AI in Bio Weekly Review: Phase 3 Trials, Mega-Rounds, and the Open-Source Shift — June 18-24, 2026

The week of June 18–24, 2026 was one of the densest in AI-bio news in the past year. Generate Biomedicines dosed the first patient in SOLAIRIA-1, the first pivotal Phase 3 trial of a fully AI-designed antibody. Insilico Medicine signed a $2.5B AI discovery partnership with SK Biopharmaceuticals at BIO 2026. Isomorphic Labs confirmed its $2.1B Series B — one of the largest private rounds in AI-bio history. And on the infrastructure side, Boltz and ByteDance’s Protenix moved to fully open-source AlphaFold-3-grade models, fundamentally changing who has access to frontier protein-structure prediction.

Below is a deep read on what actually moved the field this week, and where AI in drug discovery stands as of June 2026.

The Headlines

1. Generate Biomedicines doses first patient in SOLAIRIA-1 — first Phase 3 of an AI-designed antibody

Generate Biomedicines announced in late 2025 and confirmed through Q1 2026 that the first patient had been dosed in SOLAIRIA-1, a global Phase 3 trial of GB-0895, a long-acting anti-TSLP antibody for severe uncontrolled asthma. SOLAIRIA-1 enrolls ~1,600 adults and adolescents; a second pivotal trial, SOLAIRIA-2, is running concurrently.

GB-0895 is significant because it was engineered by Generate’s generative biology platform to bind TSLP with ultra-high affinity, an extended half-life enabling once-every-six-month dosing, and high target specificity. None of these properties — half-life extension, affinity maturation, specificity tuning — are individually novel; what Generate claims is that all three were co-optimized in a single design loop rather than tuned sequentially.

The Phase 3 readout is expected in late 2027 / early 2028. If the trial reads out positive, GB-0895 becomes the first AI-designed antibody to complete a pivotal program. That’s not the same as “first AI-designed drug approved,” but it’s the closest the field has come to the regulatory finish line.

Generate itself is one of the better-capitalized AI-bio companies: the company raised a $400M IPO in February 2026 (one of the largest biotech listing debuts in years, and the first Flagship Pioneering portfolio company to go public in 4.5 years). It’s now listed on Nasdaq under GENB.

2. Insilico Medicine + SK Biopharmaceuticals: $2.5B CNS AI discovery deal

At BIO 2026 (San Diego, June 22–25), Insilico Medicine announced an AI-driven drug discovery collaboration with SK Biopharmaceuticals worth up to $2.5 billion in milestones plus royalties, focused on neuroimmune CNS disorders — neuroinflammatory, neurodegenerative, and rare neurological indications.

The deal uses Insilico’s Pharma.AI platform (Chemistry42 + Biology42 + target discovery modules) to drive target-to-candidate discovery. SK brings US commercialization infrastructure and CNS clinical development experience.

This is the third mega-deal Insilico has signed with a major Asian pharma in the past 18 months (the previous one being a $2.75B Eli Lilly deal for the same kind of multi-target generative-chemistry partnership). Insilico’s strategy is increasingly clear: build the most productive generative-chemistry platform in the field, license it to every major pharma that wants AI capacity without building it in-house, and use the cash flow to fund internal clinical assets like rentosertib.

Speaking of which: Insilico confirmed this week that the oral rentosertib program remains on track for Phase 3 initiation in H2 2026 for idiopathic pulmonary fibrosis (IPF). Rentosertib is a TNIK inhibitor — the molecule that anchored Insilico’s landmark 2024 Nature Biotechnology paper showing a fully-AI-discovered drug reaching the clinic in under 18 months from target ID. The H2 2026 Phase 3 start would make rentosertib arguably the most important AI-bio clinical milestone of the year.

Insilico is also publicly traded — it completed its Hong Kong Stock Exchange IPO on December 30, 2025 (HKEX: 3696), raising ~$300M in what was the largest Hong Kong biotech IPO of 2025.

3. Isomorphic Labs: $2.1B Series B led by Thrive Capital

Isomorphic Labs — the Alphabet-owned DeepMind spinout running AlphaFold 3 commercially — closed a $2.1B Series B in late May / early June 2026, led by Thrive Capital with participation from Alphabet, GV, MGX, Temasek, CapitalG, and the UK Sovereign AI Fund.

This is, by some distance, the largest private financing round ever raised by an AI-drug-discovery company in Europe. It also cements Isomorphic’s position as the best-capitalized pure-play AI-bio company globally.

The strategic logic behind the round is twofold:

1. Clinical translation capacity. Isomorphic’s existing pharma partnerships — Eli Lilly (small molecule) and Novartis — together could be worth up to $3B in milestones. That capital is a long tail of payments contingent on Phase 1/2/3 readouts. Isomorphic needs deep operating capital to scale its internal clinical-readiness function.
2. The AlphaFold 3 commercial bet. Isomorphic is the only entity with a fully-supported commercial license to AlphaFold 3 (Google’s release was restrictive — open only for non-commercial use). As Big Pharma AI discovery capacity matures, Isomorphic becomes the natural “outsourced AlphaFold shop” for companies that don’t want to build the stack in-house. The Thrive-led round is funding that expansion.

Isomorphic has been quiet about its internal pipeline, but Sir Demis Hassabis’s Nobel lecture (2024) framed the company as building toward an end-to-end drug-design system. The Series B is the cash to make that real.

4. Boltz and Protenix open-source AlphaFold-3-grade models

On the open-source infrastructure side, two releases this week matter more than most commercial news:

• Boltz (from MIT PhD students) — released Boltz-2, a fully open-source biomolecular interaction model under MIT license. Boltz-2 is the first AI model to approach the performance of free-energy perturbation (FEP) simulations for binding-affinity prediction while running 100× faster. Boltz has raised $28M with a Pfizer partnership — a commercial-grade open-source bet on capturing drug-discovery infrastructure by undercutting proprietary alternatives.
• Protenix (from ByteDance) — released an open-source AlphaFold-3-class model under Apache 2.0 license, with parameter counts close to AF3 itself.

The combined effect: the technical moat of proprietary structure prediction has effectively collapsed. Any lab can now run AF3-grade prediction for free, locally. This is good news for academic and biotech research, but it puts pressure on commercial structure-prediction vendors (Schrödinger, part of the Isomorphic stack, certain Ginkgo services) to differentiate on integration, throughput, and downstream workflow rather than on raw model accuracy.

The Boltz framing is worth quoting: “When Google locked the door, three MIT students picked the lock.” That’s the right read on where open-source AI-bio infrastructure sits in June 2026.

The Other Notable Moves This Week

Chai Discovery raised $130M in a Series B in late 2025/early 2026 at a $1.3B valuation, led by Oak HC/FT. Chai-2 — their antibody-design model — reportedly hit 16% zero-shot success across 52 protein targets in two-week design loops. Chai also signed a biologics-discovery partnership with Eli Lilly in January.

Verge Labs (formerly Verge Genomics) emerged in May 2026 as a pure AI platform play after its lead ALS asset failed in clinical trials. This week, the company published a new AI model for patient stratification in neurology clinical trials — a pivot from “we will discover the drug” to “we will help you run the trial better.” Verge is the highest-profile example so far of an AI-native biotech pivoting away from internal pipeline to platform licensing.

Protillion Biosciences signed a multi-target discovery collaboration with Merck (MSD) worth up to $510M in milestones, focused on lab-in-the-loop AI drug design. This is one of the largest AI-discovery partnerships Big Pharma has signed with a small platform company in 2026.

Iambic Therapeutics — whose CEO Thomas Miller presented at BIO 2026 on Monday — has its NeuralPLexer protein-ligand structure-prediction model featured on a Nature Machine Intelligence cover article in 2026. Iambic’s Takeda partnership (signed February 2026) is now in active deployment.

Novo Nordisk × OpenAI — announced in April 2026, embedding AI agents across Novo’s entire operational stack (not just drug discovery, but commercial, manufacturing, regulatory). This is the broadest Big-Pharma × AI-platform deal to date and signals where the field is heading: AI as operational infrastructure, not just a discovery tool.

Cradle Bio — Dutch-Swiss protein-engineering company — raised a $73M Series B to scale its ML-driven protein design platform, with heavy focus on antibody and enzyme engineering for pharma and industrial biotech.

OpenEvidence — the “ChatGPT for doctors” clinical-decision-support platform — closed a funding round in January 2026 that doubled its valuation to $12B, led by Thrive and DST. OpenEvidence is a non-drug-discovery AI-bio company, but its existence signals how wide the AI-bio umbrella has become.

The State of AI in Bio, June 2026

Pulling back from individual announcements, where does AI in drug discovery actually stand?

1. The clinical-validation question is being answered in real time

The standard critique of AI-bio from 2020–2024 was “where are the clinical readouts?” As of mid-2026, that question is being answered in stages:

• Phase 1 (safety): AI-designed drugs are reportedly passing at a ~90% rate — much higher than the industry-average ~70%. This is consistent with the thesis that AI does well at producing clean, drug-like, non-toxic molecules.
• Phase 2 (efficacy): The picture is more mixed. Multiple AI-discovered compounds have failed in Phase 2 efficacy trials at a rate of ~60% — similar to the traditional pharma industry baseline of ~30% Phase 2 → Phase 3 transition. AI clearly doesn’t (yet) break the efficacy barrier. The Verge Labs ALS failure is the most visible recent example of this.
• Phase 3 (pivotal): Only Generate Biomedicines’ GB-0895 has reached a fully-AI-designed Phase 3 antibody program, and SOLAIRIA-1 only dosed its first patient recently. No AI-designed molecule has completed a pivotal trial yet. Readouts start arriving late 2027 / 2028.

The honest read: AI accelerates the front of the pipeline (target ID → IND-enabling) and may improve Phase 1 safety rates, but the fundamental biology of human efficacy has not been solved by any algorithm. Companies that built business models around “AI will solve Phase 2 efficacy” are repricing.

2. Capital concentration is real and accelerating

The top of the funding stack is now decisively concentrated:

Company	Total disclosed funding	Stage
Isomorphic Labs	$2.1B (Series B) + Alphabet backing	Pre-clinical, partnerships-driven
Xaira Therapeutics	~$1.3B disclosed	Pre-clinical
Generate Biomedicines	$400M IPO Feb 2026 + prior	Phase 3 (GB-0895)
Insilico Medicine	$300M IPO Dec 2025 + partnerships	Phase 2/3 (rentosertib)
Pathos AI	$467M across 3 rounds	Pre-clinical
Eikon Therapeutics	~$700M+	Pre-clinical/Phase 1

The middle tier is dramatically underfunded relative to the top. Most AI-bio companies in the $20–80M raised range are facing a brutal 2026–2027 fundraising window, and many will fold, get acquired, or pivot to platform-licensing-only.

3. Open source is eating structure prediction, but not yet activity prediction

The Boltz / Protenix / OpenFold-3 ecosystem has effectively made AlphaFold-3-class structure prediction a commodity. This is good news for academic and small-biotech research. What open source has not yet cracked is:

• Binding-affinity prediction at FEP-grade accuracy (Boltz-2 is the closest, but still 10-20% off clinical-grade FEP)
• Generative chemistry for lead optimization (most open generative-chemistry models still underperform proprietary Pharma.AI, Chemistry42, Iambic’s NeuralPLexer, and Generate’s Chroma)
• De novo antibody design at clinical-stage hit rates (Chai-2’s 16% zero-shot success is the highest published number; the field’s median is single-digit)

This means the defensible AI-bio IP is moving from “we have a better model” to “we have better proprietary training data + better feedback loops with wet-lab validation.”

4. Public-market sentiment is bifurcated

The publicly traded AI-bio companies have split into two camps:

• The haves: Generate (GENB, +60% since IPO), Insilico (3696.HK, +35% since listing), Novo Nordisk (riding the OpenAI partnership narrative), Eli Lilly (carrying Insilico + Chai partnership signals)
• The have-nots: Recursion (RXRX, $3.32 stock price as of June 2026, down from 2024 highs), Schrödinger (SDGR at $15.76, well below historical peaks), Verge Genomics (now private after the Verge Labs rebrand)

The market is rewarding companies with clinical-stage AI-designed assets and punishing companies with AI platforms but no clinical data. This is a healthy correction: 2020–2024 saw platform-only companies valued like clinical-stage biotechs. That arbitrage is over.

5. The FDA is catching up

In April 2026, the FDA issued a Request for Information on an AI-Enabled Optimization of Early-Phase Clinical Trials Pilot Program — the agency’s first formal step toward regulating AI use in clinical trial design, patient recruitment, and adaptive design. The pilot is expected to launch in H2 2026.

This is small in scope but important in signal: the FDA is moving from “we’ll let AI do what it does” to “we want to know what AI did and why.” For AI-bio companies, the implication is that AI-generated trial designs and AI-selected patient cohorts will need to be explainable and reproducible — pushing back against “black box” approaches.

6. The infrastructure layer is consolidating

The wet-lab infrastructure for AI-bio is consolidating around a few players:

• Sapio Sciences (LIMS / ELN / ELaiN partner ecosystem) — SapioCon 2026 announced “beyond AI hype toward smarter science”
• Benchling — embedded AI in the notebook (66% of Benchling users now use AI for “scientific reporting” per their 2026 report)
• Scispot — computational data lakes for ML training loops
• Ginkgo Bioworks — synthetic biology foundry services

The pattern: every major AI-bio company still needs a wet-lab platform to feed training data back to its models, and the winners are consolidating around vendors that natively speak both LIMS and ML pipelines. Vendors that only do one or the other are getting squeezed.

What to Watch Over the Next 6 Months

Late Q3 2026 — rentosertib Phase 3 initiation. If Insilico hits this milestone on time, rentosertib becomes the most-watched AI-drug clinical asset globally. First interim data ~2028.

Q4 2026 — FDA AI-clinical-trial pilot launch. Will set the regulatory template for how AI-designed trials are reviewed.

Q1 2027 — SOLAIRIA-1 interim look. Generate’s first Phase 3 interim efficacy readout will be the first major efficacy data point for an AI-designed antibody. Even an interim futility analysis will move the whole sector.

Throughout 2026–2027 — second-tier AI-bio consolidation. Expect 5–10 mid-cap AI-bio companies (Pathos, Cradle, Iambic, smaller Recursion peers) to either raise down rounds, get acquired by pharma, or pivot to platform-only models.

2H 2026 — open-source activity-prediction release. If Boltz or a peer publishes a model that closes the gap on FEP-grade affinity prediction, the commercial structure-prediction market collapses further.

The Bottom Line

The week of June 18–24, 2026 will be remembered as the week AI in bio stopped being a thesis and started being an industry.

Generate Biomedicines is dosing AI-designed antibodies in Phase 3. Insilico is signing billion-dollar pharma partnerships and taking its AI-discovered molecule into Phase 3. Isomorphic Labs has the capital to compete with the largest pharmas on AI-discovery capacity. Boltz and Protenix have made AlphaFold-3-class prediction free. The FDA is regulating AI in trials. The public market is rewarding clinical assets and punishing pure platforms.

The companies that will survive 2026–2027 are the ones that combined (a) AI capability, (b) proprietary wet-lab data loops, (c) at least one clinical-stage asset, and (d) commercial pharma partnerships. The companies that won’t survive are the ones that still pitch “AI platform” without one of those four.

It’s a real industry now. And it’s moving fast.

---

Sources: BIO 2026 announcements (San Diego, June 22-25), Insilico Medicine + SK Biopharmaceuticals press release (June 22, 2026), Isomorphic Labs Series B announcement (May 2026), Generate Biomedicines Q1 2026 update and SOLAIRIA-1 dosing, Boltz-2 release notes, Biohub world-model release (May 27, 2026), Protillion-Merck deal (June 16, 2026), Verge Labs STAT News profile (June 16, 2026), FDA Federal Register AI-clinical-trial RFI (April 29, 2026), Novo Nordisk-OpenAI partnership (April 14, 2026), Recursion Pharmaceuticals pipeline updates, Schrödinger SDGR Q1 2026 transcript, OpenEvidence January 2026 funding round, Cradle Bio Series B, Chai Discovery Series B, peer-reviewed publications and company press releases.