Small molecules, big bottleneck: why chemistry is still holding back AI drug discovery

TL;DR

Most AI drug discovery decks assume chemistry will somehow keep up, but small molecule chemistry is still slow, manual, and spread across vendors — that's where a lot of AI "efficiency" quietly disappears. Small molecules aren't going away: pills remain the backbone of medicine for chronic, global indications. The real unlock isn't one more model; it's the small number of teams finally treating small molecule chemistry as the main problem to solve.

Small molecules, big bottleneck: why chemistry is still holding back AI drug discovery

Most AI drug discovery decks still assume that chemistry will somehow keep up.

The model designs molecules, a CRO on another continent makes them, done.

In reality, small molecule chemistry is slow, manual, and spread across vendors. That is where a lot of AI "efficiency" quietly disappears.

And small molecules are not going away.

For all the excitement about antibodies, RNA and cell therapies, pills are still the backbone of medicine: oral, scalable, manufacturable at cost, and often the only realistic option for chronic, global indications.

The uncomfortable truth is that medicinal chemistry is still much more art than science.

A handful of senior chemists with deep tacit knowledge, long DMTA cycles, a lot of intuition and pattern recognition that lives in people's heads, not in code. We have better tools, not a genuinely automated, machine-readable medchem stack.

Excelsior Sciences just came out of stealth with $95M, and their bet is essentially this premise made concrete: chemistry is the bottleneck, so fix chemistry first.

• A "smart bloccs" system that turns synthesis into a more modular, automation-friendly language
• A fully automated, AI-steered facility in New York that runs the DMTA loop under one roof
• And, in these days of increasing global tension, an explicit reshoring story: make it viable to do high-end chemistry in Manhattan instead of defaulting to offshore CROs

They're trying to give AI a lab and factory that actually speaks its language, rather than bolting AI onto a twentieth-century chemistry stack.

They're not alone in recognizing this. iktos in Paris has been pushing from the software side, combining generative design, retrosynthesis, and robotics. Different entry point, same thesis: the real unlock is turning medchem from a craft into a system.

Until we make that shift, the interesting story in "AI drug discovery" is less "just one more model", and more the small number of teams who are finally treating small molecule chemistry as the main problem to be solved.

Key takeaways

• Most AI drug discovery decks assume chemistry will somehow keep up — but in reality, small molecule chemistry is slow, manual, and spread across vendors, and that's where a lot of AI "efficiency" quietly disappears.
• Small molecules are not going away: pills are still the backbone of medicine — oral, scalable, manufacturable at cost, and often the only realistic option for chronic, global indications.
• Medicinal chemistry is still much more art than science: tacit knowledge, long DMTA cycles, and pattern recognition that lives in people's heads, not in code.
• Excelsior Sciences came out of stealth with $95M betting that chemistry is the bottleneck — their answer is a modular synthesis system, a fully automated AI-steered facility, and an explicit reshoring story.
• iktos in Paris is pushing from the software side, combining generative design, retrosynthesis, and robotics — a different entry point, but the same thesis: turn medchem from a craft into a system.
• Until we make that shift, the interesting story in AI drug discovery is less "just one more model", and more the small number of teams finally treating small molecule chemistry as the main problem to be solved.