Investment in AI-driven drug discovery continued to grow in the first half of 2026, with a steady stream of platform deals targeting small-molecule discovery and wider pharmaceutical research. Yet faster in silico identification is surfacing a deeper obstacle: many candidates that impress in early computation still fail when formulators and manufacturers try to turn them into viable products.
The real bottleneck in AI drug discovery
Faster discovery and more computational power do not make molecules easier to develop. As AI expands chemical space and surfaces more complex small-molecule candidates, sponsors and early-phase development partners still face the same developability questions that determine whether an asset can move forward. Poor solubility, limited bioavailability, stability concerns, and manufacturability constraints still shape what happens next. For sponsors working with poorly soluble APIs, that remains the real bottleneck.
When discovery outpaces development
AI can help sponsors identify promising candidates faster, but faster discovery does not make those candidates easier to formulate or manufacture. Greater discovery output often means more programs encounter familiar developability problems later. As discovery pushes further into complex chemical space, more candidates arrive with poor aqueous solubility, limited bioavailability, high melting points, poor solvent solubility, or processing constraints that narrow the available formulation path. Generative AI platforms are typically optimized to maximize target potency and selectivity rather than developability, which biases their output toward higher molecular weight, higher logP, and reduced aqueous solubility-territory where physicochemical liabilities tend to concentrate.
A molecule may look strong in early screening but still become difficult once formulation scientists and development partners begin building a stable dosage form, generating meaningful exposure, or designing a process that can hold up beyond the bench. Those issues are not secondary. They shape whether a program can move efficiently toward the clinic and whether the product can be manufactured in a way that is scalable and repeatable.
What early development must answer sooner
Greater discovery speed puts more weight on early development decisions. If a project is going to move quickly without creating larger problems later, sponsors need clear answers to these questions much earlier:
- Can the molecule reach therapeutic exposure at a practical dose?
- Can it remain stable through processing and storage?
- Is there a manufacturing path that can hold up as the program advances?
These questions often determine where delays emerge, where material gets wasted, and where teams end up revisiting work that should have been settled sooner. In that environment, early development has less room to operate as a supporting function and needs to play a more direct role in showing whether a promising candidate can move forward in a way that is both scientifically sound and operationally realistic.
Why formulation flexibility becomes critical
As more AI-surfaced candidates move into development, formulation strategy becomes more important because many of these compounds fall outside the comfort zone of standard approaches. The pattern is especially clear when a candidate brings poor solvent solubility, high melting points, limited thermal tolerance, or narrow processing windows. In those cases, sponsors need enough formulation latitude to work around the properties of the molecule without giving up what made it valuable.
Expanded formulation latitude often requires attention to:
- Bioavailability-enabling approaches: For some poorly soluble compounds, advanced amorphous solid dispersions may offer a practical path to improved absorption.
- Processing flexibility: When heat, solvents, or material properties limit conventional routes, solvent-free fusion processing or other nonstandard approaches may open viable options.
- Broader formulation design space: Greater excipient flexibility and more tailored multi-component systems can give formulators room to balance performance, stability, and manufacturability.
Why this matters for IT and Development
For technologists building AI platforms for drug discovery, the message is clear: optimizing for potency and selectivity alone creates candidates that look impressive in silico but often collapse under real-world constraints. Models that ignore solubility, bioavailability, or manufacturability will increasingly drive wasted resources downstream. Teams that integrate formulation and process knowledge into their training data, reward functions, and candidate ranking will deliver more value to pharmaceutical partners. The bottleneck isn't a lack of compute-it's the gap between what algorithms predict and what scientists can actually make.
Your membership also unlocks:
