Everyone can name the AI drug discovery companies. DeepMind folded every known protein and put the structures online. Recursion and Insilico Medicine raised enormous rounds on the promise of AI-designed medicines.
Everyone can name the AI drug discovery companies. DeepMind folded every known protein and put the structures online. Recursion and Insilico Medicine raised enormous rounds on the promise of AI-designed medicines. Isomorphic Labs spun out of DeepMind to turn structure prediction into drugs. Anthropic just launched Claude Science aimed straight at compressing life-sciences R&D.
So here is the surprise. When you rank who actually owns the patents in this field, none of those names sits at the top. The leaderboard belongs to Chinese technology and finance platforms, and it is not close.
I ran the landscape this week in Minesoft Origin. The pattern that came back changes how you should read every “AI cures disease” headline for the rest of the year.
Search the intersection of AI and drug discovery, and about 4,021 extended patent families come back. The shape of the curve matters more than the total. Before 2018 this was a trickle, fewer than 100 published families a year. Then it inflected. Published activity crossed 1,100 families in 2025, with 2026 already past 850 by early July and still counting.
That is a roughly tenfold rise in a decade, and it is still bending upward. A field growing this fast is a field where positions are still being taken. Nothing is settled. Which makes the ownership question the interesting one.
Rank the owners and the first names are Tencent, then Baidu, then Ping An, a Chinese insurance group, then the Chinese Academy of Sciences. Keep going and you pass several Chinese universities and even a pair of power utilities before you reach the companies the market associates with AI drug discovery.
DeepMind and its holding entity, Recursion Pharmaceuticals, and BenevolentAI each sit in the mid-30s by family count. Alphabet as a whole lands near 40. These are the names on magazine covers, and by raw volume they look like mid-tier players.
Two forces explain the gap, and neither means what the raw ranking first suggests.
The first force is geography. Roughly 69% of the families in this set are Chinese filings, against about 7% from the United States by publication country. A large share of the Chinese volume comes from platform companies and universities filing broadly around AI methods that touch health data, not from companies advancing specific drug programs. When an insurance group and two power utilities show up above the labs that designed real clinical candidates, that is a signal about filing behavior, not about who is closest to a medicine.
This is the same trap that shows up in every hot-technology landscape. The biggest number at the top of the chart is often the least differentiated work. Counting families rewards whoever files the most, not whoever owns the hard part.
The second force is strategy, and it is the one founders should study. The AI-native companies are not trying to win on volume. They are concentrating on specific sub-problems and going deep.
Split the field by what the filings actually do. Property, ADMET, and screening work is the largest slice at roughly 1,477 families. Drug-target interaction and binding prediction is close behind near 1,109. Protein structure prediction is a substantial 836. De novo molecule generation, inventing new candidate molecules outright, is the smallest and most frontier slice at about 292 families.
Now overlay the owners. DeepMind and Isomorphic Labs dominate the protein structure slice; the top results there are almost entirely theirs. Insilico Medicine recurs through the de novo generation slice, the very smallest bucket. These companies are not absent from the leaderboard because they are behind. They are choosing to plant flags on the hardest, most defensible sub-problems rather than spraying filings across the whole field.
That is the difference between a vanity portfolio and a strategic one. Narrow and deep on the sub-problem that matters beats wide and shallow across everything.
Step back and the landscape connects directly to the M&A wave reshaping pharma right now. Big pharma faces roughly $170 billion in revenue going off patent by 2030, and it spent about $134 billion in the first half of 2026 buying pipelines to refill the gap. AbbVie paid $10.9 billion for Apogee. GSK paid $10.6 billion for Nuvalent. Neither target sells a product. The buyers paid for the science and the patents that fence it.
AI drug discovery is the supply side of that same story. These companies build the tools that generate the next pipeline. When one of them becomes an acquisition or licensing target, the question a buyer’s diligence team will ask is not how many families it filed. It is whether it owns something defensible on a sub-problem the buyer cannot route around. A thousand generic machine-learning-for-health filings do not answer that question. A tight cluster of families on de novo generation or structure-based design might.
The house view we bring to valuation work applies cleanly here. Patents can lift an acquisition from a 2 to 3x revenue multiple into the 10 to 40x range, but only when the market independently wants the underlying technology. The IP does not manufacture demand. It captures it, and it decides who else gets locked out.
For founders building in this space, three things follow. Know which sub-problem you can actually own, because the field is too broad to lead everywhere. File deep on that, not wide for a bigger number. And document as you build, so that when a pharma buyer’s team goes looking, the depth is already on the record.
The names on the magazine covers understood this early. They are not losing the patent race. They are running a different one, on purpose. When you evaluate any AI drug discovery story from here, do the thing the raw leaderboard punishes. Read the portfolio, not the press release.
Patent data from Minesoft Origin, extended-family landscape searches run 2026-07-06. Counts are landscape-level triage, not a claim-scope read. Market and deal figures from public reporting (Life Science Daily, CNBC), retrieved July 2026. This is strategic analysis, not legal or investment advice.
Work with ipCapital Group
From invention to monetization, our team has guided 2,000+ engagements across the full IP lifecycle. Start with a free 30-minute discovery call.
Written by
Seth Cronin