Quick Answer: AI drug discovery can speed up target identification, diagnostics, and antibody and enzyme design, according to Biocon founder Kiran Mazumdar-Shaw. But discovery is only part of the timeline. Clinical validation, manufacturing, and patient capital decide how fast treatments actually reach patients, so AI helps most when it's paired with commercial and funding discipline.
At the Bengaluru Tech Summit 2026, Kiran Mazumdar-Shaw argued that AI could transform biotechnology. As The New Indian Express reported, her pitch covered faster drug discovery, better diagnostics, and shorter paths from a molecule to a usable therapy. Most coverage stops at the lab bench. That's the wrong place to stop. The hardest part of getting a drug to patients comes after the algorithm finishes its work, and that's where her warnings about funding and commercial viability matter most.
Why Biology Is the Hardest Problem AI Has Faced
Mazumdar-Shaw called biology the most challenging frontier for AI, and her reason is worth understanding. Biology has long been an iterative science. Researchers run repeated experiments to build evidence and validate findings, and that process is slow.
The deeper problem is that many root causes remain unclear. She pointed to why certain cancers return after treatment, a question that still stumps oncologists. When the mechanism is unknown, a model can only find patterns in the data you feed it. It can't tell you why those patterns exist.
Still, the progress is real. Google DeepMind's AlphaFold system predicted protein structures with accuracy comparable to experimental methods in the CASP14 assessment in 2020. The AlphaFold Protein Structure Database, run with EMBL-EBI, now hosts predictions for over 200 million proteins. Knowing a protein's shape is a major win. It just doesn't tell you why a tumour comes back.
Where AI Could Shorten the Pipeline
Mazumdar-Shaw listed several areas where she thinks AI can help. Each one targets a different step between an idea and a treatment:
- Target identification: finding the biological molecule a drug should act on
- Enzyme design and genetic modification: engineering proteins and cells with specific properties
- Antibody development: narrowing candidate antibodies before expensive lab screening
- Cell therapies: improving how engineered cells are designed and characterised
- Biopharmaceutical manufacturing: tuning production processes and quality control
She framed the goal as shortening the path "from identifying a drug target or molecule to making treatment viable and available to patients." That phrase is doing a lot of work. Speeding up one step doesn't speed up the whole chain.
Here's the misconception most readers carry: if AI makes discovery faster, drugs will reach patients faster. That's only partly true. The gains are real where experiments are slow, expensive, or numerous. They're much smaller where the timeline is set by regulators and human trials.
The Counter-Intuitive Part: Discovery Isn't the Slow Step
This is where most AI-in-pharma coverage goes soft. Drug development commonly takes a decade or more from early research to approval, and most candidates that enter clinical trials never reach the market. The trial phases exist to test safety and efficacy in people, and no model can skip that evidence.
So an AI tool that cuts discovery time in half may save months, not years, if trials and manufacturing run at their usual pace. That's a useful gain, but it's not the revolution the headlines imply.
That said, AI isn't irrelevant to trials. It can help with patient selection, trial design, and manufacturing quality. The evidence for these uses is still developing, though, and regulators will judge them on clinical results, not on how clever the algorithm looks. Treat claims about trial acceleration with the same scepticism you'd apply to any early-stage data.
Why Biotech Startups Need Patient Capital
The second part of Mazumdar-Shaw's talk was about money. She said the sector requires patient capital because developing and commercialising products takes longer than in digital technology. That's a structural point, not a complaint.
Compare the two models. A software team can ship a release in weeks and learn from users almost immediately. A biotech team may spend months on a single validation experiment and years before a clinical readout. Funding cycles built around software timelines don't fit that pattern.
| Factor | Software startup | Biotech startup |
|---|---|---|
| Iteration cycle | Days to weeks per release | Months to years per validation cycle |
| Time to first revenue | Often months | Often a decade or more from discovery to approval |
| Capital pattern | Staged rounds, faster exits | Long runway, large later-stage raises |
| Main gate | Market adoption | Clinical evidence and regulatory approval (e.g., CDSCO in India, FDA in the US) |
| Typical failure signal | Churn, poor retention | Failed trial endpoints, safety signals |
Mazumdar-Shaw suggested policy incentives that would let venture capital investors exit through public market listings and reinvest their returns in biotech. The logic is sound. If early investors can't realise returns, they won't stay for the long haul. She also acknowledged the seed capital and other early-stage support from the Union and Karnataka governments, which addresses the earliest funding gap.
Commercial Viability Is the Test Founders Skip
Mazumdar-Shaw made a point that many founders resist: scientific achievement isn't a product. She urged startups to assess market demand, commercial viability, and profitability, not just the science.
Here's the trap. A team builds a clever AI model that predicts a promising antibody. The data looks great. Then the team asks who pays for it, how many patients need it, and what a payer will accept. If those answers are vague, the model is a research asset, not a business.
Before you commit to an AI pilot, write down three things: the patient population, the payer, and the expected cost per validated candidate. If you can't fill in all three, the project isn't ready for funding, no matter how strong the model looks.
Compute, Supply Chains, and the Cost of Using AI
Mazumdar-Shaw also urged biotech companies to adopt AI across operations, including supply chains, logistics, and manufacturing. She put it bluntly: "If we don't use AI, I'm afraid we'll be left behind."
She also cautioned that AI requires substantial computing power and resources, so judicious use is vital. This is the part that often gets dropped from enthusiastic coverage. Large models consume serious compute budgets, and biological data is often noisy, incompletely labelled, or locked inside proprietary datasets. A model trained on weak data can produce confident, wrong answers.
Here's how a responsible team can start without overspending:
- Pick one bottleneck with a measurable cost, such as antibody screening turnaround or batch yield variance.
- Confirm the data is clean and representative enough to support a useful prediction.
- Set a compute budget and a cost-per-validated-result target before the pilot begins.
- Run the model's predictions against existing wet-lab results before trusting them on new work.
- Review whether the improvement justified the spend, and only then expand.
Mazumdar-Shaw's "left behind" warning deserves weight. But adopting AI without a defined problem is just an expensive way to feel modern.
Frequently Asked Questions
What is AI drug discovery?
AI drug discovery uses machine learning and related computational methods to find drug targets, predict molecular behaviour, design candidates, and prioritise experiments. Its main value is narrowing the search space so labs test fewer and better candidates. It doesn't replace clinical trials, which still determine whether a drug is safe and effective in people.
How can AI shorten drug development timelines?
AI shortens the early stages most, including target identification, protein structure prediction, and antibody or enzyme design. The gains are largest where experiments are slow or costly. Trials, regulatory review, and manufacturing scale-up set much of the timeline, so the overall reduction is usually smaller than headlines suggest.
Why does drug development take so long even with AI?
Drugs must show safety and efficacy in humans through staged clinical trials, and regulators review that evidence before approval. Manufacturing must also be validated at commercial scale. AI can speed up parts of the process, but it can't replace the time needed to observe biological outcomes in patients.
Why do biotech startups need patient capital?
Biotech products need years of research, clinical testing, and regulatory review before they generate revenue. Investors therefore need longer horizons and reliable exit routes, such as public listings. Software startups can reach revenue in months, but biotech companies often can't, which is why funding models built for software struggle in this sector.
Frequently asked questions aside, the practical takeaway is clear
Treat AI as a way to cut the number of experiments you need, not as a shortcut through clinical evidence. Pair every technical pilot with a commercial case and a realistic funding runway. Pass this to a founder who has a strong model but no pricing plan. Before your next AI pilot, write down the patient population, the payer, and the cost per validated candidate, and check whether the numbers hold. You'll learn more from that one-page exercise this week than from another model benchmark.