TechFEATURE

AI Drug Discovery in 2026: From AlphaFold's Nobel Prize to Zero Approved AI-Designed Drugs

林紀旭 James LinEditor-in-Chief
Published · Updated
AI's role in drug discovery has produced a Nobel Prize and a near-$3 billion pharma bet, but not yet a single approved drug. DeepMind's AlphaFold earned the 2024 Nobel Prize in Chemistry, and its database now covers more than 200 million protein structures. Its spin-off, Isomorphic Labs, secured collaborations worth a combined near-$3 billion with Eli Lilly and Novartis, yet has no drug in human trials and has pushed its first clinical trial from end-2025 to end-2026. Insilico Medicine's rentosertib, the most advanced AI-designed candidate, only entered Phase III in July 2026, with no AI-designed drug approved anywhere globally to date.

From Nobel Prize to AlphaFold3: How far has AI protein prediction advanced?

DeepMind's Demis Hassabis and John Jumper won the 2024 Nobel Prize in Chemistry for developing AlphaFold, a protein structure prediction toolCITE:E1. The other half of that year's prize went to David Baker for computational protein designCITE:E1. The technology has continued to advance: AlphaFold3, released in May 2024, extended prediction beyond single proteins to complexes involving DNA, RNA, small-molecule ligands, and ions, reaching protein-ligand accuracy roughly 50% higher than the best conventional docking methods on the PoseBusters benchmarkCITE:E2. The resulting resource has also grown sharply in scale: the AlphaFold structure database expanded more than 200-fold, from roughly 1 million structures to over 200 million predicted structures, covering nearly all proteins known to science and made freely available to researchers worldwideCITE:E3.

Why are big pharma companies committing billions? How did Isomorphic Labs win over Eli Lilly and Novartis?

Isomorphic Labs, DeepMind's drug-discovery spin-off, announced two pharmaceutical collaborations in January 2024CITE:E4. Eli Lilly's deal included a $45 million upfront payment, while Novartis's deal included a $37.5 million upfront paymentCITE:E4. Combined, the two collaborations carry a potential value — excluding royalties — of near $3 billionCITE:E4.

Why is the commercial promise proving hard to deliver? Why did Isomorphic Labs' clinical progress slip?

Isomorphic Labs has not yet advanced any drug into human clinical trials, and its first trial timeline has slipped from end-2025 to end-2026CITE:E5. The company had previously stated it would begin its first clinical trial before the end of 2025CITE:E5. That one-year shift means the interval between AI-accelerated discovery work and an actual clinical readout now runs several more years before any resulting therapy could reach patientsCITE:E5.

Why hasn't any AI-designed drug been approved yet? What does rentosertib's case show?

Insilico Medicine's rentosertib, the most clinically advanced AI-designed drug candidate to date, only reached Phase III in July 2026, and no AI-designed drug has secured global approvalCITE:E6. Both the biological target and the molecule itself were identified and designed using generative AICITE:E6. In its Phase IIa trial for idiopathic pulmonary fibrosis, patients receiving rentosertib showed a mean 12-week increase in forced vital capacity of 98.4 mL, compared with a 20.3 mL decline in the placebo groupCITE:E6. Insilico Medicine initiated the drug's Phase III trial in July 2026CITE:E6.

The arc in numbers

MilestoneFigureDate
Nobel Prize in Chemistry awarded for AlphaFoldHassabis, Jumper, Baker2024
AlphaFold3 protein-ligand accuracy vs. conventional docking (PoseBusters)~50% higherMay 2024
AlphaFold structure database growth~1M → 200M+ structures (200x)cumulative
Eli Lilly upfront payment to Isomorphic Labs$45 millionJan 2024
Novartis upfront payment to Isomorphic Labs$37.5 millionJan 2024
Combined potential deal value (excl. royalties)near $3 billionJan 2024
Isomorphic Labs first clinical trial timelinedelayed end-2025 → end-2026as reported
rentosertib Phase IIa, 12-week FVC change (treatment vs. placebo)+98.4 mL vs. -20.3 mLtrial data
rentosertib Phase III initiationJuly 20262026-07

What does this mean

The record shows two timelines moving at different speeds. AI's contribution to understanding biological structure has already been validated at the highest level — a Nobel Prize and a database of more than 200 million structuresCITE:E1CITE:E3 — and has drawn near-$3 billion in pharma commitments to Isomorphic LabsCITE:E4. Translating that structural insight into an approved medicine has followed a slower path: Isomorphic Labs' first human trial has slipped a year, from end-2025 to end-2026CITE:E5, and even the most advanced AI-designed candidate, rentosertib, only reached Phase III in July 2026 with no approval yet secured anywhereCITE:E6. The gap between those two timelines is the current state of AI drug discovery.

📊 Evidence

FAQ

From Nobel Prize to AlphaFold3: How far has AI protein prediction advanced?

DeepMind's Demis Hassabis and John Jumper won the 2024 Nobel Prize in Chemistry for developing AlphaFold, a protein structure prediction toolCITE:E1.

Why are big pharma companies committing billions? How did Isomorphic Labs win over Eli Lilly and Novartis?

Isomorphic Labs, DeepMind's drug-discovery spin-off, announced two pharmaceutical collaborations in January 2024CITE:E4.

Why is the commercial promise proving hard to deliver? Why did Isomorphic Labs' clinical progress slip?

Isomorphic Labs has not yet advanced any drug into human clinical trials, and its first trial timeline has slipped from end-2025 to end-2026CITE:E5.

Why hasn't any AI-designed drug been approved yet? What does rentosertib's case show?

Insilico Medicine's rentosertib, the most clinically advanced AI-designed drug candidate to date, only reached Phase III in July 2026, and no AI-designed drug h…

📎 Sources

  1. deepmind.google
  2. isomorphiclabs.com
  3. deepmind.google
  4. isomorphiclabs.com
  5. investing.com
  6. insilico.com

Related data

Author's Take林紀旭 James Lin

The pattern here is a familiar one in translating computational breakthroughs into regulated products: the science of prediction scales far faster than the science of clinical validation. AlphaFold's database grew more than 200-fold to over 200 million structures, yet Isomorphic Labs' own clinical timeline has already slipped by a year, from end-2025 to end-2026, and rentosertib — the field's most advanced AI-designed candidate — only reached Phase III in July 2026 despite promising Phase IIa data. That asymmetry suggests the near-$3 billion in pharma commitments to Isomorphic Labs priced in discovery-stage acceleration, not clinical-stage certainty. The metric worth watching next is not another model benchmark but rentosertib's Phase III readout — the first real test of whether an AI-designed molecule's early trial signal holds up at the regulatory finish line.

林紀旭 James LinEditor-in-Chief

Related

FEATURE

Gold's 2026 Rally and Pullback: Inside the Central Bank Buying Wave and the Bitcoin Hedge Debate

Spot gold traded near $4,674 per ounce on August 24, 2026, up 39% from a year earlier but well below its January 28 record of $5,589. Central banks bought 863.3 tonnes net in 2025 — led by Poland, Kazakhstan, Brazil, Turkey and China's central bank — while global gold ETFs absorbed a record $89 billion. Bitcoin, meanwhile, fell from an October 2025 peak near $126,000 to roughly $70,000, undercutting its 'digital gold' claim.

EffectStory 編輯部 ·
FEATURE

Starlink's Lead: Inside the Satellite, User, and Revenue Gap With Every Rival

Starlink operates roughly 10,020 satellites and serves over 12 million users across 160-plus countries, while China's Guowang and Qianfan constellations each have fewer than 100 satellites in orbit against plans for 13,000 and 15,000. Amazon Leo has launched just 375-plus toward a 3,236-satellite target. Starlink's estimated $11.8 billion 2025 revenue and $800 billion valuation ahead of a planned 2026 IPO put it in a category no competitor currently occupies.

EffectStory 編輯部 ·
FEATURE

Quantum Computing's Real Distance to Practical Use: From Error-Correction Breakthroughs to IBM's 2029 Fault-Tolerant Target

Google's Willow chip and USTC's Zuchongzhi processor have both demonstrated below-threshold quantum error correction, and IonQ has hit 99.99% two-qubit gate fidelity — but these remain laboratory and component milestones. IBM's roadmap targets a 200-logical-qubit fault-tolerant machine by 2029, starting from today's 156-qubit Heron processor. Meanwhile, NIST's post-quantum encryption standards are already finalized, so cryptographic defense does not have to wait for that hardware timeline.

林紀旭 James Lin ·