AI Drug Discovery Draws $2.1B as Research Advances

Isomorphic Labs closed a $2.1 billion funding round on May 13, 2026, led by Thrive Capital — the second-largest biotech fundraise ever, behind only Altos Labs, according to trade publication Endpoints News. The raise signals sustained institutional confidence in AI-driven drug development, even as the field’s most celebrated tools remain in early clinical translation. Separately, a Phase 2 trial of a synaptic regeneration drug and a new multilingual political-fact benchmark highlight the week’s broader AI research activity.

Isomorphic Labs Secures $2.1B for AI Drug Design

Isomorphic Labs, the Alphabet-founded company best known for AlphaFold, raised the $2.1 billion round at an undisclosed valuation, Forbes reported. The new capital follows a $600 million outside round from the prior year and gives the London-based firm a substantial war chest to advance its drug design pipeline.

Isomorphic President Max Jaderberg told Forbes the funding represents “a lot of validation of what we’ve been building out the past four-and-a-half, almost five, years.” The company’s core asset remains AlphaFold 3, released in May 2024, which extends protein-structure prediction to small molecules, peptides, and antibodies — the building blocks of most drug candidates.

Building on that foundation, Isomorphic has developed what it calls the IsoDDE (Isomorphic Labs Drug Design Engine), which Jaderberg described to Forbes as “like half a dozen AlphaFold breakthroughs” combined. The company has not disclosed which specific drug candidates it plans to advance to clinical trials, keeping its pipeline details closely held despite the scale of the raise.

Isomorphic CEO Demis Hassabis — who also serves as CEO of Google DeepMind — won the 2024 Nobel Prize in Chemistry for AlphaFold’s protein-structure work, giving the company an unusual degree of scientific credibility alongside its commercial ambitions. The $2.1 billion raise is the clearest financial signal yet that institutional investors believe that credibility translates into drug development outcomes.

Tazbentetol Shows Sustained Efficacy in Schizophrenia Trial

In a separate development at the intersection of neuroscience and drug discovery, Spinogenix reported early Phase 2 results for tazbentetol, a first-in-class synaptic regeneration therapy, at the Schizophrenia International Research Society (SIRS) 2026 Annual Congress. According to Spinogenix’s press release, the drug produced a placebo-adjusted 6.3-point reduction in the Positive and Negative Syndrome Scale (PANSS) score in an add-on clinical trial.

The PANSS is a 30-item scale used to measure symptom severity in schizophrenia; reductions of this magnitude in an add-on setting — where patients remain on existing medication — are generally considered clinically meaningful, though the trial is still in Phase 2 and results have not yet been peer-reviewed.

The more notable finding may be durability: patients who discontinued tazbentetol after six weeks of use maintained efficacy for multiple days afterward. Spinogenix’s press release attributes this to the drug’s proposed mechanism — modulating fascin-1/F-actin dynamics to promote synaptic regeneration and the formation of dendritic spines with glutamatergic synapses, rather than simply suppressing symptoms acutely.

Beyond schizophrenia, Spinogenix is also investigating tazbentetol for Alzheimer’s disease, ALS, glaucoma, and diabetic retinopathy — conditions that share neurodegeneration as an underlying feature. The drug remains investigational, and Phase 3 data will be required before any regulatory submission.

PolitNuggets Benchmark Tests Agentic Fact Discovery

On the research infrastructure side, a team published PolitNuggets on arXiv (arXiv:2605.14002), a multilingual benchmark designed to evaluate how well large reasoning models (LRMs) embedded in agentic frameworks can discover and synthesize obscure, dispersed facts.

The benchmark constructs political biographies for 400 global elites, covering more than 10,000 political facts drawn from sources across multiple languages. The task is deliberately difficult: rather than retrieving well-documented information, models must piece together “long-tail” facts — details that appear infrequently and are scattered across heterogeneous sources.

To standardize evaluation, the authors introduce FactNet, an evidence-conditional scoring protocol that measures three dimensions separately: discovery (did the model find the fact at all?), fine-grained accuracy (did it get the details right?), and efficiency (how many resources did it consume to do so?).

The paper’s findings are sobering for agentic AI proponents. Across models and configurations, current systems frequently struggle with fine-grained detail accuracy and vary substantially in efficiency. The authors link agent performance back to specific model capabilities: short-context extraction quality, multilingual robustness, and reliable tool use all emerge as significant predictors of benchmark success. The work provides a concrete diagnostic framework for identifying where agentic retrieval systems break down — a gap the authors note has been under-evaluated relative to standard long-context QA benchmarks.

Cerebras IPO Validates Early AI Hardware Bets

The week also saw the Cerebras Systems IPO close as a strong market performer, generating billions for early investors including Benchmark, which holds a 9.5% stake in the AI chipmaker. TechCrunch reported that Benchmark general partner Eric Vishria co-led Cerebras’ $25 million Series A in 2016 — the year the company was founded — and has served on its board since.

Vishria told TechCrunch he nearly skipped the initial pitch: “It was five founders and a deck, and it was our first hardware investment in 10 years.” His skepticism dissolved by the third slide, when CEO Andrew Feldman argued that GPUs were architecturally mismatched for deep learning — a claim that predated Google’s 2017 Transformer paper by a year and now looks prescient given the AI compute boom.

Cerebras builds wafer-scale chips designed specifically for AI training workloads, positioning itself as an alternative to NVIDIA’s GPU-dominant supply chain. The IPO’s reception suggests public markets are willing to price in competition to that supply chain, even if Cerebras remains a distant second by revenue and deployment scale.

What This Means

The $2.1 billion Isomorphic raise and the Tazbentetol Phase 2 results together illustrate two distinct timelines in AI-assisted medicine. Isomorphic is betting that AI can compress the front end of drug discovery — target identification, molecule design, binding prediction — dramatically shortening the decade-plus development cycle. Tazbentetol, by contrast, is a more conventional drug that uses neuroscience insights (synaptic regeneration rather than receptor blockade) and happens to be studied in a period when AI tools are accelerating that science.

The PolitNuggets benchmark is a useful corrective to optimism about agentic AI. Even on a structured task with clear ground truth — political biography facts — current systems show meaningful accuracy gaps and efficiency variance. That finding matters for any enterprise deploying agentic retrieval in high-stakes domains, including drug discovery literature synthesis.

Taken together, the week’s research activity reflects a field that is well-funded, scientifically productive, and still confronting real performance ceilings in both model capability and clinical translation.

FAQ

What is Isomorphic Labs and how does it use AI for drug discovery?

Isomorphic Labs is an Alphabet-founded company that applies AI models — most notably AlphaFold 3 — to predict molecular structures and accelerate the drug design process. Its IsoDDE platform extends that work to design candidate molecules across small molecules, peptides, and antibodies, aiming to reduce the time and cost of bringing new drugs to clinical trials.

What does a 6.3-point PANSS reduction mean for tazbentetol’s schizophrenia trial?

The PANSS (Positive and Negative Syndrome Scale) measures symptom severity in schizophrenia across 30 items; a placebo-adjusted reduction of 6.3 points in an add-on trial is considered clinically meaningful by most psychiatric research standards. However, tazbentetol is still in Phase 2, meaning larger Phase 3 trials are required before regulators can assess it for approval.

What is the PolitNuggets benchmark and why does it matter for agentic AI?

PolitNuggets is a multilingual benchmark published on arXiv that tests whether agentic AI systems can accurately discover and synthesize obscure political facts from dispersed, multilingual sources. It matters because most existing benchmarks evaluate retrieval from well-documented information, while PolitNuggets specifically targets the harder “long-tail” case — the kind of fact-finding that enterprise and research deployments actually require.