How to commercialise your Frontier Software

September 2025

You’ve found the new Alphafold, good for you, but how are you going to sell it?

Consulting or anything ressembling providing it as a pure service (digital CRO) is tempting but will be hard to scale. Saas is the natural choice but in practice rarely works (Schrodinger is the exception confirming the rule). This is because when it comes to frontier tech like computational and quantum chemistry, bioinformatics, AI-for-science, most of the traditional elements of the SaaS model are not present:

• Small number of customers: only a few tens of relevant labs and companies worldwide.
  
• Configurability Requirement: researchers want to use their own stack, their own codes, or open-source tools.
  
• Procurement pain: pharma, energy, and chemical companies have long and bureaucratic buying cycles.
  
• Open-source competition: the market already has VASP, Gaussian, CP2K, Psi4 — subsidized and entrenched.

There are a handful of commercialization archetypes that frontier software and AI-for-science companies have used successfully:

Run your model for someone else:

1. SaaS (Schrodinger) or wrapped your product in a Platform SaaS (Cradle.bio)
2. Data as a Service (Scale.ai)
3. Open Source your model (Hugging Face)

Run your model for yourself:

4. Co-own the ouptut: Patnership + Milestone (Biotech/Entalpic)
5. Own the output: and sell/licence it (Biotech)
6. Own the output: and commercialise it yourself (Genentech/Orbitals)

More details for each below:

1. Vertical SaaS / Platform as a Service

Sell access to the model directly or wrap it into end-to-end applications for specific verticals. You can also sell the platform that abstracts away compute and workflows, that would extend your product through the value chain horizontally. E.g., a protein design platform.

• Revenue: monthly subscription / per-seat licenses OR usage-based billing (compute cycles, simulations, datasets processed)
• Focus: solve a business problem (reduce R&D cycles, hit regulatory milestones).
• Example: Schrödinger’s LiveDesign → not just models, but a collaborative design environment.

2. Data-as-a-Service (DaaS)

Use your models to generate proprietary (training) datasets (screened materials, candidate molecules, reaction barriers)

• Advantage: scalable, defensible if the data is unique. Re-selling potential
• Example: Scale.AI
• Risk: some customers will prefer raw models to curated data.

3. Open Source + Enterprise Layer

Open-source your core models/workflows to drive adoption.
Monetize the enterprise features: team collaboration, deployment, compliance, cloud orchestration.

• Advantage: credibility, faster adoption, community effects.
  
• Challenge: building stickiness beyond what’s free. Encourage competition.
• Example: HuggingFace

4. Partnership + Milestone Model (Biotech-Style)

Collaborate with large players who fund R&D in exchange for milestones and royalties.

• Advantage: high-value deals, credibility, aligned incentives.
  
• Example: AI drug discovery startups like Insilico
  
• Downside: not pure SaaS, but still very fundable if you land a few strong partnerships.

5 and 6. Internal Productization (Build Your Own IP)

Instead of selling the model, use it internally to discover molecules/materials or anything of commercial value. You could license/sell IP to existing players in your industry. Or become a tech-enabled biotech/materials company and bring products to market yourself.

• Advantage: biggest upside (multi-billion outcomes possible).
  
• Trade-off: capital intensive, longer horizon, but often more attractive to venture investors

Choosing a Path

The right model depends on your company’s risk profile, fundraising and ambition timeline:

• Fast SaaS-like scaling → Platform-as-a-Service + Workflow SaaS. But this is hard to pull off and few successfull examples.
• Biotech-style upside → Internal IP generation. High risk high reward.
• Community credibility → Open-core + Enterprise. Probably depending on the culture in the space you are developping. For LLM, it’s hard to image a frontier AI labs not having an open tier.
• Capital-efficient entry → Services → Product. You could probably bootstrap it but careful not to fall in the trap of capped returns.