Demis Hassabis has raised enough money to remove capital scarcity as the immediate excuse for Isomorphic Labs. In May 2026, the AI-first drug-design company secured $2.1 billion in Series B financing led by Thrive Capital. Alphabet and GV returned; MGX, Temasek, CapitalG and the UK Sovereign AI Fund joined. The round followed a $600 million external financing in 2025 and gives the company resources to expand its models, hire clinical talent, scale globally and move an internal drug pipeline towards patients.
The size of the cheque changes the burden of proof. Isomorphic no longer needs to demonstrate that investors believe artificial intelligence can improve drug discovery. It must show that its “unified drug design engine” can produce medicines with better odds, shorter timelines or lower cost than a well-funded conventional programme. In February, the company disclosed technical results for IsoDDE, saying it more than doubled AlphaFold 3’s accuracy on the hardest cases in one protein-ligand benchmark, predicted binding affinity more accurately than leading physics-based methods at a fraction of the time and cost, and identified previously hidden binding pockets from protein sequence.
Those capabilities can make discovery faster. They do not establish human safety or efficacy. Isomorphic has announced research alliances with Novartis, Eli Lilly and Johnson & Johnson, and says its internal programmes have identified viable candidates. It has not publicly provided enough asset-level detail for outsiders to judge a lead molecule, indication, development timeline or clinical result. Hassabis, founder and chief executive of Isomorphic as well as co-founder and chief executive of Google DeepMind, has built a formidable engine. The next valuation depends on what exits the factory.
Beyond AlphaFold, into development risk
AlphaFold transformed access to predicted protein structures and earned Hassabis and John Jumper a share of the 2024 Nobel Prize in Chemistry. Structure prediction, however, solves only part of drug design. A molecule must bind the intended target, avoid others, reach the right tissue, remain in the body for an appropriate period, be manufacturable and demonstrate benefit without unacceptable toxicity. Each property can defeat a programme that looks excellent in a structural model.
Isomorphic’s strategic answer is not to rely on AlphaFold alone. IsoDDE combines multiple proprietary predictive and generative models across small molecules, antibodies, peptides, molecular glues and other modalities. The company describes a curated “dataverse” and large computing fleets that allow extensive virtual experiments. Its aim is to model several constraints together rather than optimise binding first and discover pharmacological weaknesses later.
If the models generalise to novel biology, the economics could be significant. Fewer synthesis-and-test cycles reduce laboratory time. Better affinity and property prediction can narrow the number of compounds made. Finding a cryptic pocket may turn an apparently undruggable target into a programme. The benefit compounds across a portfolio: a platform that removes months from each project can create capacity without adding laboratories at the same rate.
The danger is benchmark capture. Drug-design teams can choose public tests that favour their models, and retrospective accuracy does not reproduce a prospective programme in which the answer is unknown. Even strong performance on unseen pockets may not predict absorption, toxicity or disease relevance. Isomorphic’s February disclosure is useful technical evidence; prospective candidate outcomes remain the commercial standard.
Pharma partnerships provide external experiments
Novartis was among Isomorphic’s first major partners. The original 2024 alliance covered three challenging small-molecule targets; the companies added up to three more programmes in 2025 on the same financial terms, citing progress during the first year. Expansion is a meaningful signal because an experienced drugmaker had access to work that public investors did not. It is still an intermediate signal: adding targets does not reveal whether a candidate has entered human testing.
The Lilly collaboration covers undisclosed small-molecule targets. The Johnson & Johnson agreement announced in January 2026 extends across targets and modalities, including biologics. Together, the partnerships test whether IsoDDE works outside an internal team and across different pharmaceutical practices. They also give Isomorphic access to disease expertise, assays and development infrastructure that cannot be recreated by computation alone.
The financial structure has not been fully disclosed for every agreement, limiting valuation analysis. Research payments can offset spending; milestones and royalties create upside without requiring Isomorphic to fund every clinical trial. Partnered programmes also surrender some economics and control. A platform company needs a balanced portfolio: external work validates the engine and finances operations, while wholly owned assets retain enough value to justify its capital base.
Isomorphic says its internal pipeline focuses on oncology and immunology. Those are competitive fields with expensive trials and increasingly precise standards of care. A well-designed molecule is not enough if the target lacks clinical differentiation or patient selection is weak. The company’s hiring of an experienced chief medical officer and establishment of a US presence in 2025 acknowledged that the bottleneck is moving from model building towards development.
The capital round creates concentration
A $2.1 billion financing offers freedom to run several programmes, invest in compute and recruit across machine learning, chemistry and clinical development. It also increases fixed costs before product revenue. Model training, proprietary data generation and specialist talent are expensive. Drug trials consume cash on a different scale and timetable. Isomorphic must decide whether to become an integrated biopharmaceutical company, remain a design partner or operate both models without allowing one to subsidise unclear priorities in the other.
The investor list broadens the company beyond Alphabet. Thrive led the round; Singapore’s Temasek and Abu Dhabi’s MGX add international capital with long horizons; the UK fund reinforces a national interest in retaining a London-based scientific champion. That diversity reduces dependence on one sponsor, but strategic investors may have different expectations about geographic expansion, partnerships and returns.
Alphabet remains a distinctive advantage and a governance question. Isomorphic emerged from DeepMind and continues to collaborate with Google DeepMind. Access to research, compute and talent can accelerate progress. Hassabis leads both organisations, while Google DeepMind pursues frontier models and artificial general intelligence across many domains. Clear boundaries over intellectual property, data, cost allocation and management attention are essential. Minority investors need confidence that value created inside Isomorphic will remain there.
Max Jaderberg’s appointment as Isomorphic president from January 2026 creates an operating counterweight. He built much of the company’s AI approach and can run day-to-day scale while Hassabis spans both institutions. The arrangement must eventually show that leadership is deep enough for clinical execution, where decisions about dose, safety and trial design cannot wait for the founder’s attention.
Asia supplies capital, markets and biological diversity
Asian relevance is embedded in the new round through Temasek, and in the likely market for any successful oncology or immunology product. Japan has a powerful pharmaceutical industry and deep structural-biology expertise. South Korea and Singapore invest in translational medicine; China has expanded biotechnology research and clinical capacity; India combines chemistry talent with large disease populations. Isomorphic can partner, license or build regional development capability across these markets.
The platform also needs Asian data. Protein structures are universal, but disease genetics, immune responses, care pathways and drug metabolism vary. A discovery engine trained disproportionately on Western datasets may design excellent molecules while producing weaker trial strategy for Asian patients. Diverse clinical and multi-omic data should influence target selection and biomarker design before assets enter late-stage development.
Data access carries geopolitical and regulatory constraints. Health and genomic information may need to remain within national borders. Federated modelling can help, but common definitions and quality standards remain necessary. Partnerships with trusted local institutions may be more valuable than acquiring another undifferentiated dataset. The company must also manage concerns over dual-use biology and the security of powerful generative models.
For Asian drugmakers, Isomorphic’s rise creates pressure. Buying access to a frontier engine may reduce discovery time, yet it can also transfer data and strategic learning to an external platform. Building an independent system requires compute, talent and coherent experimental data that few companies possess. A middle route—joint programmes with clear model and data rights—will determine who captures the improvement as AI becomes part of routine research.
The clinic is an adversarial benchmark
Hassabis has repeatedly succeeded by defining difficult problems that machine learning can attack: games, protein folding and multimodal reasoning. Drug development is different because the objective is noisy, delayed and ethically constrained. There is no unlimited training environment for human trials. Biology changes across patients, and the most important errors can take years to appear.
Isomorphic’s engine may still create a major advantage. Better molecular prediction can eliminate weak compounds sooner and explore chemical space that human teams would miss. The company’s partnerships and technical disclosures indicate real progress. The $2.1 billion round ensures it can run the experiment at meaningful scale.
The next proof should be asset-specific: a named candidate, the target and indication logic, the time and experimental cycles required to design it, and prospective evidence that predicted properties held. Human data must then show safety, exposure and biological effect. Without that chain, aggregate claims of viable candidates remain difficult to value.
Hassabis has financed Isomorphic as though AI drug design is ready to become an industry. The clinic will apply a harsher benchmark than any structure-prediction test, one that cannot be optimised after the answer is known. Passing it once will not prove a universal engine. Passing it repeatedly, across partners and Isomorphic’s own pipeline, would justify the scale of capital now committed to his most important healthcare bet.
