Recommended Articles
Fundraising

AI Biopharma's Infrastructure Turn: The Money Is Moving Down the Stack

By Angeled · October 13, 2026 · Fundraising
AI Biopharma's Infrastructure Turn: The Money Is Moving Down the Stack
AI Biopharma's Infrastructure Turn: The Money Is Moving Down the Stack

This week, three very different biotech financings landed within 48 hours of each other: Drylab AI raised a $3 million seed round led by Vertex Ventures to build autonomous AI research agents; Undecimal Bio launched with an oversubscribed $18 million seed co-led by 4BIO Capital and IQ Capital to solve siRNA delivery beyond the liver; and China's LigIT.AI closed a roughly CNY 50 million (about $7.5 million) seed round to develop oral cyclic peptides, according to Scrip/Citeline. Nutshell Therapeutics, an allosteric-drug maker, also raised new capital in the same window.

The common thread is not a single technology. It is where these companies sit in the drug-development stack. None of them are selling investors an "AI-designed blockbuster molecule." They are selling infrastructure: tools that make discovery cheaper, delivery possible, and modalities druggable. After a decade in which AI drug-discovery startups promised to reinvent the molecule itself, the capital is quietly migrating to the unglamorous layers underneath it.

The three infrastructure bets

Autonomous lab agents (Drylab AI). The Singapore-based company builds AI agents that plan experiments, select from more than 800 scientific tools and databases, interpret results, and troubleshoot failures — effectively software "scientists" that operate alongside human researchers. The company launched commercially in February 2026 and reports paying customers among pharmaceutical companies, cancer centers, and research institutes. The bet is that the bottleneck in R&D is not ideas but execution bandwidth.

Delivery platforms (Undecimal Bio). A spin-out from Eleven Therapeutics, Undecimal is attacking what it calls the "liver trap": RNA interference medicines work brilliantly against liver-expressed diseases, but getting siRNA into other tissues remains one of genetic medicine's hardest unsolved problems. Its DELiveri platform screens large combinatorial libraries of candidate ligands and uses agentic AI to decode structure-activity data, searching for conjugates that can carry siRNA into specific cell types — starting with renovascular diseases.

New modalities (LigIT.AI). Oral cyclic peptides occupy a coveted middle ground in drug design: larger than small molecules (so they can hit "undruggable" protein-protein interactions) but potentially deliverable as pills rather than injections. Making peptides orally bioavailable is a hard chemistry problem; AI-driven design is being applied to navigate it.

Where the drug pipeline actually bottlenecks

Understanding why investors are funding infrastructure requires a tour of the pipeline. Drug development runs: target identification → hit discovery → lead optimization → preclinical testing → three phases of clinical trials → regulatory approval → manufacturing and commercialization. The process takes 10–15 years and costs billions, with failure rates above 90% from Phase 1 onward.

AI's early promise concentrated at the very front of this funnel — predicting molecules. But the expensive failures happen downstream: compounds that fail in clinical trials, drugs that cannot be manufactured at scale, and medicines that cannot reach the right tissue. A slightly better molecule is worth little if it cannot be delivered (the liver trap) or cannot be taken conveniently (the injectable problem). Infrastructure bets attack failure modes that compound across every program a company will ever run, not just one candidate.

Where AI genuinely adds value — and where it is marketing

Separating substance from pitch requires asking one question: does the AI generate proprietary, causal data, or does it just re-analyze public data faster? Drylab's agents and Undecimal's delivery screening both generate new experimental data — the AI directs experiments and learns from their outcomes, a closed loop that compounds. That is structurally different from "AI-designed molecule" platforms trained only on public datasets, where every competitor can reach similar conclusions.

The marketing version of AI biopharma promises to collapse development timelines by an order of magnitude. The infrastructure version promises something less cinematic but more bankable: higher throughput, lower cost per experiment, and better targeting. Investors funding Undecimal and LigIT.AI are pricing platform economics — technology that can serve many programs and partners — rather than betting on a single clinical readout.

What biotech founders need now

Wet-lab partnerships are the moat. No AI biopharma platform survives on code alone. Drylab's traction came from deploying inside real labs; Undecimal sits at the Cancer Research UK Cambridge Institute. Founders should secure experimental partnerships before fundraising, because investors now treat "we have proprietary data generation" as the diligence gate.

Regulatory strategy is a funding strategy. Platforms that touch clinical development eventually face regulators. Founders should map the regulatory path for their modality early — siRNA delivery and novel peptide formats each carry distinct safety and characterization expectations — and articulate it in the deck.

Sell the pickaxes, price the platform. Infrastructure companies earn through partnerships, licensing, and milestone payments rather than single-asset lotteries. Founders should model revenue as platform economics (recurring access, per-program fees) and be honest about timelines: platform validation takes years, so raise to meaningful technical milestones, not to the next news cycle.

Angeled.com connects founders with investors.

Read Next