> ## Content Index
> Fetch the complete content index at: https://nexi.fund/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI Biotech Platforms Compared: How Insitro, Recursion, Benchling Are Reshaping Drug Discovery
- URL: https://nexi.fund/ai-biotech-platforms-comparison/
- Published: 2026-07-28T13:00:55.000Z
- Updated: 2026-07-28T13:00:55.000Z
- Description: A field comparison of five AI drug discovery platforms — Isomorphic Labs, Recursion, Insitro, Insilico Medicine, and Schrödinger — their approaches, data, and odds of surviving to approval.
- Author: Nexi.fund Labs
- Tags: Biotech & Health, #mode-3, #hook-question, #track-D

Roughly $15 billion has flowed into AI-native drug discovery companies since 2020\. Yet only one AI-discovered molecule has posted positive Phase 2 data. That ratio , hundreds of companies, a single clinical win , is the number that matters.

🎯

**AI drug discovery is real, but most platforms will not survive to see a drug approved.**  
  
The field has bifurcated: three companies (Isomorphic Labs, Recursion, Insitro) have raised over $1 billion each and secured Big Pharma partnerships that validate their approach. Everyone else is fighting for the remaining attention — and the clock is running.  
  
The convergence of foundation models, automated labs, and multi-modality platform design is compressing discovery timelines from 12 years toward 4\. The question is not whether AI changes drug discovery. It is whose platform the industry will depend on when that happens. 

Two hundred startups is too many for a field that has produced exactly one positive Phase 2 readout. Insilico Medicine's rentosertib, for idiopathic pulmonary fibrosis, hit its endpoint in 2025 and was published in Nature Medicine. Every other AI-native platform is still proving it works in humans. The market has responded accordingly: capital is concentrating into a handful of companies with clinical data, pharma partnerships, and differentiated technology stacks.

Benchling surveyed 100 biotech organizations actively using AI for its 2026 Biotech AI Report. The findings: literature review has 76% AI adoption, protein structure prediction 71%, scientific reporting 66%. The breakthrough use cases share one trait , they work on clean, verifiable data that fits naturally into a scientist's daily workflow. Drug discovery remains the hardest integration. Only 58% of organizations use AI for target identification, and fewer than a third trust AI-designed molecules enough to advance them to preclinical testing without extensive wet-lab validation. As we wrote in July, the $347 million flowing into AI protein design platforms reflects investor belief that the tools will get there. The data says they are not there yet.

Five platforms dominate the conversation. Their approaches, data, and odds are worth comparing.

## Five platforms, five bets on how AI finds drugs

Isomorphic Labs raised $2.1 billion in May 2026 , one of the largest private biotech financings ever. DeepMind's spinout builds on AlphaFold 3, which solved protein structure prediction. The company has signed deals with Eli Lilly and Novartis worth a combined $3 billion in milestones. Its bet: structure-based design for undruggable targets is the highest-value application of AI in biology.

Recursion Pharmaceuticals takes the opposite approach. Instead of modeling proteins, it runs millions of cellular images through machine learning at its Phenomics platform. Five clinical programs have emerged from this pipeline, including a Phase 2 candidate for cerebral cavernous malformations. The company merged with Exscientia in a $688 million deal, adding chemistry capability to its imaging stack. Revenue: partnership milestones, no approved product.

Insitro, founded by Stanford ML pioneer Daphne Koller, has raised roughly $800 million to build TherML , a single AI engine that designs small molecules, antibodies, oligonucleotides, and complex biologics. TherML launched in January 2026 after the acquisition of CombinAbleAI. The company claims it is the only platform covering all four modalities. Partners include Eli Lilly and Bristol Myers Squibb. No wholly-owned candidate has reached clinical trials yet.

Insilico Medicine is the only company with published Phase 2 data. Its PandaOmics and Chemistry42 platforms identified both the target (TNIK) and the molecule (rentosertib) for IPF entirely through generative AI. The company went public on the Hong Kong Stock Exchange in December 2025 at a $2.7 billion market cap. The question is whether rentosertib was a one-off or the first of a repeatable pipeline.

Schrödinger is the most commercially mature. The 36-year-old company reported $58.6 million in Q1 2026 revenue, with all top-20 pharma companies as customers. Its FEP+ platform for free energy perturbation is the industry standard for lead optimization. In January 2026, the company launched Bunsen, an agentic AI co-scientist that integrates with its existing computational chemistry stack , moving beyond pure physics simulation toward AI-guided decision support. Schrödinger sells software, not drugs. Its bet is that every pharma company will need computational chemistry tools, and it is the default supplier.

| Parameter           | Isomorphic Labs | Recursion         | Insitro               | Insilico          | Schrödinger    |
| ------------------- | --------------- | ----------------- | --------------------- | ----------------- | -------------- |
| **Funding raised**  | $2.1B           | $1.3B+            | $800M                 | $400M + IPO       | $201M ACV      |
| **AI approach**     | AlphaFold 3     | Phenomics imaging | TherML multi-modality | Generative AI     | FEP+ physics   |
| **Clinical data**   | Preclinical     | 5 Phase 1/2       | Preclinical           | Phase 2a positive | N/A (software) |
| **Pharma partners** | Lilly, Novartis | Roche, Bayer      | Lilly, BMS            | Lilly, SK Bio     | Top 20 pharma  |
| **Revenue model**   | Milestones      | Milestones        | Milestones            | Pipeline + IPO    | Software ACV   |

Sources: company filings, press releases, Artificial Intelligence Companies 2026 comparison report

## Where each platform wins

Isomorphic Labs owns the hardest problem: undruggable proteins that have resisted small-molecule approaches for decades. AlphaFold 3 has changed what is structurally predictable. The company does not need to discover a drug to prove its model works , it needs to prove its predicted candidates survive Phase 1\. The $2.1 billion round suggests investors believe that happens.

Recursion wins on scale. Its Phenomics platform has generated 50 petabytes of cellular imaging data , a proprietary dataset that no competitor can replicate. The Exscientia merger added chemistry that Recursion lacked. The combined pipeline of five clinical programs is the deepest in the AI-native space. The risk is that phenotypic screening produces candidates faster but does not improve Phase 2 survival rates.

Insitro's TherML is the only platform designed for all four major drug modalities from the start. Most AI drug discovery companies focus on small molecules. It explicitly covers antibodies, oligonucleotides, and biologics. If multi-modality drug design becomes the standard, it owns the infrastructure layer. If the field stays specialized, TherML's breadth becomes a cost burden.

Insilico has the only data that matters in drug discovery: a positive clinical trial. Rentosertib's Phase 2a result , +98.4 mL FVC improvement versus -20.3 mL placebo , is statistically meaningful and was peer-reviewed. The question is repeatability. Insilico's pipeline includes candidates for cancer and longevity, and the HK IPO gave it capital to run those trials without diluting further. The market cap of $2.7 billion prices in at least one more success.

Schrödinger is not trying to discover drugs. It sells the picks and shovels. The company reported $201 million in software annual contract value in Q1 2026, up 124% year-over-year in drug discovery revenue. Its FEP+ platform is embedded in every major pharma's lead optimization workflow. The bet is that computational chemistry becomes as standard as NMR spectroscopy. The risk is that a platform like Isomorphic Labs or Insitro eventually replaces Schrödinger's tools with end-to-end AI.

## The money behind the platforms

The capital concentration tells a clear story. Isomorphic Labs raised $2.1 billion in a single round. Xaira Therapeutics launched with $1 billion. Recursion raised $688 million to acquire Exscientia. Insitro has $800 million cumulative. These four companies account for roughly $5 billion of the estimated $15 billion that has flowed into AI drug discovery since 2020 , one third of all capital going to four platforms.

The remaining 196 startups divide the other $10 billion. Most will not raise another round. The biotech funding environment in 2026 is selective: VC is concentrating into fewer, larger transactions, and IPO windows have narrowed. The Stifel biotech outlook reports that over 40% of Big Pharma revenue is at risk from patent expirations in the next six years , which drives dealmaking but only for late-stage assets. Early-stage AI platforms without clinical data or a pharma partnership face a hard fundraising landscape.

Partnership structures reveal the underlying economics. Isomorphic Labs and Insitro operate on a milestone-based model: pharma partners pay $30-50 million upfront and commit $1-2 billion in "biobucks" contingent on clinical success. The realistic value is 10-20% of headline figures, reflecting 50-60% Phase 2 attrition. Schrödinger collects software licensing fees of $500,000 to $5 million per site per year , smaller per deal, but recurring and independent of clinical outcomes.

## The limits no platform has solved

The central problem is that AI discovers targets and designs molecules, but drug discovery fails in humans for reasons that models cannot predict. Toxicity, bioavailability, and off-target effects emerge in Phase 2 and Phase 3 , stages where AI's predictive power drops sharply. Insilico's rentosertib success is encouraging. It does not mean the next AI-designed molecule will clear the same bar.

Data quality is the second limit. Every platform trains on different datasets, and those datasets have systematic biases. Recursion's 50 petabytes come from cellular assays , rich in phenotypic signal but blind to molecular mechanism. Isomorphic's models train on protein structures from structural biology databases, which are heavily skewed toward well-studied protein families. Insitro generates its own data from automated labs, which gives it control over quality but limits dataset scale to what its own facilities can produce. A model is only as good as the feedback loop that trains it, and no platform has closed that loop through a full clinical cycle yet. The hard truth: every AI drug discovery platform is making predictions about humans based on data from cells, proteins, and animals. The jump from there to a Phase 3 patient is where the industry's $15 billion bet will be tested.

Intellectual property is the third. Cloud AI platforms require submitting proprietary compound structures and protein targets to third-party data processors. If a structure is disclosed before patent filing, it compromises novelty. Several large pharma companies have raised this concern internally, and it has slowed adoption of cloud-based AI drug discovery tools in therapeutic areas where IP is the primary competitive moat. The largest pharma companies maintain internal computational chemistry groups specifically to avoid sending proprietary structures to external platforms , a structural headwind that no AI-native company has fully addressed.

Talent is the fourth. The hybrid scientist who can navigate both machine learning and biology is the scarcest resource in the industry. The Benchling report found that talent is the single most cited barrier to AI adoption in biotech R&D. Both have hired aggressively , roughly 250 and 900 employees respectively , but the broader market for ML-competent biologists is thin. Every platform competes for the same 2,000 people globally. That constraint caps how fast any single company can scale.

## The counterargument

Skepticism is warranted. The 200-startup count includes many that will never generate a clinical candidate. The single positive Phase 2 readout in five years of intensive AI-drug-discovery investment is not a good batting average. Platform companies have raised enormous sums based on the promise that AI reduces attrition, but that promise remains unproven at scale.

Even the winners face structural risk. Recursion has five clinical programs , a reasonable pipeline for a traditional biotech. But none is close to approval. Insitro has not advanced any wholly-owned asset to the clinic. Isomorphic Labs has no clinical data at all. Schrödinger's revenue is real, but it is a tools company, not a drug company. The market may eventually decide that AI drug discovery platforms should be valued as service providers, not as potential pharmaceutical giants.

The counterargument also applies at the technology level. Foundation models for biology , AlphaFold 3, the various protein-language models , have improved rapidly, but the gap between predicting a structure and predicting clinical outcomes is vast. Knowing what a protein looks like does not tell you whether a molecule that binds to it will be safe in humans over 10 years. That gap is where the industry's $15 billion will be tested.

The regulatory dimension adds another layer of uncertainty. The EU AI Act classifies AI systems used in drug discovery and clinical development as high-risk, requiring conformity assessments before deployment. No AI-native drug discovery platform has been through that process. The FDA has not issued formal guidance on AI-designed investigational drugs. Every platform is operating in a regulatory vacuum, and the first company to navigate it will have an advantage , but the timeline for that navigation is measured in years, not quarters.

📊

**Key signals to track**  
  
IND filing for a wholly-owned candidate from Insitro or Isomorphic Labs — signals platform confidence beyond partnerships  
  
Phase 2 readout from a second AI-native platform — converts Insilico's data point from anomaly to pattern  
  
A pharma partner exercises its option on a platform-designed molecule — real money validation vs. biobucks  
  
Schrödinger's software ACV growth rate — reveals whether pharma is buying tools or platforms 

## What an investor should watch

The AI drug discovery market is not a single bet. It is a portfolio of five distinct theses, each with a different risk profile and timeline. Isomorphic Labs is the highest-upside, longest-duration bet on structure-based design. Recursion is the most clinically advanced, with the deepest data moat. Insitro is the most architecturally ambitious , multi-modality or nothing. Insilico is the only one with clinical proof, but the repeatability question is open. Schrödinger is the safest revenue play but the lowest ceiling.

The field will not sustain 200 companies. It will not sustain 20\. The capital concentration already visible , $5 billion into four platforms , is the market voting on which theses are credible. An IND filing from Isomorphic or Insitro, or a Phase 2 success from a platform other than Insilico, would be the signal that the thesis is real. For an investor evaluating these companies, the partnership structure matters as much as the science. Milestone-based deals with upside caps of 10-20% of headline value do not justify a $7 billion valuation on their own. That requires a wholly-owned pipeline. Until one exists, the ratio of $15 billion invested to one clinical win is the most important number in the sector.

[ Best AI Drug Discovery Companies 2026 — AI Pharma Platforms Compared Comprehensive comparison of AI-native drug discovery platforms with valuation, clinical data, and partnership details across five leading companies. Artificial Intelligence Companies ](https://artificialintelligencecompanies.com/best/ai-drug-discovery-companies/?ref=nexi.fund) 

Primary source for platform valuations, clinical programs, and partnership terms — the benchmark comparison used throughout this article

[ Insitro to Acquire CombinAbleAI to Complete its Full Stack, Modality-Agnostic AI Platform for Drug Discovery and Design Official announcement of the TherML platform — Insitro's unified AI engine covering small molecules, antibodies, oligonucleotides, and biologics. Insitro ](https://www.insitro.com/news/combinabletherml/?ref=nexi.fund) 

Primary source for TherML architecture and multi-modality platform claims — the company's own platform announcement

[ 2026 Biotech AI Report Survey of 100 biotech organizations using AI — adoption rates, workflow integration patterns, and organizational shifts toward AI-native R&D operating models. Benchling ](https://www.benchling.com/biotech-ai-report-2026?ref=nexi.fund) 

Industry-wide data on AI adoption in biotech R&D — provides the macro context for the platform comparison