What Makes a Health Company Truly AI-Native?

Insitro & Lilly TuneLab: Data Moats in Multi-Modal AI

Listen to this article · 8 min listen

When Insitro’s machine learning models for predicting small molecule in vivo pharmacokinetic behavior showed up in Eli Lilly’s model access platform, Lilly TuneLab, it was a pretty big deal for anyone tracking AI in drug discovery. The collaboration was announced back in September 2025, but the models actually going live on October 6, 2026, forces a fascinating question. This move shows how a model distribution deal can completely shift the data-moat advantage between partners, especially when one company’s entire value is built on a complex data fusion strategy.

Insitro’s Models in Lilly TuneLab: A Distribution Play

Insitro, a company we track for its Multi-Modal Data Fusion strategy, just made its purpose-built predictive models available inside Lilly TuneLab. Practically speaking, Eli Lilly’s own researchers and their partners can now point Insitro’s AI at their small molecules to predict how they’ll behave in a living system, a huge part of de-risking drug development. The announcement states these in vivo pharmacokinetic prediction models are live on the platform. For a company like Insitro, which was born from the idea of fusing genetic, cellular, and clinical data to build its predictive engine, dropping its models into a pharma giant’s internal platform is a major move. It’s a firehose for getting its AI used by the people who matter in drug discovery. But the implications are way bigger than just market reach. The deal immediately raises hard questions about who really owns the models, who owns the data coming off them, and how an AI-native company stays special.

Multi-Modal Data Fusion: The Core of Insitro’s Data Moat

We’ve been classifying Insitro’s data strategy as Multi-Modal Data Fusion for a while now. The whole approach is about building predictive models by fusing together wildly different datasets, genetics, cellular assays, clinical observations, you name it. This fusion is an incredibly difficult technical problem that demands serious infrastructure, deep expertise in data integration, and a ton of computational horsepower to find patterns. These models now sitting in Lilly TuneLab are the direct output of that entire data strategy. For AI-native companies, the holy grail is a strong data moat, a competitive edge from proprietary data that’s hard to copy and makes your models better. Insitro’s moat is theoretically dug deep by its unique and proprietary way of combining all that multi-modal data. The models on Lilly TuneLab are basically missionaries for this fused-data strategy, sent to demonstrate its predictive punch.

Where Does the Advantage Sit: Platform vs. Creator?

So the big question this partnership raises is who in the end holds the strategic high ground when models born from a fused-data strategy get used inside someone else’s walled garden. You can argue it two ways, and both are compelling.

  • Compounding the Data Moat: One argument is that getting its models inside Lilly TuneLab actually strengthens Insitro’s data moat. Every single time a model runs, it’s an advertisement for the power of the underlying fused-data strategy. If Eli Lilly’s teams see consistent success with these models, it validates Insitro’s entire approach, which could bring in more partnerships and cement their reputation as the go-to for AI in drug discovery. You could see a positive feedback loop where this platform exposure just drives more demand for Insitro’s core AI.
  • Leaking Advantage to the Platform: On the other hand, the advantage could easily leak away from Insitro and pool inside Eli Lilly. The customer relationship is Lilly’s. More importantly, all the usage data, what molecules are being tested, what parameters are being used, which predictions are most valuable, is generated on Lilly’s turf. This would give Eli Lilly a treasure trove of insight into how Insitro’s models actually perform in the wild, which inputs matter, and where they should point their own internal AI efforts next. That proprietary usage data, generated by Insitro’s models but captured by Lilly’s platform, could be used to build up Lilly’s own AI and erode Insitro’s unique edge over the long haul. That’s the nightmare scenario for any AI company that doesn’t own the rails. What we don’t know, because it wasn’t in the press release, are the commercial terms, model ownership details, and data rights. For anyone reading this who works in AI-native healthcare, those are the only things that matter. Without knowing the fine print, this distribution “win” is just a headline, not a proven long-term advantage. Analysis of AI model ownership in pharmaceutical partnerships

    Ownership and Terms: The Unanswered Questions for AI-Native Platforms

    The real strategic upshot of putting Insitro’s models on Lilly TuneLab comes down to the confidential agreement between them. Nothing in the public record mentions the financial arrangement with Eli Lilly, any performance benchmarks for the models, or who owns the data they’re built on. These aren’t just details. They’re everything when you’re trying to figure out if Insitro’s data moat is getting deeper or shallower.

  • Model Ownership: Does Insitro keep full ownership of its models, especially any improvements made based on how they perform inside Lilly TuneLab? Or does Eli Lilly get some rights to the models or their derivatives? This is a huge deal, especially when you consider algorithmic drift, where a model’s performance gets worse over time as new real-world data no longer matches the old training data. Who’s responsible for fixing that? Explanation of algorithmic drift in AI models
  • Data Ownership and Usage: Who owns the new data that’s created every time a scientist at Lilly runs one of these models? Is that usage data fed back to Insitro? If so, on what terms? Getting access to that real-world evidence (RWE) would be priceless for Insitro to keep refining its models and strengthening its whole Multi-Modal Data Fusion strategy. But if Eli Lilly gets to keep all that RWE for itself, it could rapidly build up its own internal AI muscle. In any deal like this, data governance and privacy rules like HIPAA and the NIST frameworks are obviously front and center, especially with sensitive biological data. NIST guidance on AI data governance
  • Clinical Guardrails and Efficacy Evidence: The announcement talks about model accessibility, but any serious AI health company knows the game is really about operating within clear clinical guardrails and having published proof that your stuff works. This partnership is still in the early discovery phase, but the precedents they set right now for data sharing and model validation are going to be incredibly important as these tools get closer to actual clinical use. So until we get clarity on these ownership and usage terms, putting Insitro’s models on Lilly TuneLab is a win for market reach, but it’s not necessarily a win for Insitro’s long-term Multi-Modal Data Fusion strategy. This relationship with the Eli Lilly platform is the exact place where a fused-data moat either gets wider and deeper, or it starts to spring a leak. For anyone working in drug discovery and healthcare AI, watching how these platform partnerships are structured is the only way to understand who is actually going to win in the end.

Frequently Asked Questions

What is the significance of Insitro’s models being available on Lilly TuneLab?

This development is a critical juncture for understanding how model distribution decisions can shift data-moat advantage between partners. It represents a direct channel for Insitro’s AI capabilities to be consumed by a broader audience within the drug discovery ecosystem, impacting questions of model and data ownership.

What is Insitro’s core data strategy and how does it relate to their data moat?

Insitro’s core data strategy is Multi-Modal Data Fusion, which involves building predictive models by fusing diverse datasets like genetic, cellular, and clinical data. This unique combination and integration of multi-modal data ostensibly deepens their data moat, a competitive advantage derived from proprietary datasets that enhance AI model performance.

How might distributing Insitro’s models through Lilly TuneLab affect Insitro’s data moat?

One perspective suggests it could compound Insitro’s data moat by validating their approach and attracting more partners. Conversely, the advantage could leak to Eli Lilly, as the platform provider gains valuable usage data and insights into model performance, potentially informing their own AI strategy and diminishing Insitro’s unique advantage.

What key information is missing to fully assess the strategic implications of this partnership?

The public announcement lacks details regarding the terms of the arrangement, specifically model ownership, data ownership, and financial terms. Without clarity on these aspects, it is difficult to definitively evaluate whether this distribution is a long-term strategic advantage for Insitro or if the advantage shifts to Eli Lilly.

Share
Was this article helpful?

Editorial Team

The Editorial Team at AI-Native Health Companies is a group of writers and editors who cover health, workplace wellness, and everyday care. They research each story and publish under a shared desk byline — experts in their fields, working as one team.