A molecule designed by AI can begin life on a screen. To become a drug, it must still undergo years of experiments, clinical trials and negotiations. This journey is where the economic battle plays out: who gets paid early, who funds the failures, and who retains a share of future sales? Behind the headline-grabbing partnership announcements, the industry is arranging less a division of treasure than an allocation of uncertainty.
Looking ahead to September 2026, this question is shaping competition among AI specialists, biotechs and pharmaceutical groups. The examples below draw on public agreements and events from 2021 to 2024; the developments envisaged for 2026 are forward-looking analysis, not established outcomes.
The billion-dollar headline is not the check cashed
In January 2024, Isomorphic Labs, an Alphabet company that emerged from the DeepMind ecosystem, announced collaborations with Eli Lilly and Novartis. Their combined potential value runs into billions of dollars. Yet upfront payments account for only a fraction of that total. The rest depends on research programs, development progress and commercial successes that may never materialize.
This structure is far from unusual. The collaboration announced in 2021 between Recursion and Roche, alongside Genentech, and the agreement reached in 2022 between Exscientia and Sanofi illustrate this approach. The pharmaceutical company buys access to a platform and programs without immediately paying as though a drug had already been validated.
An announced maximum deal value is therefore neither guaranteed revenue nor a measure of scientific worth. It sometimes combines several projects and payments spread over a long period. To interpret an announcement correctly, look at how much is paid upfront, which commitments are binding and how many programs must succeed to reach the ceiling.
Three tiers for sharing risk
The upfront payment: funding the present
The upfront payment generally compensates for access to a technology, rights to certain assets or the launch of a collaboration. For a biotech, it provides cash without necessarily issuing new shares. It can also signal credibility to investors, although it does not constitute evidence of therapeutic efficacy.
But the check comes at a price. Overly broad exclusivity can prevent the young company from working with other partners in a promising field. Rights granted over multiple targets can also tie up some of its potential. Funding that is welcome today can therefore prove costly tomorrow.
Milestones: paying as uncertainty recedes
Milestone payments, or milestones, are triggered by events defined in the contract: selecting a candidate, starting a trial, securing approval or crossing a sales threshold. Each milestone turns a promise into an asset backed by stronger evidence. The pharmaceutical company limits its initial exposure; the biotech retains a financial stake in the program’s progress.
Much depends, however, on the wording. Who decides whether to continue the trial? What happens to the program if the partner changes priorities? Are there development commitments or rights to reclaim the program? A candidate may be abandoned for strategic reasons without having been shown to be ineffective. For the biotech, this distinction matters scientifically, but it does not replenish its cash reserves.
Royalties: retaining a share of the future
Royalties provide a share of sales under terms that are often disclosed in little detail. They can become highly lucrative, but only after development and commercialization. Their value depends on the territories covered, their duration, permitted deductions and the competitive landscape. An attractive rate on a product that never reaches the market generates nothing.
AI accelerates certain tasks, not all of medicine
Identifying a target, predicting a structure, proposing a molecule or analyzing cellular images: AI plays a role at various points in research. AlphaFold marked a major advance in protein structure prediction, extended in 2024 by AlphaFold 3 to the modeling of molecular complexes. This does not mean that a predicted interaction will produce a safe treatment in humans.
The transition to the laboratory remains crucial. Among other requirements, a molecule must demonstrate relevant activity, reach the right tissue and avoid unacceptable toxicity. Then come patients, with their biological diversity and concomitant treatments. The entry into Phase II, announced in 2023, of an Insilico Medicine candidate for idiopathic pulmonary fibrosis marked an important step, not definitive validation of the business model.
The true measure of performance is therefore not simply the number of molecules proposed or the time saved on a task. It also includes candidate quality, probability of success and the total cost of producing a drug. Accelerating the production of mediocre candidates may simply shift spending toward trials.
Power belongs to those who can wait
In negotiations, differences in cash reserves carry considerable weight. A major pharmaceutical company can fund several bets simultaneously. A biotech facing a funding deadline may accept a larger upfront payment in exchange for fewer future rights. The division of value then reflects financial bargaining power as much as the quality of the algorithm.
The pharmaceutical partner also brings expertise that is difficult to replace: clinical development, regulatory knowledge, manufacturing and market access. Conversely, a young company strengthens its position if it has proprietary experimental data, a platform validated across several projects or an asset advanced enough to attract multiple offers.
Data and patents: the hidden value in the contract
The product is not always the only important asset. A collaboration generates experimental results, including failures, that may improve subsequent models. Who can reuse these data? Does the AI specialist retain the right to train its platform on them? Does exclusivity cover a molecule, a target or an entire therapeutic area? These clauses can determine its future competitiveness.
Patents also remain central, but protection for a molecule, a use and a method must be distinguished. AI’s contribution removes neither patentability requirements nor the need to properly establish human inventorship under the relevant jurisdiction’s rules. A discovery that is difficult to protect may have medical value without delivering the returns investors expect.
What next? For September 2026 and beyond, the scenario to watch would be more stringent selection of platforms: less compensation for algorithmic promise alone, and more for reproducible experimental results and differentiated assets. Biotechs able to fund their independence could retain more rights; others may prioritize immediate cash. The winner will not necessarily be whoever designs a molecule fastest, but whoever turns that head start into clinical evidence while retaining a negotiated share of its value.