All right, great. Welcome, everybody. I'm Casey Woodring from the Life Science Tools and Diagnostics team here at J.P. Morgan. Welcome to our conference. Pleased to be joined by Tempus AI CEO Eric Lefkofsky. We'll go through the corporate presentation, then leave time for the Q&A afterwards. Eric, all yours.
Thank you. Welcome. I don't have seats up here. I'm going to have people in the back, but my homies, feel free to walk up. So I'll try to give you a little context of what we've been building at Tempus and a little bit about where we're at now. So 10 years ago, we started Tempus to solve a single problem. Could AI-enabled diagnostics unlock precision medicine? Essentially, could we use diagnostics as a vehicle to make precision medicine as opposed to targeted medicine a reality? In order to do so, you really need two things. You need access to vast amounts of proprietary data to train models and uncover insights, and you need a distribution system to deliver those insights to the hands of physicians and patients. Most people have failed historically at having one or both of those. We have spent the last 10 years building both.
We've built up a data set that is now quite substantial. And equally importantly, we built up distribution capabilities to take those insights and distribute them to tens of thousands of physicians across the country, reaching millions of patients to advance therapy selection, what therapeutic path should my patient be on, to advance clinical trial matching, is there a trial that they're eligible for, and ultimately to advance research and drug discovery. The idea we had 10 years ago was that in order to bring AI to health care, you had to start somewhere. And we thought you should start with diagnostics. It's the only external data modality that physicians interact with all day long.
So if you want to bring AI to the health care system, you either ask doctors to go somewhere, you take systems like Epic and make them AI-enabled, or you have to find some alternative pathway. Given that physicians order laboratory test results and routine diagnostics when they make almost every major decision, we thought if you could wrap AI around the diagnostic itself, you could kind of infuse the benefits of technology into health care. And so the platform we built is now broadly connected to about more than 5,000 providers across the United States. To put that in perspective, it's probably two-thirds or so of the United States who are connected to Tempus in some way, shape, or form. They're either ordering our tests or using our technology to match their patients to clinical trials or close care gaps, things of that nature.
The scale of the platform is unique in that we touch more than two-thirds of all academic medical centers, more than 55% of all oncologists, 7,000 of which are regularly ordering our tests and interacting with our products. And the data set we've amassed as a result of that is more than 400 petabytes of rich multimodal health care data. When I say multimodal, what I'm referring to is phenotypic, morphologic, and molecular data. So think of it as text, image, and molecules at scale. So who's this patient? What drugs are they taking? How are they responding? And what's their molecular composition? Understanding all of those and critically understanding all of those is necessary to advance AI. Everything else is just a point solution. It will solve part of the problem, but not solve all the problem. We've been focused on amassing this data set.
We had a theory that it was crazy that people would try to do research 10 years ago when I started Tempus, building these very small and expensive data sets, writing grants, raising money, trying to procure samples in small denominations, sending them off to places like the Broad to be sequenced, and as a result, you amass 50 or 100 patients' worth of data over a year or two, and it was crazy to me that at the time, people were sequencing patients at scale, but that data was lost. It either wasn't being handed back to hospitals by the companies that were sequencing, or it wasn't being matched with other critical data, like what drug is that patient on and how are they responding, what's their longitudinal journey. It wasn't being matched with other key data sets, like pathology slides or radiology scans.
And as a result, it couldn't be used to advance research. We started focusing on building these large data sets that could be de-identified and used broadly. And as we sit here today, some 10 years later, we've amassed a data set that's truly unique. It spans over 45 million patients. Over eight million have digitized imaging records that are annotated. We've sequenced over four million samples. And at the very bottom of this funnel, there's more than 350,000 records that have rich molecular data, rich genomic data, rich transcriptomic data, rich imaging data, rich clinical data. And that data set really powers the majority of our data business today, which operates at scale. The idea of marrying these two concepts of taking clinical diagnostics and allowing them to power research data sets is still in place today. The network effects that result from these two is pretty profound.
The more patients we sequence, the more data we collect, which allows us to provide additional insights for their enhancing our businesses and allowing us to collect more data. That flywheel that began to kind of turn six or seven years ago is now in full flight, and as you can see, we're not just collecting some data. We're really collecting the totality of data that's necessary to figure out if a patient's on the right drug or the right trial or how I would advance drug discovery and development. If you think about the problem that we have been solving for the past 10 years, it really exists on two levels. The first is you have to acquire enormous amounts of data. We have this data set that now is unique, and yet it still is not at appropriate scale.
So we will continue to amass a lot of data over the next several years as we really try to get a significant percentage of the U.S. market's data in our hands. The second is you have to build a platform that allows you to make sense of the data. Otherwise, you just have exabytes of data, and to make sense of the data, you need a platform that is sustainable. You have to be able to invest in harmonizing and structuring and making sense of this data and generating insights, so we effectively had to clear two hurdles. We had to find a way to amass lots of data and find a way to do so without burning tons of cash.
What we're most proud of over the last 10 years is not only that we built this data set and touched so many patients, but that we now are completely self-sustaining. We generate positive adjusted EBITDA. We have great momentum in terms of our financial performance. The machine is both humming and sustainable. To think about these two businesses and their integration, you have to understand the two businesses in isolation. We'll start with diagnostics and then get to what we call data and applications. Our diagnostic business consists of us running tests broadly in oncology and other disease areas and billing insurance. We are effectively a provider similar to companies like LabCorp, Quest or similar to the other sequencing companies that we work with or that we compete against, namely Guardant and Caris and Natera and folks like that. We run these tests.
We bill insurance. We get paid. But in our case, we're connected to those hospital systems bringing in data. And so our tests are getting smarter. The system's getting smarter. So we're contextualizing these tests, and we're learning, and we're producing as an offshoot of those diagnostic tests vast amounts of data. That data gets de-identified, and we license it pretty broadly. So let's start with diagnostics. Our diagnostic business today is made up of really two main components, a genomics business and a genetics business. We essentially span the entire cycle from understanding hereditary risk, who's going to get cancer or who's going to get a particular disease, to how do I treat that person once they've been diagnosed with a disease, whether that's through solid tumor profiling or liquid biopsy, and then ultimately how do I monitor that patient post-treatment to see if their disease has come back.
Tempus is unique in that it spans the entire spectrum. We are just as good in almost every part of that continuum. In terms of hereditary risk, we acquired a company called Ambry, which is where we do the majority of our germline testing. We operate at significant scale. I think we're the largest player in that market. In treatment selection, we're unique in that we are just as good at solid tumor profiling as we are at liquid biopsy. And both of those businesses continue to grow, and growth is accelerating. We do a variety of other testing, in particular, a whole litany of targeted tests or algorithmic tests. And then we recently launched both a tumor naive and tumor informed MRD product that's also gaining traction. The backbone behind all of this is something we don't spend a lot of time talking about.
We don't tend to overly emphasize studies we're running or readouts that will occur, but we have significant technical capabilities. We have about 700 software engineers, about 400 PhDs, 50-100 MDs across the entire organization. So it's a fairly large technical team that allows us to do best-in-class research, publish that research, and ultimately protect the intellectual property. We operate at a scale that's pretty extraordinary as it relates to how science is applied across our platform. This has resulted in us being kind of first to market with a whole litany of insights. We wrote seminal papers in terms of the benefits of concurrent testing, seminal papers in terms of how incidental germline findings influence care, and seminal papers on the benefits of transcriptomic profiling, RNA profiling, and how it enhances fusion detection.
These papers were often first in the market and led to a change of behavior across the field, and we're proud of that fact, and we intend to continue that as we move forward. It's one of the reasons that our platform is not just growing, but growing faster than most others. If you look at the output of that growth, we ran over 800,000 clinical tests in 2025 and had a 28% growth rate in Q4. Our oncology business is growing quite rapidly and has actually accelerated in growth over the last three quarters, and our hereditary business maintains really strong growth rates despite some concern that that business wouldn't operate at such high growth. In addition to our unit growth being best in class, we also have rising ASPs.
And our ASPs, from our perspective, we have really good tailwind that we expect to materialize over the next several years. In particular, as we migrate from our LDT version of our solid tumor assay to our FDA-approved version, it comes with enhanced reimbursement through our ADLT offering. We're in the midst of getting approval for our liquid biopsy assay from the FDA. That will come with enhanced reimbursement. Commercial coverage continues to go up every quarter as more and more people cover our tests. And as a result, we expect our ASP to rise from $1,630, which is what it was this quarter, to somewhere in the $2,200 range over the next several years. One of the best parts of Tempus is the strength of our business.
Even with ASP that's dramatically lower than our competitors that will rise over time as a function of reimbursement just getting better, we still operate with margins in the mid-60s%, I think, across the business, in large part because we're highly efficient, so our genomic business still has high margins, and our data business has even better margins, so we have significant leverage in the business, high growth, and leverage even with lower ASPs than others, which means we're well protected and likely to get significant upside. Switching to the data business for a minute, we have a large and growing data business. When we started Tempus, most people thought our data business would never get to $10 million in revenue. I think we did $316 million last year with a business growing at about 30%. and so we do the math. We're operating at significant scale.
Our data business is really both accelerating and pulling ahead of others. We've just never seen greater strength in this business. We've never been better positioned. In large part, it's because people are more and more realizing how instrumental the kind of data sets we can build are to their entire discovery and development platform. Biopharma uses our data across the board. They use it for trial design, for figuring out how to enroll patients, figuring out what companion diagnostics they should consider, which targets they should go after, the kind of evidence they need to generate to get regulatory approval, how to design Phase 2s , and so on and so forth.
I suspect that this will kind of have these moments where, as we begin licensing data to companies, typically they'll start with a small amount of data and then license a bit more data, then maybe more for several years. And at some point, they realize that this data is just too instrumental to their programs to not have within their environment at scale. And so that's typically when we move from a one- or two- or three-year deal to a much larger multi-year agreement. What people don't realize is you can license one file from Tempus. There's no minimum. So the fact that people sign $100 million multi-year deals is only a function of the fact that they want access and they want a discount. Otherwise, they could license one file at a time. Our data business, as I mentioned, operates at scale.
We work with 19 of the 20 largest pharmaceutical companies, over 250 biotechs. We've signed licensing agreements for data that are north of $2 billion over the last several years, which is pretty extraordinary. We had $316 million of data revenue last year, 31% growth. We have partnerships with numerous companies: AZ, GSK, BMS, Pfizer, Novartis, Merck, so on and so forth, and we've delivered over 8 million de-identified patient records to biopharma to advance drug discovery and development. When you think about people running trials with 50 patients, 100 patients, 300 patients, 500 patients, the fact that we've delivered over 8 million de-identified patient records to advance discovery and development is pretty extraordinary. We announced last night that we have over $1.1 billion of total contract value in place, so this essentially is the, I think most people think of this like a bookings metric.
It's the total value of the data licenses we have under contract to deliver in future periods. There's a small amount of this that's opt-ins, but the majority of it is just contracted revenue where people have, they're just licensing our data and have to fulfill their contract terms. We don't include milestones or bio bucks or any weird number in this. This is just real money, real cash. We also disclosed that our net revenue retention, which is the second metric we look at related to our data business, was 126%. The way to think about this is like same-store sales. So this is effectively if a client was licensing $10 million of data last year, how much are they licensing this year? That went up to 126%. So in this case, they'd be licensing $12.6 million.
It speaks to the strength of the enduring strength of the data business as people consistently, year after year, license more data and renew contracts. We also tried to give a breakdown of how this TCV relates to our forward bookings. As I mentioned, we've never been in a better spot in terms of visibility to our performance next year. We have $350 million of TCV related to 2026, which is an extraordinary amount given that it's not atypical for us to generate about a third of our data revenue within the year. We have 100 salespeople selling these products. There's needs all the time. People come to us and say, "I'm looking for this data or this analytic product." As a result, that gets recognized within the year.
So one of our biggest challenges going into 2026 is how do we control the growth of this business that is in just an incredibly strong spot. Switching for a second to algos, which is our third product line. The one business is we call it diagnostics. The other we call it data and applications. Applications, which is that component, is made up of a series of algorithmic diagnostics or algos that help route patients to the right therapy, route patients to the right trial, or make some kind of prediction. And that's our applications business. It operates at scale in that we have algos deployed across a significant population, but it operates at very low revenue because essentially there's minimal reimbursement in this category. So even though we have these AI algorithms clinically at scale, the revenues are quite small.
So we think long-term that changes, and this becomes quite a large business, but today it's operating in a small state. As I mentioned, it's made up of predominantly three product lines: matching patients to trials, closing care gaps, or producing algorithmic insights. Our trial matching business we call TIME. This is essentially we're scanning clinical data in real time, identifying patients who are a perfect fit for a trial based on inclusion and exclusion criteria, and then matching them to that trial. Again, operates at significant scale. We find and enroll a huge number of patients, small revenue. Our next product closes care gaps. Same thing. It's reading clinical data in real time. It's finding patients that have fallen through some very clear care gap. They're a non-small cell lung cancer patient. It's NCCN guideline. They should be tested for EGFR. They weren't tested for EGFR.
The provider thought they were or made some kind of mistake. We want to catch that and ensure that no patient falls behind, and so we close that care gap. Operates at significant scale, very low revenue. Same thing in our algos business. This is us essentially running an algorithm maybe on a pathology slide or radiology scan and producing some kind of insight, typically unpaid for. We expect that over time, AI will get paid for. I can't think of any other way that we avoid cataclysm in this country besides bringing AI to healthcare given the rising costs. And so as it gets paid for, companies like ours that happen to have broad distribution of AI in theory should be beneficiaries of that.
So we try to show in this investor deck one use case of that, which is our, in addition to oncology, we operate in other disease areas: radiology, pathology, neuropsych, and cardiology. We showed one use case for cardiology, which is essentially we developed a series of algorithms based on 12-lead electrocardiograms. We developed one to predict undiagnosed atrial fibrillation and one to predict undiagnosed low ejection fraction. We have several others in flight with the FDA now, and we expect to have a whole suite of these things in market. These algorithms are deployed, again, as I mentioned, at scale. So we could have 50 different providers that have these algorithms deployed, touching thousands of physicians and millions of patients. So the impact is large, but the revenue at the present moment is small. One of these algorithms, our ECG algorithms, recently got reimbursement from CMS.
It got a CPT code and approved reimbursement at $128 per algorithm run. So we've just begun. We're in the early stages of rolling this out with provider partners. We announced Northwestern Medicine. His leadership happens to be six feet in front of me a few weeks ago. And this is just one example of rolling these algorithms out to world-class providers who are starting to say, "Hey, it's not okay that I have 3% of patients who get an ECG are told they're fine and yet have a heart attack or stroke within a year. If I can use technology to predict that, I want to predict it." And so as these things get rolled out, we expect they will scale, including financially, given the reimbursement codes that now exist.
To put this in perspective, just this one algorithm, this one ECG-based AFib algorithm, if it were rolled out at scale, would produce hundreds of millions of dollars of revenue. And I would suspect over time, some company, I think it'll be Tempus, but some company will have dozens or hundreds of these things operating, and you can kind of do the math. Even though that product line is small, AI is at the center of everything we do. It literally touches every product we have. It's instrumental to every product that we have built. It's integrated in our diagnostic pipeline. People consistently ask, "How are you growing so fast? How is it possible that you have best-in-class unit growth rates?" And the answer is we've been saying for a long time is we have the most comprehensive and technology-advanced platform. It helps doctors make decisions, and so they use it.
They use it regularly, and AI is embedded into that entire product suite. Plain and simple, we make sure patients are on the right therapy and the right trial more often than others, and as a result, physicians increasingly are coming to our platform. In addition to that, AI is embedded throughout our data business, and in particular, we announced about seven or eight months ago that we've begun building the largest multimodal foundation model on oncology in partnership with AstraZeneca and Pathos. This is a large multimodal model. We procured a compute cluster of 1,008 H200s, which is being leveraged for this model. We've begun building a second model using a cluster of 524 GB200s.
So for those that are thinking about AI, these are very large compute infrastructures now leveraging our 400-plus petabytes of data to build very large multimodal models, which we assume will be catalytic both to our diagnostic business and to our data business. And this is largely just happening in oncology in the United States. We are in these other disease areas: neuropsych, cardio, radiology, rare disease, and digital path. But they're still quite small. By % of revenue, I don't know the number, but it's unbelievably high, the % of our revenue that's U.S.-based and oncology-based. And I would suspect over time that won't be true. If we fast forward a decade from now, I would imagine that Tempus will have meaningful revenue in cardiology, in I&I, in neuro, so on and so forth, and in other markets outside the United States.
Really quickly, just in terms of updates, we did $1.27 billion last year. It was ahead of our guidance. The growth rate was fairly extraordinary given the acquisition of Ambry, but even without Ambry, it's about a 30% growth rate of our core business, again, best-in-class. Our data business grew even faster in Q4. It was 68% growth, I think, when you take out the one-time AZ warrant in Q4 of 2025. So diagnostic business growing well, data business growing well. And as a result, that puts us in a great position for our long-term guidance. We've told the world we expect to generate 25% growth over the next three years. That would equate to about $1.59 billion this year.
And even though we could generate lots of EBITDA given the growth of our gross profit dollars, given that we're early in our growth cycle, we're investing two-thirds of those back in the business. So we should generate about $65 million of adjusted EBITDA this year, which is an improvement of maybe $100-$150 million. So not small. On that note.
All right. Great.
Any questions?
Yep. Thanks. Yeah. I guess maybe to start the Q&A session, I have a few on the pre-announcement. So you pre-announced the top-line beat driven really by diagnostics. So can you just walk through the performance in that business in the quarter? How should we think about volume growth across xT, xR, and xF? I think historically xF has been growing a little faster than xR and xT. So just curious if that was the case in the quarter.
Do I have to push this thing or does it work?
I think you're good.
I'm good?
Yeah.
We had significant growth across all of our therapy selection assays. Our liquid product did well. Our solid product did well. We didn't have one particular area that carried the majority of the growth. It was really quite diffuse. Liquid is growing slightly faster than solid, I think, across the board for a variety of reasons. But at the end of the day, we're lucky that we have kind of the gold standard of solid tumor profiling. We have a liquid product that people love. And so both are growing. Of our 28% growth or whatever it was in Q4, a couple of points was related to our MRD product line, but the majority is non-MRD products, which just speaks to the strength of our therapy selection assays.
Oncology ASPs came in above our expectations. Can you talk about what drove the strength there? Was this largely from the xT CDx migration? And then I guess on the latter point, where are you in terms of the portion of xT tests that are being converted to the CDx version relative to your 40% goal by the end of the year?
Yeah. We haven't disclosed the exact percentage. I mean, obviously, we put out kind of flash numbers for J.P. Morgan, and the full numbers come out when we file the 10-K. But there wasn't. We continue to be up and to the right in terms of adoption of our xT CDx assay, but nothing material. We're fortunate that we have a business that has, as I showed in that slide, really significant ASP tailwinds. And if we are concerned in trying to jam those through too quickly, is that all it does is create accelerated growth that you have to lap at some future period. Same with our data business, which is like on fire. So we think a lot about sustained growth over long periods of time as opposed to accelerated growth that you have to lap.
So whereas other companies might be trying to grow even faster and then have these very wonky 30% growth down to 15%, we believe it's more prudent to just be slow. So I suspect ASPs will be in our favor for quite some time. I suspect you'll see significant improvement over the next several years, but we want that to be every quarter up and to the right instead of lumpy.
No, that makes sense. Maybe moving on to data and services. So 4Q is typically a seasonally strong quarter, and you did see a nice step up there relative to 3Q. Maybe walk us through customer demand trends, including any bookings color if possible, and then what the setup looks like for data in 2026. During the presentation, you noted you have $350 million of the current TCV that relates to 2026. So maybe just talk about what percentage of that you think is locked and loaded and non-cancelable, and how should we think about additional bookings on top of that in 2026?
Yeah, it's all non-cancelable. I mean, we have opt-ins that are way out in 2028, 2029, whatever it is, but the 350 is essentially the contracts we have that we'll deliver in next year. So I think we're just at a moment where our data business is kind of firing on all cylinders. And as I mentioned on the podium, our biggest challenge is going to be containing growth. One of the challenges we have in this space is that a bunch of folks who understand diagnostics don't understand data. And so they don't know how to think about it or value it, and so it can have no value. But in reality, the data business is, in many ways, far more durable than the diagnostic business. It's just that people understand the diagnostic business. The data business is made up of an enormous number of clients.
You have 19 of the 20 largest pharma companies licensing our data, 250 biotechs. These are lots of contracts across lots of people who have our data embedded across their entire portfolio, and when I say embedded, what that means is that people have downloaded our data. They've brought that data into their environment, their data warehouse. They have it as a part of multiple tools that they're using internally to make critical R&D decisions. It's likely a part of multiple regulatory filings, and so it's embedded, and that's one of the reasons that people want to sign up for multi-year access and make sure they have discounts because they realize they're going to need this data. They can't just turn it off.
And so we just see our data clients licensing more, buying more, signing up for longer periods of time, making bigger commitments, again, even though they don't have to. And we saw that trend pick up at the end of last year. It continues to pick up. And so the data business is just crazy strong.
Yeah. Maybe we can dig into the data business a little bit. A lot of us are healthcare folks here. So maybe help us understand, first of all, the moving pieces of Tempus's. It's now greater than $1.1 billion TCV. So there are around $300 million of future opt-ins, as you've talked about. As of the 3Q filing, there was around $360 million of non-cancelable performance obligations related to multi-year contracts. So that leaves around $440 million of the remainder, call it. Just help us understand that last piece, whether those bookings are cancelable, your level of visibility into those, and what those customers are saying about eventually committing to data sets.
Yeah. So as we've disclosed historically, the only amount of our TCV that's cancelable is what we call opt-ins. So we'll say it, and people will be like, "But we keep saying it." There's only two things that matter. There's bookings, and there's revenue. So at the end of the day, you can get as lost as you want to get, but in the world, we have like, "I sold something, and I delivered it, and nothing else." So everything ultimately has to basically flush itself out in that. So if our bookings are a certain amount, and our revenues are a certain amount, and our bookings are a certain amount, and our revenues are a certain amount, you can kind of back into the strength of both. So we have $1.1 billion of effectively bookings.
There's a certain amount that's opt-in, which we've disclosed, which means the rest of it is contractual, and it will deliver. Even the opt-ins, by the way, we call out that they're opt-ins, but I don't know how they would kind of go away. In other words, people might say, "I'm signing up for five years of data, and I'm going to pay $100 million, but I want the right to continue in year six and year seven." So now I get to the end of five years, and they have our data embedded across 100 people in their company, 100 tools, 100 projects. If the contract ends, they have to delete all that data, remove it from every server, take it out of their systems, retrain everybody. How does that happen?
So we haven't seen anyone ever leave us, and I don't know how easy it would be to even do so. It's not simple. So if you're buying $500,000 of data, it's easy to say, "I'm done with Tempus." But once you get to the level that we're talking about, these strategic partnerships where people have us embedded across their ecosystem, very hard. I don't even know how it would happen. So if you said to me, "Place a bet on the likelihood these opt-ins become real revenue," I'm like, "100 to 1, any amount of money you want to bet.
Okay. Fair enough. Maybe one more on data. We've seen other diagnostics companies start focusing on building out data licensing offerings. Are you starting to run into other players in the market in data? And beyond that, maybe the size of Tempus' data, is the direct data pipeline with hospitals a key advantage for you guys relative to some of the other players that are coming into the space? And do diagnostic peers that run larger panels have an advantage just given the added data that's collected per patient?
Yeah. I mean, so the good news is that the companies we compete against in diagnostics, the good news is that all but one are public. The other one's also public, which is part of Roche. So Caris is public, and Guardant's public, and Natera's public, and we're public. And so you can see it. And so you can see the fact that, at least at present, nobody has a real data business. And if you go back and look three, four years ago of some of the thoughts people had about building a data business, they thought it would be much larger. They thought it would be much larger. I thought it would be much larger.
If you had just said to me three years ago, "How good are your diagnostic competitors going to be in data?" I would have said, "Very good." And I would have said, "It's going to be-" and I think I actually did say this. If someone can go back and look at old JPM, I think it's going to be like, "We're going to be like AWS, but they're going to be like GCP and Azure." It has not turned into that. It's like we're AWS, GCP, and Azure in one company, and they're like half of Oracle or like a third of Oracle. So it's that extreme. And as a result, if you look at last year, I don't think I ever heard any one of our sales reps or leadership come to me at any point in time ever and say, "We didn't get a data deal.
Somebody else got it." "Zero." So that's extraordinary. And I don't think, "Great." I would much prefer they had better data businesses so we could collectively educate the market on how valuable this data is. But at the present moment, probably because we're a tech company at heart and a diagnostic company second, we've invested enormous amounts of money, well north of $1 billion, in amassing data and building tools that make that data useful. And so when somebody gets our data and our tools and somebody else's data and their tools, they're like, "Oh, that's horrible, and this I like." And as a result, even people who've signed up other people come to us afterwards and say, "I need your data." In fact, if you look at everyone who's been announced to our competitors over the last three or four years, all of them are licensing our data.
Every one of them. So I think it just speaks to the fact that we've invested in the technology and tools.
Okay. Gotcha. That's pretty helpful. I guess moving over to genomics in the last few minutes we have here. I have a question just on MRD. When looking at xM's test performance relative to peers, the longitudinal specificity is notably lower at around 90%. Peers are closer to 98%. Has that come up in conversations with MolDX at all? And what gives you confidence that physicians won't prefer to use competing tests with higher specificity at comparable sensitivity?
So it has not come up with MolDX even in the back and forth, but I don't have any confidence that they're not going to prefer to use other people's tests. They do. They will. And that is happening today. We have two products in MRD and market. We have a tumor-naive xM test largely in CRC, and we have a tumor-informed product in partnership with Personalis in a bunch of other subtypes. This is directionally right. The vast, vast, vast majority of our volume is tumor-informed. In CRC, it's even more pronounced because you have enormous amounts of tissue. So why not do tumor-informed? So I'm not worried that in the back and forth, we could eventually get MolDX reimbursement. I am worried that that product is not taking off relative to our informed product or other informed products.
We could see this like six months ago, and I think we talked about it in our last earnings. We began to realize that even though we could get this thing reimbursed at some point, we're not going to get adoption, and so we worked on a second version of the naive product, which we talked about, I think, last quarter, and we have studies running in many disease types at the present moment, first being non-small cell lung cancer. We'll bring it to CRC quickly now. That is a better version of that assay. It has significantly improved sensitivity and specificity, and it does compete on the numbers, at least at present, more appropriately with assays like Signatera, and I think reimbursements are relevant.
Market adoption, we're going to have to have a better naive product to win the market, as I think our others, or it's going to be an informed market, so we're investing in both of those, so we're lucky that our economics relative to our informed product are really good. We make a bunch of money by selling those tests instead of lose a bunch of money, and we're getting adoption. We're seeding the market. We're getting share, and eventually, we'll have to figure out the right portfolio to maximize, but in the near term, it's going to be largely informed.
Okay. Maybe last couple of minutes here. Just looking ahead to 2026, you're pointing to $1.59 billion in revenue. Maybe walk us through the moving pieces there in terms of what you're expecting from contribution from diagnostics versus data and algos. And where would you see the most upside in 2026 between the businesses?
I don't have the breakdown of the exact percentages diagnostics to data, and we haven't given that yet. We'll give some color, I'm sure, when we announce full earnings. Both businesses are performing super well. We've got really strong unit growth in diagnostics with rising ASP, and our data business is crushing it. I mean, I think to the extent there's significant overperformance, I mean, our goal, this is in a meeting, outside of diagnostics, in the normal IR world, you want to give guidance and slightly beat and slightly raise for a variety of reasons, and so we think that's the right approach. We don't want to give guidance that might be sandbagged and have massive beats based on cash you collect that you probably knew you're going to collect from three quarters ago. We don't do that game. We try to give guidance that's relatively intelligent.
That's why we give people a number and in theory, we want to beat and raise that number, but in appropriate amounts. If I give a number and beat it by 20%, it means I'm not great at forecasting. To the extent that happens and we have a significant beat in 2026, it likely will come from either the data business, which we're having a hard time controlling the growth because it's growing that fast, or one of these ASP levers, we just can't stop it. It just happens, and it creates really a one-step-up function in ASP. I would much rather have 25% growth for three years than have that happen but it is possible that we can't contain the growth but we're very focused on predicting our business conservatively and intelligently and being in a position to beat, but not beat in crazy ways.
All right. Well, it looks like we'll have to leave it there. Eric, thank you. And the Tempus team, thank you. Enjoy the rest of the conference, everybody. Thanks for coming.