TD Cowen, Tools and Diagnostics team. It's my pleasure to welcome Geoff Porges, CFO of Schrödinger.
Great. Thanks very much, Steven, and welcome everybody. Thanks for coming to the presentation. I will be making some forward-looking statements, so please refer to our SEC filings for all the necessary disclaimers. Let me try and give you a quick overview of how I explain what Schrödinger does. It's really captured in this graphic. Computationally, we're trying to enable protein structures and then model the interaction between small molecules, typically, but in some cases also large molecules with those protein targets. This graphic here represents all the possible range of R groups that we are in the process of modeling for its fit in the binding pocket in the case of this particular protein. This technology is built on more than 30 years of really intense research.
We've calculated that cumulatively more than $800 million has been invested in the development of this platform. It enables the design of new molecules with specific desirable attributes. We can model these attributes computationally with the same reliability as actual physical experiments. And it really opens up the possibility of discovering truly novel chemical structures that are way outside the domain of existing molecules in particular therapeutic areas and indications. Now, we combine physics-based methods, which are the basis for those calculations of the attributes of protein targets and ligands and the interactions between them, with machine learning. Machine learning enables us to execute these tasks with much greater scale. Instead of doing thousands of molecules, we can model the interactions of billions of molecules.
So our technology doesn't start with AI or machine learning, but it is dramatically accelerated by the deployment of machine learning, and we have a substantial machine learning or AI group in-house. And so the combination of the physics-based methods and the machine learning enables us to execute this exploration of novel chemical space very quickly on a massive scale. And this is where our large customers are now headed. And of course, the goal of deploying this technology is to select the best molecule. And there's, as you know, a nearly endless array of attributes that we can model. We started with potency, binding affinity, selectivity. We're now modeling things like solubility, bioavailability, clearance, half-life, and we're actually moving into the area of drug-drug interaction liabilities and off-target toxicity with the modeling recently of the interaction of small molecules with hERG.
We're now at the point of predicting things like QT prolongation liabilities for molecules that might come out of the explorations that we're doing. So how do we make money out of this platform? Historically, the company was built on software licensing. We have some 2,000 customers spending over $1,000 a year licensing our technology or 1,785. And we've. I'll share some KPIs, but our largest customers are now spending well into the millions. We have collaborations with large and emerging companies. Those are being quite successful, and we have 17 collaborations that we've built up so far. And then we have a proprietary pipeline that's growing.
We have two programs in the clinic now that I'll talk about briefly, and one more going into the clinic this year, and a number that are just behind heading towards the clinic over the next year and a half or so. So just on the software business, its business generated $159 million in revenue last year, grew 17.5%. There were 54 customers spending over $500,000 a year, 27 spending over $1 million a year. The ACV of our top five customers was $6.7 million. We've disclosed that there were four customers spending over $5 million. So obviously, the top end, it's well north of $5 million, though I can't give you the exact number. We had a very high retention rate in those larger customers, 98% in 2023, 100% in 2022. One customer was acquired, which was the reason it wasn't 100%.
We did see an increase in the drop-off of our customers. So our retention in the 100,000-500,000 went down somewhat, and that's the overall retention went from 96%-92%. That increase was due to some of the smaller biotech companies who'd been customers getting out of doing research. We still see a lot of opportunity here. There's a 20x difference in the contract value for us between our top customer, global pharma customer, and our number 10 global pharma customer. So, as you can imagine, we're very focused now on bringing all of the other global companies up to that upper range. Just in terms of the collaborations, we now have a number of approved molecules with the Agios drugs that you're all familiar with.
The TYK2 with Takeda is now in phase 3, a number of programs in phase 2, the ACC inhibitor at Gilead, and then a number of the Morphic programs, or one Morphic program is in phase 2, a number of Morphic programs are earlier. And then we have collaborations ongoing with Structure, Lilly, etc. So a broad group of collaborations. We signal that we are, I would say I don't think it's a pivot, but let's just say we're evolving from collaborations more towards proprietary molecules. We're just seeing that, that once we have the financial resources we have now, we can retain a lot more of the value of, of the programs if we take them further. So these are the, the proprietary programs that, that we've disclosed. We have a MALT1 inhibitor that's in phase 1, in, in B-cell malignancies. That's recruiting nicely now.
We have a CDC7 program that's in phase 1 for AML. We have 3515, which is the WEE1 inhibitor that we've indicated is going to go into the clinic this year. And then behind that, we have an interesting group of programs, PRMT5, EGFR, NLRP3, and LRRK2. We've signaled that we, our goal is for at least one of these programs to be IND ready in 2025. A little bit more about the lead molecules. The MALT1, we completed a 73-subject phase 1 healthy volunteer study last year. We announced those results at our pipeline day in December. I thought these, this was pretty intriguing, but it was only 10 days of exposure. So the pushback on that was, okay, you've got a safe molecule. It looks like it's pretty well tolerated, but it's only 10 days of exposure, and it's not cancer patients.
What we confirmed on our call last week is that we've now been able to accelerate the recruitment in the cancer study, and the safety and tolerability and, frankly, PK and PD that we're seeing are exactly the same as we'd seen in the healthy volunteers, and that all the subjects enrolled in the cancer patient study, which has been ongoing now for about 15 months, remain on study. So encouraging, we think that we can expect to see data late 2024 or 2025, depending, of course, on the recruitment and when we get enough data that we think it's ready to disclose at a conference. The CDC7 program, this is a target that we think is potentially really implicated in AML. However, this is a challenging patient population, and we're aware that this is sort of a Bermuda Triangle for drug development.
So, that study is now recruiting. There's a lot of enthusiasm among the academic community about this study, about this different mechanism, but we're going carefully, as we dose escalate, because of the risk of toxicity in this patient population. We still think that we can get data from this study also late 2024 or 2025. Lastly, the WEE1 program, you're all familiar with the Zentalis data. We view that as a sort of prototype molecule. That molecule, we know, is very similar to the precedent AstraZeneca molecule, and ours is a very, very different structure. We think has the opportunity to have quite a different profile. We're filing the IND in the first half of this year.
We've indicated that we plan to start clinical trials in the second half of the year, and it's reasonable to expect that that will be a reasonably quickly recruiting patient population to be mixed solid tumors. Quickly, financial highlights. We just reported Q4 results. This is Q4. Total revenue was up 30%, $74 million. Software revenue was $69 million, the biggest quarter that we've ever had. We indicated that there were a couple of large multi-year deals that contributed to that, and I'm sure we can talk about that with Steve. Drug discovery revenue was down because there was a large milestone payment in 2022 that contributed to that $9 million. Gross margin was up nicely because of the incremental revenue on the software side. Operating expenses came in at $87 million, so our net loss was around $31 million.
Our cash at the end of the year was $469 million. That cash doesn't include roughly $80 million in equity and public companies such as Structure and Morphic, which we have some restrictions, as you imagine, our ability to liquidate, and that's why it's not included there. Just for the full year, total revenue was $217 million. Operating expenses were $318 million. Our net income for the year was $41 million because of the distribution from Nimbus that we received in the first half of the year, $147 million. Our guidance for the year, this is, beginning of the year guidance based on what the information that we have now. Software revenue we're guiding to 6%-13% growth. That's coming off, and this is in revenue, not ACV. This is coming off that very strong contribution we had in the fourth quarter from the multi-year deals.
If I look at 2023, we had 17.5% software growth. ACV growth was only 10%. So those two have to reverse themselves. The structural growth features of the business we think are in the mid to high teens, but the revenue necessarily is going to be lower because of the fourth quarter bolus that we have to absorb and get past. Drug discovery revenue, we've guided to $30-$35 million. We've stripped out business new business development, milestones, from progressing programs in other companies' portfolios where we don't have visibility, so again, we're, we're taking a pretty cautious approach to that. We expect a gross margin to be similar. Operating expense growth coming well down from 2023, guided to around 10%, and then, cash use and operating activities will be higher. And then lastly, milestones, I mentioned already. We're excited about submitting the IND for 3515.
That's the WEE1 inhibitor and starting the phase one in the second half of the year. We're hopeful that we'll get some data either late this year or next year for both, the MALT1 inhibitor and the CDC7. We're preparing for an IND submission in 2025 for one of the other molecules I mentioned. We do anticipate that we'll be launching a significant update to the force field, which is a core part of our technology suite. You're probably familiar that we have a funded research project with the Bill & Melinda Gates Foundation exploring new battery chemistry, and we're making really significant progress with understanding and modeling computationally different materials for batteries, and we think there'll be a publication at least coming from that next year. So there we are. All right. Great. Thanks, Geoff, for the introduction.
If people have any questions, you can email me at Steven.Mah@cowen.com or just raise your hand, and we can do it interactively. So let's, let's great. Let's dig in a little bit more on the software business. You know, talk about Lilly. I mean, they did a significant kind of re-up on their software agreement last quarter. You know, what does that say about in terms of like validating the platform and the value add that you know partner like Lilly is getting? Yeah, we were really excited about the Lilly announcement. We'd hope that we could actually put numbers beside it, but let's just say they, it's fair to say they're our largest customer. And it's also a really interesting journey.
Lilly's been a customer for a better part of a decade, and we went back and looked at the progression, and, you know, they sort of went through several plateaus before they got to the level, and this level is significantly above where they were before. They sort of stepped up, got experience with the technology at a certain level, found its utility, and then stepped up again, rolled it out to, you know, new programs, new teams, new therapeutic areas, new sites, and then several years passed, and they stepped it up again. So we're really excited about this step up. We do think that they're kind of a bellwether for the rest of the industry. We've already started having conversations with other companies where they're saying, "Wow, that's a pretty big sort of deal." And we can't say specifically what it is.
They, they don't want us telling people what the number is, but you can kind of, if you take apart our results, you can pretty much see most of the revenue growth, let's say, came from the multiple multi-year deals, and there was another one as well that I can't mention. So we had two or three in the fourth quarter, but the biggest component by far was Lilly. And just to explain, you know, since we've done a lot of it, gone through a lot of this, the Lilly contract is what we call on-prem, which means that we distribute the licensed server to Lilly. Because of that, we recognize the most of the value of the contract when that transaction occurs. And so let's hypothetically say it's, I'll just say it's $5 million.
It's much significantly more than that, but let's just say it's $5 million times three years, $15 million total value, 65% of that is recognized at the time that we sign the contract. Now, there's no cost associated with that. That's why the gross margin goes way up, and it causes a big spike in revenue. So we absorb that on a very large contract in the fourth quarter. We have to grow beyond that in 2024, which is how you get to the revenue guidance. Now, the ACV is adjusted. It's just the annual contract value so that, in effect, surplus revenue in the Eli Lilly contract or in the other multi-year contracts is stripped out from ACV. So, you know, ACV was, as I said, around 10%. We expect that to accelerate this year.
We see lots and lots of opportunities for ACV to grow faster, but revenue in 2024 will necessarily grow more slowly than ACV. Right. Okay. No, that makes sense. And is, do you see a broader shift to on-premises versus hosted? No. Originally, pretty much all of our large customers were on-prem, because they wanted to host the licensed server. They, they were concerned about security, concerned about that we could track the utilization of different parts of the software suite. That concern has fallen away, fallen away. And if we host the licensed server, which is not actually hosting the software, it's just hosting the little, it's almost like we're, we're hosting the toll booth on the turnpike. If we host it, we get to see their utilization of different parts of the technology stack, how many tokens they're using at any one point in time.
Some customers like that. Other customers don't like that. Over time, we would certainly like our business to migrate towards hosted because it's much smoother. It's recognized readily across the contract. But we're not doing that quickly. And if a customer says, "Absolutely, we're adamant about staying on-prem," then we're going to respect that. Okay. Yeah, understood. You know, and just to further dig into the software business in terms of, like, growing, is like Lilly a good example of how you can get a kind of, you know, an early user who maybe isn't using a lot of software licenses? Give us some color on how that kind of builds up. Yeah. And if there's any way you can tweak that. And is Lilly like a good representative example of what, you know, a long-term 10-plus-year partnership could look like?
So first of all, there are three other companies above $5 million, and you can see in the ACV, there's $6.7 million for the top five customers. So they're not at $5 million and one. They're significantly above $5 million. Those are all accounts that we've had long-standing relationships with. That trajectory, that is, is very similar. It's sort of 5- to 15-year duration of relationship with kind of build, plateau, build again. Now, we have 160 people in our sort of customer-facing organization, tech support, and sales. Mm-hmm. They, many of them, in fact, most of them, have scientific background, and they will work with these accounts to deploy the technology and make sure it's working. They give us feedback on what we need to do with the technology. But more importantly, they help them use the full capacity of their license.
So it's our expectation that within 6, 12, 18 months of signing a contract, they hit that cap in terms of the size of their license, and they come back to us and say, "Okay, next time around, we need more tokens," which is how we license the software. So the interesting thing is the Lilly agreement clearly was a significant step up for where Lilly was before. But we have even more opportunity in sort of the smaller large companies. And some of the large companies that, you know, are household names in the industry, largest R&D spenders in the industry, are way down. They're literally an order of magnitude lower. And we're having discussions with them now because they're saying, "Okay, Lilly's figured this out. A few other companies have figured this out.
You know, we should be looking hard at this." And I think that a number of those discussions will come to fruition. Some this year. It's just a little early to be certain about how much they'll contribute. Right. Right. Is there a sense of an arms race? I mean, I've heard that before from a, you know, number of companies that, you know, especially in, like, AI. Mm-hmm. You know, there's sort of like an arms race. People are trying to get as many NVIDIA chips as possible. People are building their own supercomputers. So it very much feels like an arms race. You know, how does that, you know, is that sort of the sense you get? And, you know, how does that bode for your software business? Yeah. I'm so our software is almost exclusively run on their cloud instances.
So they already have a relationship with Amazon or whoever it is. And so, you know, those, those cloud resources aren't really restricted. But as they demand more, then that drives more demand for NVIDIA GPUs, for sure. So they are not really, they're not capacity constrained. It's expensive. And they, if, if they're at the sort of Lilly level, they could easily be spending millions of dollars on the compute as well as on the software. So, and, and that, you know, if you look through our financial statements, we have a significant line for our compute costs to support our program. So, you know, I don't think they're running up against that as an issue. One of the things that we are finding is that there's still a shortage of trained computational chemists. Okay. And these are not medicinal chemists. They're not synthetic chemists.
They're, they're sort of, it's a completely separate discipline. And they are being hoovered up. We have more of them we think than anyone else in the world, but they're being hoovered up quickly. And a number of our largest customers, global companies, household names, say to us, "We want to use more software, but we can't find someone to run our computational chemistry group." And so we're, we're sort of almost in the business of helping them find computational chemistry. In fact, one of our large customers has hired one of our employees. So frequently, those scientifically trained folks in the field go to work for our customers and become heads of computational chemistry or running projects for them, and they're very familiar with the software. So probably the biggest bottleneck is that, which is trained computational chemists to run the software at scale on multiple programs. Okay.
No, that makes sense. No, but I guess more in terms of, like, you know, maybe, like, competitors to Lilly or peers, you know, saying, "Okay, well, they're taking all these licenses. You know, we're only taking, you know, a fraction of that. Like, shouldn't we, like, re-up our game?" That's sort of what I was getting at as well. We're definitely having those conversations. I mean, the interesting thing is there's nobody out there who doesn't know who Schrödinger is, what we do in the land of sort of synthetic chemistry, computational chemistry, etc., and drug discovery. So, it's a question of what the scale is. I mean, every company that comes up with a novel pharmaceutical product is on our customer list.
But there's some companies who consider themselves global leaders, who consider themselves advanced computationally, who are, let's say, more than tenfold smaller than Lilly's purchase, who are starting to ask themselves, "What are we doing?" And what we're seeing now is, you know, there's a trickle-down effect. Either CEO changes or head of R&D changes. Then they say, "Okay, we've got to get a new head of drug discovery, or we get a new head of digital." That new head of digital comes in and says, "Okay, Schrödinger is a core kind of foundation of what we're going to do going forward." And they come to us and ask for, you know, a significantly increased token library. And that's a discussion that we're having right now with, you know, a number of global companies.
A lot of that continued from last year, but there's new conversations happening as well after the Lilly announcement. Okay. I appreciate that color. Maybe let's pivot over to your internal drug candidates. You know, as, you know, are there certain, you know, stage gates, you know, you're going to go to before, you know, monetizing them, or is, you know, what is your intention on how far you'll take these into the clinic? Yeah. I think it's clear that we can't possibly take all of the programs that we've come up with already, through even mid-stage, let alone late-stage clinical development. That would be overwhelming in terms of its cost. I think it's also clear that some of the programs we have aren't really suitable for a small company to be investing in.
So a good example would be, let's just say, NLRP3, which is one of our late discovery programs. If indeed NLRP3 has utility in obesity and a broad array of inflammatory diseases, we're not going to be doing pivotal trials in those sort of indications. Now, we will do enough to really drive value and to demonstrate the unique attributes of our molecules. So in the case of MALT1, I mean, I think most people are familiar with the J&J data that we think validates the target. But we have to get some human clinical data to show that our molecule is differentiated, that the attributes that we designed into the molecule have panned out in people. So we're in the process of doing that.
But we think that something like MALT1 could be pretty attractive if we show that we don't have, you know, bilirubin grade 3-4 elevations, that we don't have renal cardiac tox, then that's what's necessary for that molecule. So it's a case-by-case basis. Got it. But I don't think we're going to be taking a whole portfolio through to late-stage development. Okay. And what about you, you know, kind of your prior strategy of kind of spinning off, you know, assets into, you know, VC-funded entities? Is that still a strategy? Yeah. No, no, we're really excited. The returns have been terrific from our venture activity. Those returns have mostly come from companies that we've co-founded, things like Nimbus, Morphic, Structure. We have another one that we're working on now in the hematologic malignancy area.
We're getting very interesting proposals coming into us, because I think people understand that we're open for business in terms of deploying our technology and our expertise into a company, helping them come up with a really unique molecule in return for equity as well as in return for cash. So, yeah, that's a really active part of our business model. Okay. Got it. And then, maybe just, you know, going into, you know, other applications. So right now you guys are focused on small molecules. You know, can the platform be applied to biologics as well? Yeah. Look, we can model the interaction between an antibody and a target very effectively. So we can measure things like binding affinity. We can measure pH, pH-dependent, loss of binding and degradation. That's useful for predicting half-life.
I think that if you think about the breadth of the universe of small molecules, which is massive, the breadth of the universe of large molecules against a particular target is much narrower to begin with. And so the utility there is lower. But we think that our technology is still very valuable. We also have a thing called LiveDesign, which is an enterprise informatics platform. That was built really to accommodate all of the elements of our small molecule suite. We're now deploying LiveDesign for biologics, which we think will make it easier for people to use some of our other tools in for the design of biologics. So I don't think I wouldn't say that I think biologics is going to be a large piece of our business, but I think it can be significant.
O kay. Got it.
And this enterprise informatics platform you're talking about, you know, are you guys monetizing that, or are you, is it part of the software package to your licensees?
No, we actually price it separately. Okay. And it's priced on a per-seat basis. And there are some customers that have hundreds or thousands of licenses to that and then a relatively small number of licenses to the core suite. And there are other customers who have a lot of licenses to the core suite and very few licenses to LiveDesign. So it's sort of, it's marketed by a commercial organization. It's complementary to our core suite of technologies. But some customers gravitate towards one or the other. Okay. Got it. And then maybe just continuing on with kind of these foundation models and platforms.
I know you have a partnership with NVIDIA. We just hosted Kimberly Powell next door, you know, 30 minutes ago where she was saying, you know, AI and drug discovery is, you know, going to be NVIDIA's biggest opportunity here.
You know, are there ways to monetize your software on, for example, their BioNeMo platform or on the Google platform?
Yeah. That's an interesting question. I mean, we don't actually control the data that is produced using our technology in our customers. We're arm's-length because, to use a Wall Street term, we don't want to be seeing their flow and trading on the basis of their flow because then they wouldn't sort of buy our software.
So we don't accumulate a lot of data, which is part of what BioNeMo is trying to do, is to create a sort of large data repository. So that hasn't really made sense for us. Now we do have a longstanding relationship with NVIDIA because of the demand for GPUs. And they did a lot of work in the early days of that relationship to enable our technology to work on their GPUs. And then we have a big relationship with Google to access their cloud for our own purposes in terms of our drug design. Because, of course, we're, if you looked at our software business, if, if we were selling to ourselves, we'd be 10 times our largest, we'd be 10 times Eli Lilly, nearly, in that range. So, so we, we require a lot of cloud resources ourselves. And so we have that relationship with Google.
We definitely have conversations about how we can accelerate our front-end using AI. Right now we're doing that with, with the sort of exploration of chemical space. That's a really good application of AI. But can we make our technology even easier to use? Can we make it so that you don't have to be a computational chemist to program up these calculations so you can be just a regular medicinal chemist, of which there are, you know, 20 times more medicinal chemists and computational chemists, maybe 100 times more?
So we're considering that. I just don't think that we're ready to make that commitment yet. But, you know, there are fairly obvious partners that we talk to about that. Right. Right. Right. Understood. All right. Are there any questions in the audience? You want to repeat the question? Oh, yeah.
The question was, you know, their internal pipeline, is there any potential to compete with, you know, other companies, essentially?
No. Other way around. Oh. Sorry. But I know. So the question was whether our collaborations restrict us from going after those targets. And the answer is yes, but for a very finite period of time. So if we are working very closely with Lilly or BMS or Takeda on a particular target, then while we're engaged in that collaboration to come up with novel chemical matter against that target and they're sharing data with us, we're basically embargoed from working on that same target with anyone else or on our own account. Usually, there's a sometimes 6-month, 12-month sunset. Sometimes it's immediate.
As soon as that collaboration ends, we hand over either the lead or the DC or whatever the transition period is. Then we can go back and work on that target again on our own account. So, there's not a long tail, is the answer, even though we still get downstream participation in the program once we've handed over. Do you do any deals, Geoff, where it's exclusive? In other words, if somebody comes and says, "I want to explore this area, I don't want to build anything else," is that something you've done? We haven't done it. We definitely contemplate it. The economics would have to be really attractive for us to be closed out of either one target or a therapeutic area. I can't imagine we would be closed out of a whole therapeutic area.
But yeah, there are some targets that are sufficiently exciting that under the right circumstances, we would do that. All right. Great. We're out of time. Everyone, appreciate. Thank you. Thanks, Geoff. Appreciate it, Steven. Yeah. Thanks for coming. Thanks, everyone.