectal cancer, where you have lots of tissue, it's a perfectly good solution. In lung, where there's less tissue, people might want a liquid offering. So we have long felt that, again, you need to have both tumor informed on the solid side and tumor naive on the liquid side. You need to have solutions for both and that the market by subtype might evolve in certain ways where you need both. It is possible that over time, the liquids -- the naive side of this becomes so sensitive, so specific where the limits of detection are so low that you could see significant displacement. But I don't see that happening for quite some time. So I think we think our approach of being in market with both, which I think Natera now has a similar approach is probably right.

And just given Caris' recent IPO, one question we've been getting is whether doing a whole exome really adds that much diagnostic yield versus a broad but targeted DNA panel. What are your thoughts on that?

So I think the -- at the present moment, therapeutically, on the DNA side, there's only really several hundred markers, biomarkers that are clinically relevant. So whether you do whole exome or whole genome, most of that information is being compiled for research use only. It doesn't really have any clinical significance. There's no drugs tied to it. There's no therapies tied to it. So we have long felt like doing whole transcriptome was probably generating more insights because you get to see things downstream from a genetic mutation. That said, I do think the market, and we've discussed this at JPMorgan, we launched our whole genome-based heme assay, which will come out later this year. And we do expect over time to migrate our solid tumor portfolio to whole genome, less because the data is so relevant today, more because it's an easier workflow. You now can generate rich data at lower cost. And I suspect it's just going to be nice to move to one chassis where you don't have to run these [indiscernible].

And the number of paid oncology tests has seen a nice steady improvement. Just what's the latest progress on securing reimbursement among the commercial payers?

So yes, we're on the same journey as everybody else. Obviously, our lab is a little bit younger than some of our competitors. So we see positive trends, although the commercial payer landscape for us is very fragmented. So it doesn't necessarily always show up in the numbers in any given quarter. But I think generally, we've seen more coverage than what we saw 5 years ago, and we'll continue to kind of chip away at that on the commercial payer side.

Are there any assays that are lagging others on the commercial coverage side?

I mean, obviously, MRD for us is the big one because we don't have any reimbursement today. So as we get that kind of in the tail end of this year, first through CMS and then we'll kind of approach the commercial payers. But I would imagine that will be the same case for [ MRD ].

Got it. And I want to spend some time on Ambry as well. I think you recently called out long-term growth rates there could be maybe higher than originally expected. The market on the hereditary cancer side is seen as maybe more established than some other areas. So I'm curious why could you see this reinvigoration of growth?

Yes. I think the -- there were some -- there was this belief, I think, based upon the performance of a bunch of companies in the space that the growth rates in hereditary profiling had kind of capped out and it was becoming a commoditized space. I don't really know how that narrative evolved, but it -- we thought it was an inaccurate narrative, and I think the unit volume is kind of playing that out. I mean there are far more people that are at risk of getting disease than have disease. And we benefit -- Tempus and others like us benefit from the collective R&D efforts of the entire ecosystem. Every academic medical center researcher, every biopharma company, everyone doing work to find some molecular biomarker that's connected to disease turns into something you have to watch, whether that's for various types of cancer where markers beyond BRCA will become equally relevant or whether it's outside of cancer, early onset dementia, type 2 diabetes, pick up -- pick up -- pick a risk. So I would not be surprised if companies like Ambry end up sequencing 10x or 20x the amount of patients, companies like Tempus sequence in oncology alone just because I think there are far fewer people that have disease that are at risk. And so it does feel like a space where ASPs have normalized, where the unit growth rate will be much higher than people anticipated. And so we're optimistic. That said, we've only owned it for 2 quarters. And so we told the world like, let's wait and see the next few quarters go.

Yes. How -- I think, north of 30% growth for Ambry in Q2, how sustainable is that for the rest of the year?

Yes. So we talked at the Q2 earnings call that about half of that growth came from share gains from competitors, the other half from kind of organic growth within their accounts. The share gains obviously can't continue forever. And so we would anticipate those taking down over time, but they're performing ahead of where we anticipated when we bought them.

And what does the mix look like between hereditary rare disorders and pediatric for Ambry specifically?

So we don't disclose. Obviously, the majority of the business is still hereditary oncology and then a smaller component is rare.

What are some of the investments you're making on those other 2 areas for Ambry specifically?

Most of the investments in rare have been made. So we have quite a good portfolio there, both on the exome side and the whole genome side and a whole platform around tracking these patients over time. So I think we're well positioned. We're one of the largest players in the market today. Obviously, [ GeneDx ] is a large part of the market, and it's really [indiscernible]. And I suspect, given the fact that we just now are ramping that up, we'll grow quite quickly. Historically, for Ambry reimbursement wasn't there, and so they put more of their energy on the cancer side. But like each one of these areas where all of a sudden, we demonstrated enough clinical utility for reimbursement to normalize, you get the opportunity to invest.

Okay. I want to shift on to data now. What are some of the challenges you had to overcome to build the infrastructure you have today? And why is it that someone else couldn't come in and emulate it?

Yes. I mean -- so we're -- as a tech company, we just have a different orientation than most of the big labs we compete with. We have something like 700 software engineers and folks in that world. and we make enormous investments in cloud and compute on top of our investments in engineering talent. And so -- and we've been making these investments for a very long time. And so we've built up a very large and very mature technology stack that allows you to make sense of all this disparate multimodal health care data so that clients, especially R&D clients can get real benefit from it. And when we first started licensing the identified data to biopharma clients years ago, it didn't go well. They didn't like the data. They couldn't generate insights. And so we had to really invest heavily in building products that would make the data useful. I suspect if you fast forward today and look at our data business relative to others, one of the reasons our data business is so much larger than anybody else is because we've made those investments. So it's not just that we have more data than other people or that it's real time in nature based on all these thousands of connections. It's also the amount we've invested in software products and tools that make that data useful. So it's -- we said this during the IPO, I kept using this example. It's like mowing 3,000 lawns. It's not that somebody can't do it. It's that it takes enormous effort and you can't cheat it. There's no way to snap your finger and say, oh, my 3,000 lawns are mowed. You have to mow them.

And what's in it for the health care institutions to provide you with the health care data essentially for no payment? I guess, what's in it for them? Could they and, I guess, have they ever changed their minds?

So I don't believe we've ever had anybody that turned off the data. I mean for them, they see value by getting kind of a more intelligent diagnostic results. So by sharing clinical data with us, we're able to contextualize the results for the individual patient for which the test is ordered, recommend trials that they are actually eligible for based on the inclusion/exclusion criteria, removing therapies that they've already received in prior line and failed and giving access to the broader database for physicians to kind of sort through and see how other patients were treated. So it's really by sharing the data, they're getting more insights and actionable insights back from the diagnostic test.

Yes. On insights, the vetting and trial period with potential customers, how does that process work? What are the typical studies you work through with customers?

On the...

On insights, specifically. So what's that kind of trial vetting period like? Just maybe talk us through how that works.

I mean, typically, what happens -- so we have hundreds of biotech clients and then we work with most of the big pharma and oncology. And often the way it works is somebody will license a very small amount of data. Several hundred thousand dollars, whatever small amount of data, maybe $1 million. And then they say, in one subtype to answer one set of questions. And then maybe a year or 2 later, they realize it's adding real value and they'll expand that into multiple subtypes. And at some point, once they realize that data is helpful across their entire oncology portfolio, you begin having these conversations about, okay, if I'm going to license lots of this data, what's the best price I can get. And the way our pricing for data works is also when you think about the fact that we've got kind of a total contract value north of $1 billion, meaning people have signed up for data to be delivered in the future at that significant rate. What's kind of wild about that is you can license our data, you can license one file. Like it's not like you have to sign a massive deal to get our data. The only difference between one file and 10,000 files or 20,000 files is price. So it's a bit like the way AWS or GCP or Azure prices their cloud products where you can use it in very small denominations, but you're going to pay kind of retail. And if you want to make a longer-term commitment, multiyear commitment with a certain dollar amount, you get a discount. And so I think the fact that so many people are signing long-term agreements means that the data is obviously adding a ton of value.

And obviously, a lot of agreements there in play. Can you give us some examples of pharma use causes -- cases of your data, just how they've used it to better outcomes?

Yes. I mean, well, I mean, AZ has kind of published on this, so you could read about it. They had published about a year or 2 ago that they saw a roughly 5% PTRS lift, probability of technical and regulatory success lift, across big parts of their oncology portfolio. If you accelerate -- it's just simple. If I increase the probability of success by 5% or if I increase the time for a drug to get to market by 12 months, either one of those 2 produces something like $90 million of NPV per asset. So if you've got 10, 20, 30, 40 drugs in your portfolio and you can use our data to build a synthetic cohort against a single-arm Phase II or figure out that a shelf asset should be unshelf or attach a biomarker to a drug that didn't have a biomarker or remove an exclusion criteria that you really don't need because in the real world is no longer there. Any one of those, these things are like massive. So I would find it kind of hard to believe that 10 years from now, every major oncology company isn't licensing significant amounts of data from us or someone like us. I just would find that hard to believe.

That's helpful. How does the data piece aid the genomics part of the business model? Are there particular aspects of the genomics offering that you could point to that makes your clinical tests more differentiated versus those by competitors, thanks to the data?

Yes. So one of the advantages, and this kind of leads into the AstraZeneca Pathos agreement is by structuring all this data for purposes of the data licensing business, it also gives us a really robust data set to kind of train models on, identify insights and kind of embed those back in the genomics or in our diagnostic offering. So the AZ Pathos agreement where we're building this foundation model on the entire data sets or training on the entire data set, we're confident it is going to yield those types of results that are going to differentiate our genomics business. And so we often have talked about this flywheel where genomics is kind of the data provider for the data business. Then we mine it for insights and we embed those back kind of giving us an advantage. So these businesses are definitely interconnected.

And we said this during the -- so no one's ever run this kind of foundation model in oncology, talking north of 300 petabytes of data being moved into a cluster of essentially 1,008 H200 GPUs that are going to be running for like 3 years on that data set with kind of all the tools we built. It's something like 1,200 proprietary agents that make sense of multimodal healthcare data. This is like a non-small effort. So we don't really know what's going to come out of that. We're finishing pretraining now. We'll run compute into Q4. But I suspect what will come out of it, if you look at our growth rate, which even at our scale, I think we ran 212,000 tests last quarter, like growing units at 26% year-over-year at that growth rate is pretty extreme. And the main driver of that is that our tests are just more personalized, more contextualized than others. And so physicians like them. And what they're really trying to figure out is, okay, in light of this molecular insight, whatever it is, this RNA expression level, this DNA education, what do I do? What drug do I give? How do I change therapy? And I think what I'm hoping comes from the foundation model is a plethora of insights that we couldn't see until we ran compute at this scale. Associations, for example, where you can look at non-small cell lung cancer patients where frontline therapy might be an EGFR inhibitor if you're EGFR mutated, but we can see, oh, wait a minute, here's 20% of the population that never responds. So you as a physician you could do something different because this patient moved back every 3 months. So I think it could be transformative in terms of those level of insights, but we'll have to see [indiscernible].

How should we think about total contract value from here? Obviously, now peaking over $1 billion. How much fluctuation can we expect there in the coming years?

Yes, we get this question, I think, a lot. I mean, if you go back several years, that number was $300 million. So you look at kind of on an annual basis, it has grown kind of steadily over the last 4, 5 years. Within any given quarter, if you sign a $200 million deal, obviously, there's big fluctuations. There may be a quarter where less is signed and so it comes down a little bit. But largely, when you look over multi-years, it should be kind of growing at similar rates to kind of the growth rate of revenue over that kind of same time period.

Just how hard is it for a company? So you have these long-term contracts, they use your data for a while. Surely, they don't want to let that go, right? They want to keep using it and expanding beyond that.

We only had 2 -- many of these contracts are kind of 4, 5, 6 years in duration, the bigger ones. So they haven't come up a lot. We had a few. We had -- one of them which we announced, which is Merck KGaA came up for renewal after a very large 3-year contract, and they renewed for another 3. So that was one of the few. And then AZ agreeing to do this foundation model, they had a few years left on their old deal. This new foundation model is a really big investment by them into leveraging our data. So that's another great proof point that the data is adding a ton of value because otherwise, [indiscernible]. So -- and I suspect as other contracts come up, I have no reason to believe they all won't.

Yes. Who would own the foundation model once it's completed? I guess, how do you anticipate using this in other pharma partnerships?

In this particular case, I think each one of these things might be slightly different if we do other deals. But in this deal, each party gets a copy of the foundation model. So we get to own it for diagnostic and data purposes and then Pathos and AZ each get a copy of the model that they get to use for their own internal drug discovery efforts. So Pathos is a small biotech, AZ global pharma, but they can use it for their internal R&D work.

On the Q2 call, you noted a flow in the U.S. health care system when there's no kind of mechanism for reimbursement for AI and algos. Is there anything you can do to further accelerate that evolution or drive awareness of the potential benefits?

I don't know. I mean we're in the middle of a lot of those conversations now. The system has to fundamentally change. You can't not pay for AI in health care when you spend as much as we spend as a system and produce the results we produce, it's just not sustainable. So we're going to -- the only solution I can think of this technology and AI that in theory could produce better outcomes. And so we have to find a way to pay for that. That said, I don't have any -- there's no like this is coming next month, it's going to be game changing. I think the system, at least with this administration, in particular, in HHS and CMS, I think you have people that recognize they're going to do something different.

Have you been engaging with the FDA on this?

We've engaged with the FDA for quite some time. The FDA is not -- what's interesting is the FDA is not the problem, which is most people think they are. We have, I don't know, a dozen more FDA-approved AI-based algorithms. FDA approved, meaning the FDA has approved an algorithm for Tempus to look at a 12-lead ECG. And from a normal 12-lead ECG run by [indiscernible] or others, we can basically say that result is wrong and this person has undiagnosed AFib or this person has undiagnosed low ejection fraction and more are coming. We have the same thing in ditch path. We have the same thing in radiology. We can detect pulmonary nodules. So the FDA is not the problem. They're -- if you're willing to go through their process, they're willing to give you approval if the bar is met. The problem is once you get approval, you get no money. That's the problem. The problem is we don't bundle FDA approval of these tests with reimbursement.

Understood.

That's what has to change.

Yes. So you've got a number of analytical tools to support researchers use those insights from your data, Loop and Lens are a couple of them. Maybe just tell us a bit more about these solutions and any recent traction you've seen with [ Lens ].

Yes. So I think, as I mentioned, one of the tools we built that is differentiated and drives our data business is this application called Lens. And Lens is basically an analytic tool that we built that allows you to build cohorts of interest, interrogate those cohorts, run your own models on our technology stack. And so it allows you to kind of move around a lot of data and interrogate the data at high fidelity and low cost. And that product is starting to get some real traction. On the modeling side, we also -- when we started sequencing patients, we started thinking a lot about the kind of data you would -- the kind of multimodal data you would need to generate these insights. And we used to talk about this notion of phenotypic morphological molecular data or text images and molecules. But we also were cognizant that no matter how much data we had, there would be a certain amount of data we didn't have and that we would want to generate on our own. So we built a modeling infrastructure, in our case, based on organoids where we began bringing in cryo -- basically frozen or fresh tissue that we would then cryopreserve and build these organoids and do drug screening on these kind of mini tumors across really every epithelial cell category. And that bank has now gotten quite large. And so in addition to our normal data business, we also have an emerging kind of synthetic data business where we're able to interrogate all these different drug combinations across these many tumors. And that has had some really nice wins where clients have come in and said, I want to do a data licensing deal and in part bring in that capability. We call that Loop. So I just highlighted, I think, in one of our calls, these are just 2 of the kinds of products that we built that add some fortitude to our data business.

And then just on the physician apps, Tempus One, Tempus Next and Hub, maybe just give us a brief overview on those and how they differ from maybe what's out there, if anything, is out there?

Yes. I mean, Hub is kind of as I mentioned before, I mean, physicians can go in there, they can kind of view the broader database. They can filter down for similarly situated patients, either from a genomic standpoint or phenotypic, see how those patients were treated and how they responded. Tempus One is embedded both in Lens and in Hub. So physicians can actually talk to the diagnostic test. They can ask what is linking on to the guidelines? What are the side effects of this therapy? All those things are kind of built right in the Hub. Again, kind of getting back to the point of, in addition to providing the diagnostic, we want to provide tools and technology that make physicians' life easier, and that's why we embed all these things into kind of their workflows.

Paige AI recently announced, maybe, again, an overview on that. What do you think it brings to the business?

So the Paige acquisition was for us really interesting on a few levels. One is we think ditch path is an exciting space. And over time, will be one of the main cornerstones of bringing AI to diagnostics will be through digital pathology. So they had a whole bunch of capabilities in that space. They had built a viewer. They had a series of FDA-approved or pending algorithms to make predictions clinically. They had built a foundation model called Virchow that was -- gotten some real scale. They had an incredible team that was good at manipulating ditch path data and building these models. And then they also had a unique relationship with Memorial Sloan Kettering that gave them access to all of MSK's digital pathology data connected to a certain amount of clinical insight. So we wanted all that. And at various moments in time, we had talked to the company and priced didn't align. But as we got a few months ago, price didn't align, and we were able to bring them on board.

And post July's convert, you're pretty well capitalized now organically investing or also through M&A. So how are you thinking about capital allocation from this point?

Yes. I mean as we've said, we historically have bought smaller things that kind of solve some kind of problem for us. So for example, we want to increase our digital pathology data set. We want to double down our capabilities there. We can make a relatively small acquisition that allows us to kind of move that chess piece forward. We don't make big acquisitions unless we can find a company that we think is, relatively speaking, as good as us. So we operate at significant scale. We have a highly diversified business. All main parts of the business are growing rapidly, kind of 30% growth, and we trade at some multiple. So if we were going to buy something big, we would want it to fit into that paradigm. And it's hard to find things that fit into that paradigm. So we tend to be way more cautious there.

And a small raise after Q2, I think adjusted EBITDA was kept the same. Just maybe walk us through the philosophy underpinning the guidance? And any color on what you've seen in the last couple of months, I think, through Q3?

Yes. So I mean, I think from a guidance standpoint, it was very important for us to be self-sustaining as we just turned 10 about a month ago or a couple of weeks ago. And so for us, it was very important to be adjusted EBITDA positive for 2025. And so we're very focused on that. I think we raised revenue a little bit. We kept the adjusted EBITDA the same. We've always said that the opportunity in front of us is still very, very large. And so we're not in a position where we want to just be harvesting profits. And so if we're running ahead of track in any given year, we may reinvest some of it back into the business. And so that's why the adjusted EBITDA remains the same.

Anything more recently, any color? Again, it's 2 months [indiscernible]

Not the one forced us to file an 8-K -- September 9. The business is doing in the aggregate. So we're fortunate, as Jim mentioned, that we're able to keep reinvesting in that long-term growth trajectory. And bringing AI to health care, despite the fact that we operate at some scale, we're still in the very earliest part of the cycle. So we don't want to win 2026 and lose 2036. So we want to make sure that we're kind of appropriately aligning. That said, we thought it was important to be EBITDA and free cash flow positive. We're knocking on that door, and we'll get there and it's a nice spot to be.

Yes. Maybe on that, Jim, can you maybe talk about just bridge us from where we are today to some like longer-term targets on the profitability side?

Yes. So we haven't commented on kind of long-term profitability for the exact reasons of my previous response. So each year, we're going to assess of, okay, the business -- each of the businesses are growing x percent, that's generating its increase in gross profit dollars, what's the appropriate amount to drop down to the bottom line versus reinvest in the business. And each year, there's a laundry list of things that we go through to say, is this something that we're doubling down on? Or is this something that we may be putting aside. So as we approach kind of 2026, we'll provide more and more kind of color on our thinking there. But as we sit here today, it's a year-by-year effort.

And I think we've got a couple of minutes left. So as we think about incorporating AI into the health care industry more broadly, I think AI to some investors is met with some skepticism, maybe to some degree because of the weight of accountability it has for patients at times. Do you think that AI will serve a big role in health care, maybe relative to other industries? Just how are you kind of thinking about the impact that it could have?

It's interesting. So I mean, obviously, we're the leaders in bringing AI to diagnostics. And yet we actually talk very little about the impacts of that financially because I think they're very small right now. So I think we're fortunate that our main diagnostic business and our main data business are big and growing, and those are tangible. You can see it. There's no doubt that AI will come to health care. I think it comes through diagnostics first. But either way, it will come to health care, and it will have an enormous impact. But it is very hard to see that today. So anything we were to say on that topic would be highly speculative. And it could just as easily be completely wrong. So I tend to -- the way I think about it is, given that you know AI is coming to health care at some point and given how massive the health care space is, this is a space that probably has $1 trillion or $2 trillion worth of movable free cash flow based just on inefficiency. Like it's massive. You just want to be in the game. You want to have a broad portfolio of AI products that you can bring to market when there's actually money to be made by bringing them to market. And I think that's how we think about AI, which is we're making all those investments. And I think one day, they'll yield really positive results. But for right now, that's kind of [indiscernible] the future.

I think we have time for one more. So what's something you wish investors ask you more often?

I don't think it's necessarily that. We have this conversation. If you look at the people who have kind of made the most money investing in Tempus, they tend to be -- they've been more thesis oriented. They believe that AI is coming to health care. They believe that Tempus has as good an approach as anyone, and they're long that thesis. And so they have invested in us and there's other things that look like us, and they're just true believers. And I think they will likely probably do well. No different than if you were investing in e-commerce or search 20 years ago, you would have done well. You may have invested in 5 other things, but Google would have been one of them, and you've done well.

Got it. Okay. Eric, Jim, thank you so much.

Thank you.