Raw Reads: AI Trends with 10x Genomics CEO Serge Saxonov

10x Genomics CEO Serge Saxonov mentioned AI's impact on his company many times during his Q2 conference call with Wall Street analysts.

Raw Reads is a series of refined, comprehensive transcripts of conversations from leading figures covered by Ion Genomics.

Single-cell sequencing has for the last several years already formed the basis of AI foundation models of cells, so it’s no surprise that 10x Genomics CEO and Cofounder Serge Saxonov is tracking that field closely.

I’ve spoken to him about it personally, but wasn’t sure if he’d commented so much about it publicly, as he did during the company’s second quarter financial results call. AI and its impact on 10x came up in both Saxonov’s prepared remarks as well as in the Q&A session.

The following is everything Saxonov said about AI during the call:

Serge Saxonov (00:00) — CEO and Co-founder, 10x Genomics

… Over the past several quarters, products like Flex Apex have enabled a new generation of this work, particularly in biopharma and translational research. … A significant trend in single-cell has been an increase in large-scale perturbation experiments to map biological mechanisms and resolve causality. We’re finding that Flex Apex is becoming the standard assay for these experiments because of its scalability, robustness, and sensitivity. While we see significant flex Apex adoption across all customer segments, the uptake of Apex in biopharma has been particularly strong, driven by the application of perturbation screening to target identification.

The value of these studies is also increasing because of the progress in AI, which helps derive mechanistic insights from the large amounts of data generated by these experiments. As we have discussed before, we believe AI represents a significant and structural tailwind for our business. AI has enormous potential to transform biology and human health, but realizing that potential depends on generating vastly more of the right kinds of data. The key bottleneck for AI-driven progress in biology is the same bottleneck we identified when we started the company. Biology is incredibly complex; we understand only a tiny fraction of it, and solving that complexity requires measuring biological systems at massive scale and high resolution. We build single-cell and spatial technologies for precisely that purpose, which is why they are now being deployed by so many of our customers to train AI models.

In fact, AI as an influencer of demand is now becoming pervasive across our customer base. Today, most significant biological data generation efforts are conceived, at least in part, with the goal of training AI models.

On the academic side, there are multiple well-known pioneering efforts, such as those led by [BioHub] and the Arc Institute, dedicated to building virtual biology models, but we’re also seeing a wider shift, where more of basic scientific research entails training AI models. This shift is driven bottom up by decisions of individual scientists, as well as top down by philanthropic and government funding priorities, such as those outlined in recent proposals from the White House, a similar shift is also starting to happen by pharma with a rapid growth in AI-focused investments. Initially, much of the AI work in drug development has focused on the chemistry side of the process, on creating molecular interventions once a target is known.

Going forward, we expect increasing investments to be made in modeling biology at the cell and tissue level to unlock new targets and to predict drug response in patients.

We believe this is where the biggest bottlenecks are and where there are the greatest opportunities to transform drug development.

This work is also precisely what our tools enable, and why we anticipate a very large opportunity for our technologies over time. Most pharma companies now have strategic mandates to leverage AI to speed up drug development and increase the probability of success. At the same time, there is a rapidly growing number of biotech companies that seek to transform drug development using AI. More and more of them are focused on building sophisticated virtual models of human biology.

The vast majority of the companies building such models are using 10x single-cell and spatial technologies.

Customers overwhelmingly choose our products because they deliver the highest data quality, the largest scale, the widest biological context, and the most powerful multiomics capabilities. It has become increasingly clear in the field that all of these considerations are critical for building high-quality, generalizable, and useful models.

It should be noted that building better models is only a part of the AI story. For years, one of the biggest barriers to broader adoption of single-cell and spatial biology has been the bioinformatics expertise required to analyze increasingly rich datasets. Advances in agentic AI are beginning to remove that bottleneck.

Researchers who previously required dedicated computational experts are starting to analyze complex datasets through natural language interactions with AI agents.

We believe that will make single-cell and spatial analysis accessible to a much broader community of scientists, while increasing the value of the underlying data.

Together, these trends reinforce our conviction that single-cell and spatial biology are foundational to the future of basic science and drug discovery research. AI is increasing both the demand for high quality biological data and the ability of researchers to extract insights from that data. We believe those two forces will reinforce one another over the coming years.

…

Q&A

Kyle Mikson (Canaccord Genuity) (17:31)

Hey guys, thanks for the questions. Congrats on the quarter. Multi-part question first on instruments. So, on Atera, great to hear all the demand and the interest and all that, but why remain conservative with that expectation for the second half… And secondly, just quickly, Serge, on AI, getting a lot of questions on your differentiation. You obviously have a lot of, you know, strong push there, but what really sets 10x apart, to directly benefit from that? And can you talk about any tangible revenue that you’ve recognized so far. Thanks.

Saxonov (18:28)

Thanks, Kyle. So just like for your first question on Atera, as we talked about in our prepared remarks, the demand has been extraordinary, and that’s not the constraint here. Right? The constraint is actually you know the manufacturing capacity to to ship the units this in the second half of the year, and we continue to expect to be able to ship 40, which is what we said in the last call, and still continues to be the case going forward.

[Editor’s note: Saxonov did not appear to address Mikson’s question about AI-related revenue.]

Michael Ryskin (Bank of America) (31:21)

Hey, thanks, guys. I want to go back to chromium and the single-cell platform. You know, chromium consumables are kind of flat, effectively quarter over quarter, just sequentially throughout the year. And instruments were a little bit on the lighter side. You know, I understand a lot of focus on spatial and maybe Xenium, but just we’ll have to dig into more to what you’re seeing there. You know, you’ve got things like the Billion Cell Atlas ongoing. You’ve got things like Perturb-seq ongoing. You’ve been talking about AI-driven drug discovery, which I think should tap into single-cell a lot. Why aren’t you seeing a little bit stronger numbers in Chromium? Is it all really tied to Flex, or is there anything else going on? And maybe if you could just quantify what you’re seeing, give us any tangible metric that we could sort of latch onto for AI-driven demand, just so we could sort of figure out, you know, how big it is for you right now in Q2. Thanks.

Saxonov (32:18)

Yeah, Mike, thanks for the question. So yeah, I mean, on single-cell. I would say the first order dynamic by far is that sort of transition that I talked about earlier, the rise of Flex Apex and transition of some of the other products to that to that assay. Again, it’s had a really nice pickup, a really nice momentum. We expect that to continue, and as a result of that, yeah, the the volume growth has been quite consistent and very robust and very encouraging, and a lot of the growth is in fact driven being driven by kind of emergence of AI applications and also more large scale experiments that involve larger cohorts and kind of distributed sample collection that that Flex is particularly great for.

That also has a bit of an influence, in fact, on instruments in the sense that because of the capability enabled by Flex of having this distributed sample collection and then centralization of processing, that naturally leads to more centralization to our service providers, to core labs, to big labs, which which kind of reduces the necessity to be placing instruments at every single lab, and also you know at this point we do have a lot of chromium instruments out there, so accessibility is generally not an issue at all.

So, overall, the dynamics around single-cell are very similar to what we’ve been describing for the last couple of quarters, and we expect that to continue in the over the rest of the year, and I do expect, like I said earlier, yeah, all the people or like large majority of the people who intend to transition to Apex will have largely done so by the end of the year, and that should put us in a good position to you know keep driving that sort of robust reaction growth, but also having it be translated into more top-line impact as well.

As far as the AI question is concerned, there’s a lot of layers to that, and I think it’s really, really important to set context here, which is what I did earlier as well with my my prepared remarks,

Like I said earlier, and I think there’s like a wide sort of recognition of that that AI is now a major structural tailwind for us, and because you know what these big AI models need is precisely what we have built over the years, and AI at this point, one way in one shape or form, is becoming pervasive across just about all of our customer segments.

At this point, you know there isn’t really a large project out there where AI isn’t like either a big driver or at least like a you know an an important influencer. And even small scale projects, I think in many instances are performed with an eye toward feeding the data into AI models, and so you know in some cases AI is a driver of demand. In some cases it’s an influencer, and sometimes an accelerator.

One issue here is that AI can mean a lot of different things, and also the landscape is changing quite fast, and so that’s why I went specifically in in a bit more detail in explaining how AI, for example, is used potentially in drug in the context of drug development for the parts specifically to measure biology. You know, if you think about the three stages of drug development: target ID, to understand the biology of what targets to go after, the chemistry; the middle part actually making the drug, the molecule; and then figuring out which patients to give the drug to.

The middle part is chemistry, where a lot of current AI investments up to now have been focused, but the big big opportunity is really around the biology, the target ID, and patient selection, and that’s where our tools are becoming are really really compelling and are becoming increasingly important.

And so there are some parts of our revenue where unambiguously AI revenue is coming from those, like tech bio companies, some large academic projects, but there’s also others where there is a mix of big pharma companies that we know are developing these AI models of biology, but also using our products for other for other goals, and same thing in academia, and so there’s a mix.

So right now at this stage, I’d say it’s still very early, but the opportunity is massive, and our products and technologies are particularly well positioned for this opportunity. And you know, as we go forward, and as these categories grow, we’ll provide more granular color on them and and and how to think about sort of numbers around them.

Matt Larew (William Blair) (38:03)

Hi, good afternoon. You referenced a number of the larger projects you’re working on with respect to AI, and also on the translational side. And in some cases, you know, customers adopting or increasing use of multiple platforms. I’m curious as you’re having these these discussions about larger projects, multi-year projects, kind of how important the suite of products that you have, and you know, software and analysis tools where there’s perhaps some integration or at least familiarity, you know, how that kind of ecosystem might be having an effect as customers think about you know longer term projects versus you know the merits of the platforms on their own.

Saxonov (38:48)

Yeah, really interesting question. Well, you know the first order answer I would say is that the platforms by themselves, whether you look at single-cell and spatial, have really strong merits just to stand on their own. I would say, and certainly is really appealing in many ways to our customers. You know, if you think about, for example, Flex Apex, really, really high sensitivity, incredible scalability, huge robustness. This is actually really important for AI in particular, a subtle point where it works across many different tissue types, many different cell types, many different contexts, and increasingly is becoming critical. So, if you want to build AI models that are useful that generalize, you really need to be able to measure lots of different contexts. You can’t be measuring the same cell line over and over again, for example, and Flex is incredibly great for that.

Also, another sort of emerging trend is, and maybe it’s a little bit your question too. Here is multiomics, being able to measure other modalities, and again, we have unmatched strength on that on along that dimension, and and yeah, certainly there is a there is an attractiveness to be able to do yeah to to train your models from a spatial perspective as well as from single-cell, and we certainly provide kind of these kinds of solutions to our customers, and also yes on the software side something we haven’t talked that much about, but of course we have invested in software fairly materially from the beginning of really the beginning of the company, and in particular is becoming important here because the data sets with Apex and with Atera are getting to be very large, especially for training AI models, and we have made quite a number of advances specifically to enable people to to run large-scale experiments in a straightforward kind of ergonomic manner.

So all of these pieces together do tie out to provide really compelling solutions for our customers, certainly much much more compelling than any other potential alternative on the market.

Puneet Souda (Leerink) (45:51)

Yeah, hi Serge and team. Thanks for taking my questions here. … On the AI side, I mean it does appear that the biology foundation models or the virtual cell models will require fundings in the scale of 100 million dollars or something closer in order to build the data for those models. Can you maybe just elaborate on what are the line of sight to the major large grants or funding sources right now that you see and the timing for those to land into you know 10x’s revenue. Thank you.

Saxonov (47:11)

[...] You know, as far as AI and funding is concerned, I think there’s a lot of work at the highest levels happening where I kind of spoke to that, where there’s a general kind of reappraisal of funding flows.

I think, in fact, a lot of large scale projects, anything that has to do with sort of large-scale science, is now having AI as a driver.

And if you think about it, kind of the fundamental model, if you want to understand biology using AI, you have to use single-cell and spatial because they are the scale technologies to measure biology, and so whether it’s the biopharma kind of world, or the world of academia, or various consortia, that’s sort of the case.

As people put in their priorities, whether it’s sort of from governments or from various philanthropic organizations, we anticipate that that will ultimately translate into you know more deployment of single-cell on spatial and ultimately more revenue to us.

Operator (57:21)

There are no further questions at this time. We have reached the end of the Q&A session. This concludes today’s call. Thank you for attending. You may now disconnect.