What is it
Milestone-based pricing is a billing unit used in drug discovery and biotech AI where payment is tied to achieving defined research milestones rather than time or compute consumed.
The structure borrows from the pharmaceutical industry’s long-established R&D collaboration model and applies it to AI-powered platforms. Instead of charging per seat or per API call, a vendor receives an upfront payment at deal signing, then earns a series of predetermined payments as the engagement clears verifiable scientific gates: a target validated, a candidate molecule identified, a preclinical study completed, an IND filing submitted, a Phase I trial initiated, or a drug approved and on the market. Isomorphic Labs — the DeepMind spinout led by Sir Demis Hassabis — uses exactly this structure in its partnerships with Eli Lilly ($45M upfront, up to $1.7B in milestones) and Novartis ($37.5M upfront, up to $1.2B in milestones), both announced in January 2024.
What distinguishes milestone billing from other outcome-based approaches is the explicitness of the gate. Each milestone is written into the contract as a binary event — it either happened or it didn’t — with a specific dollar amount attached. This removes the ambiguity that plagues looser outcome models and gives both parties a shared, auditable record of when payment is owed. Recursion, which operates one of the largest AI drug-discovery platforms (models trained on more than 50 petabytes of proprietary biological data and a 504-GPU BioHive-2 supercomputer), structures its Roche/Genentech partnership around exactly this logic: the $150M upfront unlocked access to the platform, while each program carries more than $300M of individual milestone potential payable only as science progresses. For a primer on how these value metrics compare to seats and tokens, see the guide to usage-based pricing models.
Because the largest payments arrive only after years of clinical and regulatory work, milestone pricing is radically back-loaded. The headline “biobucks” figures that populate press releases — Recursion’s Sanofi deal at up to $5.2 billion in aggregate milestones across 15 programs, Insilico Medicine’s US$4.6 billion in cumulative collaboration value — are ceilings reachable only if many gates clear across a decade-long arc. Upfront and near-term payments are far smaller, and most of a deal’s potential value is never realized because most drug candidates fail. The conditionality is deliberate: the vendor shares the risk rather than collecting regardless of results, which is what makes large pharma willing to sign these deals at all.
How it works
A milestone-based deal has three structural layers that stack over the life of a drug program.
Layer 1: Upfront payment. At signing, the pharma partner pays a cash sum that compensates the AI platform for its prior investment and secures access to the platform for the collaboration. Upfronts in the corpus range from $20M (Exscientia’s Merck KGaA deal, and Recursion’s own Merck KGaA deal) to $150M (Recursion’s Roche/Genentech deal). The upfront is earned regardless of downstream outcomes — it is the one non-contingent element of the deal.
Layer 2: Milestone payments. The contract defines a ladder of discrete research, development, regulatory, and commercial events, each with an associated payment. The milestone types and their typical relative sizes appear in the table below.
| Milestone category | When triggered | Typical share of deal ceiling |
|---|---|---|
| Research / hit identification | AI-designed molecule meets target engagement criteria | Small (5–15% of total) |
| Preclinical / IND-enabling | Candidate clears in-vivo studies; IND filed | Moderate (10–20%) |
| Clinical entry (Phase I) | First patient dosed in a Phase I trial | Moderate (10–20%) |
| Clinical advancement (Phase II / III) | Positive Phase II or Phase III primary endpoint | Large (25–40%) |
| Regulatory approval | NDA/MAA submitted or approved | Large (20–30%) |
| Commercial sales milestones | Drug reaches defined annual net-sales thresholds | Variable (cumulative, often largest) |
Layer 3: Tiered royalties. Once a drug is commercialized, the AI vendor receives a percentage of net sales that escalates as revenue crosses defined thresholds. Royalty rates in the corpus run from mid-single-digit percentages to the mid-teens — Exscientia’s Sanofi agreement quoted high-single-digits to mid-teens, with a co-investment option that would have elevated royalties to 21%. Choosing the right billing gate is itself a design decision; the guide to choosing the right usage metric covers the trade-offs.
Unit math example: A three-program collaboration with a $30M upfront, five milestones per program totalling $120M per program, and a 10% royalty on $200M net peak sales would yield: $30M (upfront) + up to $360M (milestones across three programs) + ~$20M/yr (royalties at peak). Total potential: ~$390M before royalties, with royalties compounding annually until patent expiry or drug withdrawal.
The payment trigger is always a binary event verified externally — a regulatory filing date, a trial initiation confirmed in a clinical-trials registry, a published Phase II readout. This makes milestone accounting more straightforward than many outcome-based models: there is no disputed measurement of what “resolved” or “completed” means, only whether a dated, public event occurred.
Companies using this
Four companies in the corpus use milestones as a primary billing unit, all operating in AI-powered drug discovery. Isomorphic Labs and Recursion are the largest by deal volume; Exscientia (now absorbed into Recursion after an ~$688M all-stock merger completed in November 2024) and Insilico Medicine complete the group.
| Company | Product | Pricing model | Billing units | Free tier | Verified |
|---|---|---|---|---|---|
| Exscientia (now part of Recursion) | AI-driven drug discovery & design platform | No | 2026-06-16 | ||
| Figure | General-purpose humanoid robots (Figure 03) & Helix AI | No | 2026-06-14 | ||
| Insilico Medicine | Pharma.AI generative drug-discovery platform + clinical pipeline | Yes | 2026-06-14 | ||
| Isomorphic Labs | AI-first drug discovery & design (Isomorphic Drug Design Engine) | No | 2026-06-14 | ||
| Recursion | AI-enabled drug discovery platform (Recursion OS) — pharma partnerships, internal pipeline & NVIDIA-powered compute | No | 2026-06-10 |
Explore this theme in the knowledge graph
FAQ
What is milestone-based pricing in AI drug discovery?
Milestone-based pricing is a billing structure where payment is triggered by defined research achievements — such as a drug candidate identified, a clinical trial initiated, or a regulatory filing submitted — rather than by time elapsed or compute consumed. Companies like Isomorphic Labs and Recursion use this structure for pharma partnerships, with an upfront payment at signing followed by a ladder of payments as each milestone is cleared.
How are milestone payments different from subscriptions or usage fees?
A subscription charges a fixed recurring fee regardless of outcomes; a usage fee charges for compute or API calls consumed. A milestone payment charges nothing until a specific, pre-defined result is verified — and then releases a lump sum. This makes milestone-based pricing heavily back-loaded and contingent on scientific or commercial success, which aligns the vendor's revenue with the buyer's real goal rather than with the volume of software used.
How large are milestone deals for AI drug-discovery companies?
The headline 'biobucks' totals are ceilings, not invoices. Recursion's Sanofi collaboration is worth up to $5.2 billion in aggregate milestones across 15 programs on top of a $100M upfront, while Isomorphic Labs' Eli Lilly deal carries up to $1.7B in milestones on a $45M upfront. Most of that value is contingent and is rarely realized, because most drug candidates fail before the largest gates are cleared.
Who uses milestone-based pricing?
In this corpus, the four companies that bill on milestones all operate in AI-powered drug discovery: Isomorphic Labs, Recursion, Exscientia (now part of Recursion), and Insilico Medicine. Each earns a modest upfront payment and then a ladder of milestone payments plus royalties as partnered drug programs advance through research, clinical, and regulatory stages.
Related billing units
- Credit-Based BillingA billing unit where customers pre-purchase or are allocated a pool of credits that deplete as they use the product, often at variable rates per feature.
- Token-Based PricingA billing unit common in LLM and AI products, where customers are charged per input and output token processed.
- Per-Seat PricingA billing unit where the vendor charges a fixed fee per named user, regardless of how much each user consumes.
- Per-Resolution PricingA billing unit unique to AI customer-support products, where the vendor charges only when an AI agent resolves a customer issue without escalation.
- Bandwidth-Based PricingA billing unit where customers are charged per gigabyte of data transferred out of the platform.
- Per-Function-Invocation PricingA billing unit where customers are charged per serverless function invocation, often combined with a separate compute-time charge.
- CPU-Hour PricingA billing unit where customers are charged for the CPU time their workloads consume, typically measured in vCPU-seconds or vCPU-hours.
- GB-Hour PricingA billing unit where customers are charged for the memory their workloads consume over time, measured in gigabyte-hours.
- GPU-Hour PricingA billing unit where customers are charged for GPU time consumed, typically measured per-second or per-hour by GPU type.
- Per-API-Call PricingA billing unit where customers are charged per API request, regardless of payload size or processing time.
- Per-GB Storage PricingA billing unit where customers are charged per gigabyte of data stored on the platform per month.
- Media-Minute PricingA billing unit where customers are charged per minute of audio or video processed — used by speech, voice, and video AI vendors.
- Per-Request PricingA billing unit where customers are charged per request served — the generic meter for inference endpoints, search, scraping, and browser infrastructure.
- Per-Event PricingA billing unit where customers are charged per event ingested — the native meter of observability and billing-infrastructure platforms.
- Vector Storage PricingA billing unit where customers are charged for vectors stored or indexed — the storage dimension of vector database pricing.
- Per-Character PricingA billing unit where customers are charged per character of text processed — the standard meter for text-to-speech and translation.
- Per-Document PricingA billing unit where customers are charged per document processed or generated — common in AI writing, SEO, and document-intelligence tools.
- Per-Page PricingA billing unit where customers are charged per page crawled, parsed, or rendered — the meter for web scraping and document parsing.
- Per-Transaction PricingA billing unit where customers are charged per financial or billing transaction processed — the meter of billing and accounting platforms.
- Active-User PricingA billing unit where customers are charged per monthly or daily active user rather than per provisioned seat.
- Per-Task PricingA billing unit where customers are charged per task an automation or agent executes — Zapier's historical unit, now spreading to AI agents.
- Per-Unit PricingA billing unit used by robotics, hardware AI, and some SaaS companies where the metered object is a physical or abstract 'unit' — a robot deployed, a device sold, or a defined deliverable.
- Workflow Execution PricingA billing unit where each end-to-end workflow or automation run is metered and billed, regardless of the compute steps it contains.
- Per-Message PricingA billing unit where each individual message or reply in a conversation is metered, common in AI chat and voice platforms.
- Per-Invoice PricingA billing unit used by billing infrastructure platforms where each invoice generated or processed is metered as the primary cost driver.
- Per-Action PricingA billing unit where each discrete action taken by an AI agent or automation is metered — common in browser automation and agentic workflow tools.
- Per-Image PricingA billing unit where each AI-generated image is metered, common in image generation APIs and multimodal AI platforms.
- Per-Conversation PricingA billing unit where each complete customer conversation — from first message to resolution — is metered as a single chargeable event.
- Per-Record PricingA billing unit where each data record processed, labeled, or extracted is metered — common in data platforms and web scraping services.
- Per-Word PricingA billing unit common in translation and localization platforms where the metered object is the word count of content processed.
- Per-Video PricingA billing unit where each AI-generated video is metered, common in video generation and synthetic media platforms.
- Per-Outcome PricingA billing unit where payment is triggered by verified outcomes delivered — distinct from outcome-based pricing models, this refers specifically to 'outcomes' as a countable billing unit.
- Per-Datapoint PricingA billing unit where each individual data measurement or signal ingested is metered — common in cloud cost intelligence and ML evaluation platforms.
- Per-Interaction PricingA billing unit where each patient-agent or user-agent interaction is metered, common in healthcare AI and customer engagement platforms.
- Data Licensing PricingA pricing structure where access to proprietary datasets or data assets is licensed separately from the software or services, common in AI training data and clinical data platforms.
- Robot-Hour PricingA billing unit where each hour a robot or autonomous system operates is metered — the robotics equivalent of a GPU-hour.
- Per-Contact PricingA billing unit where each contact or lead in the database is metered, common in AI sales development and outbound automation platforms.
- Per-Mailbox PricingA billing unit where each connected email mailbox or sending account is metered, common in AI outbound sales and email automation platforms.
- Browser-Hour PricingA billing unit where each hour of headless browser compute time is metered, common in web scraping and browser automation platforms.
- Per-Generation PricingA billing unit where each AI-generated creative asset — image, video, or design — is counted as a 'generation' and metered accordingly.
- Per-Ticket PricingA billing unit where each customer support ticket handled by an AI agent is metered — common in AI customer service platforms.
- Per-Log PricingA billing unit where each LLM request log ingested or stored is metered — common in AI observability and evaluation platforms.
- Per-Trace PricingA billing unit where each distributed trace — a complete record of an LLM request chain — is metered, common in AI observability platforms.
- Per-IP PricingA billing unit where each IP address or proxy endpoint allocated is metered — used by web scraping proxy providers.
- Per-Device PricingA billing unit where each hardware device or endpoint connected to the AI platform is metered.
- Per-Case PricingA billing unit used in legal AI platforms where each case or matter processed by the AI is metered.
- Per-Report PricingA billing unit where each AI-generated report or analysis document is metered as a discrete output.