A Vienna-based biotech just raised €3 million to attack drug discovery for autoimmune diseases—not by building another chatbot wrapper, but by constructing proprietary biomedical knowledge graphs that most AI companies can’t replicate.
The Funding Details: What Graph Therapeutics Actually Secured
Graph Therapeutics closed its €3 million pre-seed round in January 2025, with Squareone and Merantix Capital leading the investment. This represents the company’s first institutional funding and positions it within a crowded but increasingly differentiated Q1 2025 cohort where over 150 AI startups collectively raised more than $5 billion.
The timing matters. While generalist AI funding showed signs of cooling after the 2023-2024 frenzy, vertical AI plays with defensible data moats continued attracting capital. Graph Therapeutics fits this pattern precisely: instead of competing on model architecture, they’re competing on domain-specific data assets that would take competitors years to assemble.
The target market provides context for the valuation conversation. Inflammation and immunology therapeutics represent a projected €257 billion opportunity by 2032. Pre-seed valuations in this space typically range from €10-15 million for teams with strong scientific credentials, suggesting Graph Therapeutics likely gave up 20-30% equity for this round—standard for European biotech AI plays at this stage.
Why Vertical AI in Drug Discovery Isn’t Just Another Hype Cycle
The AI drug discovery market has been littered with overpromised and underdelivered results. Recursion, Insilico Medicine, and Exscientia all raised hundreds of millions on similar premises. Yet the actual number of AI-discovered drugs in clinical trials remains embarrassingly small relative to the capital deployed.
Graph Therapeutics enters this skeptical landscape with a specific bet: inflammation and immunology diseases have enough shared biological pathways that a focused platform can achieve better hit rates than broader approaches.
This isn’t about building better foundation models—it’s about structuring the right data.
The inflammation and immunology space makes biological sense as a target. Autoimmune conditions like rheumatoid arthritis, psoriasis, and inflammatory bowel disease share mechanistic similarities despite different clinical presentations. A knowledge graph capturing these relationships could theoretically surface drug candidates that work across multiple indications, dramatically improving the economics of development.
Current investor trends in 2025 show a clear shift toward AI startups with proprietary data strategies rather than those building on top of commodity LLMs. Graph Therapeutics fits this thesis perfectly: their value accrues not from the model weights but from the structured biomedical knowledge they’re assembling.
The Technical Architecture: Knowledge Graphs Meet Drug Discovery
Understanding why Graph Therapeutics might succeed where others have struggled requires unpacking what “AI-powered drug discovery” actually means at a technical level.
Traditional computational drug discovery relies on three primary approaches: molecular docking simulations (computationally predicting how drug molecules fit into protein targets), QSAR modeling (quantitative structure-activity relationships that predict drug properties from chemical features), and phenotypic screening (testing compound libraries against cell or animal models).
Each approach has fundamental limitations. Docking simulations produce high false-positive rates. QSAR models struggle with novel chemical spaces. Phenotypic screens are expensive and often don’t translate to human efficacy.
Where Knowledge Graphs Change the Game
Knowledge graphs offer a different paradigm. Instead of asking “will this molecule bind to this target?” they ask “given everything we know about this disease biology, these patient populations, these failed and successful drugs, and these genetic associations, which targets and molecules are most likely to matter?”
The technical implementation typically involves:
- Entity extraction from scientific literature, clinical trial databases, patent filings, and proprietary experimental data
- Relationship modeling between genes, proteins, pathways, diseases, symptoms, and drug molecules
- Graph neural networks that learn representations capturing multi-hop relationships invisible to traditional methods
- Inference engines that traverse the graph to propose novel hypotheses about drug targets and mechanisms
For inflammation specifically, this matters because the field has accumulated decades of partially understood data. We know TNF-alpha inhibitors work for multiple autoimmune conditions. We know JAK inhibitors work differently but also across conditions. We know IL-17 and IL-23 pathways matter for psoriasis and psoriatic arthritis.
A well-structured knowledge graph can find the patterns human researchers miss because they can’t hold 50,000 papers in working memory simultaneously.
The competitive moat here isn’t the graph neural network architecture—those are increasingly commoditized. The moat is the curation, validation, and continuous updating of the knowledge graph itself. Every relationship edge requires biological validation. Every entity needs proper disambiguation. The difference between “IL-6 promotes inflammation” and “IL-6 promotes inflammation in specific tissue contexts under specific conditions” represents years of expert curation.
What Most Coverage Gets Wrong About AI Drug Discovery Valuations
Here’s the contrarian take most AI drug discovery coverage misses: the €3 million pre-seed round actually reflects market maturation, not market excitement.
In 2021-2022, European AI biotech pre-seeds routinely hit €5-10 million with less specificity than Graph Therapeutics is showing. The compressed round size in 2025 signals investors demanding more focused theses and clearer paths to value creation before writing larger checks.
This is healthy. The AI drug discovery space got ahead of itself, and the correction is forcing companies to make narrower, more defensible bets. Graph Therapeutics’ focus on inflammation and immunology—rather than “all of drug discovery”—represents exactly this kind of discipline.
The Real Competitive Landscape
The relevant competitors aren’t the well-funded giants like Recursion or Exscientia. Those companies raised during different market conditions and face different pressures around clinical validation of their platforms.
The actual competitive set includes:
- Causaly: UK-based, focused on biomedical knowledge graphs but without the therapeutics development focus
- Iktos: French generative chemistry company with AI design capabilities
- Peptone: Vienna-based (interesting geographic overlap) working on protein engineering
None of these competitors have Graph Therapeutics’ specific inflammation focus combined with knowledge graph methodology. Whether this specificity helps or hurts depends on execution—it narrows the initial market but could accelerate pharma partnership conversations with the inflammation-focused giants like AbbVie, Johnson & Johnson’s Janssen, or Bristol-Myers Squibb.
The European Biotech AI Ecosystem Advantage
Vienna as a headquarters location deserves attention. The European AI biotech ecosystem offers structural advantages that American coverage consistently undervalues.
January 2025 AI developments showed continued regulatory clarity in Europe around AI in healthcare applications, with the EU AI Act providing frameworks that, while sometimes burdensome, create predictability that pharma partners value.
European academic institutions also produce exceptional computational biology talent at lower salary expectations than Bay Area equivalents. Vienna specifically hosts strong programs at the University of Vienna, Medical University of Vienna, and the proximity to major pharmaceutical research centers in Basel (Roche, Novartis) and Munich (multiple biotech clusters).
The funding environment in Europe has matured significantly. Merantix Capital, one of the lead investors, operates as the investment arm of Merantix, a Berlin-based AI venture studio that has built and scaled multiple AI companies. They bring operational support, not just capital—a distinction that matters enormously at pre-seed stage.
European AI biotechs often achieve comparable milestones to American counterparts at 30-50% of the capital intensity.
This capital efficiency becomes competitive advantage when markets tighten. Graph Therapeutics can potentially reach proof-of-concept partnerships with pharma on €3 million that would require €10 million in the Bay Area simply due to labor cost differentials.
Practical Implications for Technical Leaders
If you’re a CTO, senior engineer, or technical founder watching this space, several actionable insights emerge from the Graph Therapeutics raise.
For Those Building AI-Native Products
The knowledge graph approach Graph Therapeutics appears to be taking offers a playbook applicable beyond drug discovery. Any domain with:
- Large corpus of semi-structured expert knowledge
- Complex multi-hop relationships between entities
- High cost of wrong predictions
- Regulatory or validation requirements that commodity LLMs can’t satisfy
…represents a potential vertical AI opportunity where knowledge graphs beat pure language models.
Consider financial compliance, materials science, legal research, or supply chain risk assessment. The technical architecture is similar: entity extraction, relationship modeling, graph-based inference, and human-in-the-loop validation for high-stakes decisions.
Architecture Considerations
If you’re evaluating knowledge graph approaches for your domain, the technical stack typically involves:
- Graph databases: Neo4j remains the dominant enterprise choice; Amazon Neptune for AWS-native stacks; emerging options like TigerGraph for scale
- Graph ML frameworks: PyTorch Geometric, Deep Graph Library (DGL), or specialized biomedical tools like GraphSAGE variants
- Entity extraction: Fine-tuned NER models on domain corpora, increasingly augmented with LLM-based extraction for complex relationships
- Embedding approaches: Knowledge graph embeddings (TransE, RotatE, CompGCN) that capture semantic relationships in vector space
The key technical decision is build versus buy for the underlying knowledge graph. In biomedical applications, open datasets like PubMed, UniProt, ChEMBL, and DisGeNET provide starting points, but competitive advantage comes from proprietary curation and relationship validation.
Partnership Opportunities
For those in adjacent spaces—clinical trial software, real-world evidence platforms, electronic health records infrastructure—AI drug discovery companies like Graph Therapeutics represent potential integration partners.
The inflammation focus creates specific opportunities. Any platform with data on patient journeys through autoimmune disease diagnosis and treatment becomes valuable for validating computational predictions. If your system touches rheumatology, dermatology, or gastroenterology patient data, you’re sitting on assets these companies need.
Forward Look: What Happens in 12-18 Months
Graph Therapeutics’ trajectory will follow predictable biotech AI milestones if execution goes well.
Months 1-6: Team building. A €3 million pre-seed typically supports 10-15 people for 18-24 months. Expect hiring focused on computational biologists, ML engineers with graph expertise, and potentially a chief scientific officer with pharma industry credibility if they don’t have one already.
Months 6-12: Platform validation. The company needs to demonstrate that their knowledge graph approach can identify targets or compounds that conventional methods missed. This usually means retrospective validation—showing they would have predicted known drugs or identified known targets—plus at least one prospective prediction that generates novel experimental data.
Months 12-18: Partnership or Series A. The standard path is either a pharma partnership that provides both validation and capital, or a Series A that extends runway for independent development. Given the inflammation focus, likely partners include AbbVie (dominant in TNF-alpha inhibitors), Eli Lilly (active in immunology), or mid-sized specialty players like UCB or Galapagos.
The €257 billion market projection by 2032 provides context but also a trap. That figure includes blockbuster drugs already on market and their biosimilars. The addressable market for new AI-discovered therapeutics is a fraction of that—likely €5-10 billion in potential royalties and milestone payments over the next decade for all AI drug discovery companies combined.
Graph Therapeutics doesn’t need to capture a huge market share to deliver strong returns on a €3 million pre-seed—a single successful pharma partnership in the €20-50 million range would represent massive validation.
The Bigger Picture: Why This Deal Signals Where AI Investment Is Heading
The Graph Therapeutics raise crystallizes broader patterns in 2025 AI investment that technical leaders should internalize.
First, vertical beats horizontal. The companies attracting smart money aren’t building general-purpose AI tools—they’re building AI systems with proprietary data moats in specific domains. The generalist opportunity has been captured by OpenAI, Anthropic, Google, and Meta. Everyone else needs domain specificity to compete.
Second, Europe matters more than American VCs admit. The combination of regulatory clarity, academic talent, capital efficiency, and proximity to major pharmaceutical headquarters makes Europe genuinely competitive for AI biotech. Merantix Capital’s involvement signals European AI infrastructure maturing beyond just providing talent to American companies.
Third, the bar for AI claims has risen. “We use AI” no longer opens checkbooks. Investors now ask: “What data do you have that others can’t easily replicate? What’s your specific claim about where AI outperforms existing methods? What’s your validation pathway?” Graph Therapeutics apparently answered these questions satisfactorily.
Fourth, pre-seed rounds are getting more disciplined. The €3 million size reflects investors wanting to see milestone achievement before committing larger sums. This is healthy for the ecosystem even if it makes early-stage fundraising harder. Companies that can execute efficiently will stand out.
What This Means for Your Organization
If your organization touches healthcare, life sciences, or any domain with complex expert knowledge, the Graph Therapeutics model offers strategic considerations.
For pharmaceutical companies: AI biotech partnerships will become table stakes over the next 3-5 years. Building internal evaluation capabilities to assess these partnerships—distinguishing genuine technical advantages from well-marketed vaporware—is urgent. Consider what data assets you’re sitting on that could feed into knowledge graph approaches.
For healthcare technology companies: The boundaries between “healthcare IT” and “healthcare AI” are dissolving. Any dataset with structured medical information becomes potential training data for drug discovery or clinical decision support. Understand the regulatory and partnership implications of your data assets.
For AI teams in other verticals: The knowledge graph architecture pioneered in drug discovery applies to your domain. If you’re doing anything with complex domain knowledge, semi-structured data, and high-stakes decisions, study how biomedical AI companies structure their technical approaches. The playbook transfers.
For founders considering the space: The combination of vertical focus, proprietary data strategy, and European capital efficiency that Graph Therapeutics exemplifies represents a viable 2025 model. The days of “we’ll use AI to do X better” securing funding are over. You need a specific data moat thesis and a credible path to validation.
The inflammation and immunology focus matters less than the underlying strategy: pick a specific domain, build knowledge infrastructure competitors can’t easily replicate, and execute efficiently enough to reach validation before capital runs out.
Graph Therapeutics’ €3 million bet isn’t about whether AI can help discover drugs—it’s a test of whether focused, capital-efficient approaches can compete against the heavily-funded platform plays that have thus far underwhelmed.
