Isomorphic Labs lands $2.1B to make AI drugs real

Alphabet’s biotech moonshot just got a $2.1 billion engine

Isomorphic Labs has finally priced its ambition. On May 12, 2026, the DeepMind spinout closed a $2.1 billion Series B led by Thrive Capital, vaulting its outside capital base to roughly $2.6 billion and making it one of the best-funded private companies in the history of AI drug discovery. Per Isomorphic Labs’ official announcement, the round added MGX, Temasek, CapitalG, and the UK Sovereign AI Fund to a cap table that already included Alphabet and GV — a sovereign-wealth-heavy syndicate that signals where the next phase of frontier AI financing is heading. Per Fierce Biotech’s coverage of the raise, the proceeds are earmarked for the company’s IsoDDE drug-design engine and a push to file its first Investigational New Drug application before the end of the year. That commitment is the real news. Isomorphic missed Demis Hassabis’s earlier pledge to put an AI-designed molecule into human trials by end-of-2025, and the Series B is the capital cushion that turns a missed deadline into a credible 2026 retry.

The dollar size matters because the AI-drug category has been waiting for a benchmark capital event that legitimizes the thesis at scale. Per R&D World’s analysis of the round, Isomorphic’s $2.1 billion private check is the largest disclosed Series B in the AI-drug-discovery space to date, eclipsing the comparable raises that turned Recursion Pharmaceuticals and Schrödinger into public companies. A round of this scale signals that growth-stage investors now believe the AI-drug-discovery thesis has matured past its skeptical adolescence — not because any AI-designed molecule has cleared the FDA (none have), but because the technical case for IsoDDE has become specific enough to underwrite. Thrive’s repeat lead position — the firm anchored both Isomorphic’s $600 million Series A and this round — is the cleanest signal in the cap table that the operating progress is real rather than rhetorical.

The strategic geometry around Isomorphic is what makes the deal more than a private financing footnote. Hassabis sits at the apex of three of the most consequential entities in the AI economy: he co-founded DeepMind, he serves as CEO of Isomorphic, and he is now a Nobel laureate in chemistry following the 2024 prize awarded for the AlphaFold work. Per Fortune’s profile of Hassabis, the unifying ambition across those roles is to compress what he calls “a century of biology into a decade” by reframing drug discovery as an engineering problem. The line lands as hyperbole until you parse the partnership stack: collaborations with Eli Lilly worth up to $1.7 billion in milestones, an expanded Novartis agreement, and a fresh Johnson & Johnson program announced in January. Per pharmaphorum’s earlier raise coverage, those partnerships now anchor a portfolio of 17 active programs spanning oncology, immunology, and cardiovascular disease.

The competitive context tightens the stakes further. Isomorphic raised at the same moment that the broader AI-drug landscape is consolidating around a small number of credible platforms. Recursion Pharmaceuticals, Insilico Medicine, Insitro, and Schrödinger collectively occupy most of the public and private attention in the category, but their technical strategies diverge sharply — phenomics, generative chemistry, machine-learning biology, and physics-based simulation respectively. Per a category survey from axis-intelligence, more than 173 AI-discovered drug programs are now in clinical development globally, with 15 to 20 expected to enter pivotal Phase III trials during 2026 — the cohort that will decide whether the entire decade-long AI-drug-discovery thesis was real or fashion. Isomorphic’s bet is that AlphaFold-derived structural prediction, married to an in-house engine trained on a far broader chemical space, can produce molecules with both the affinity and the developability that the rest of the field has struggled to deliver at scale.

The Series B also crystallizes a quieter pattern in AI financing this year: the dominance of sovereign-wealth-adjacent capital. MGX is the Abu Dhabi vehicle co-leading the largest AI infrastructure financings globally. Temasek is Singapore’s $300 billion sovereign investor. The UK Sovereign AI Fund signals that the British government wants a stake in any biotech with DeepMind’s pedigree headquartered in London. The lineup parallels the structure I unpacked in my May 8 piece on Anthropic’s Project Colossus orbital-compute partnership and reinforces a thesis the rest of the year will keep validating: at frontier scale, private capital alone no longer clears the bar, and any ambitious AI company will end up with a sovereign on its cap table within two rounds. The cost of admission to the AI-drug-discovery elite is now north of $2 billion, and the gatekeepers increasingly answer to governments.

Inside IsoDDE: what the money actually buys

The technical story is more interesting than the capital story, and it starts with how IsoDDE differs from AlphaFold 3. AlphaFold 3 — the molecular-structure prediction system Hassabis’s team published in 2024 — is a research artifact licensed broadly across academia, with limited commercial use restrictions. IsoDDE, by contrast, is a proprietary stack that uses AlphaFold-derived primitives as one input among many to generate, screen, and optimize novel drug candidates. Per Isomorphic Labs’ own technical post on the IsoDDE engine, IsoDDE achieved roughly 76% accuracy on antibody-antigen interface prediction in the FoldBench benchmark — compared with AlphaFold 3’s 48% — and more than doubled AlphaFold 3’s performance on protein-ligand binding-pocket identification. Those are the predictive subtasks that determine whether a candidate drug actually fits its target. If the numbers hold up in head-to-head external validation, IsoDDE has roughly the same step-function lead over AlphaFold 3 that AlphaFold 3 had over its predecessors.

The technical edge feeds directly into the partnership economics, which is where the platform thesis becomes a revenue thesis. Per TechCrunch’s January 2024 coverage of the original Lilly and Novartis deals, Eli Lilly committed $45 million in upfront cash and up to $1.7 billion in milestone payments to access IsoDDE for an unspecified portfolio of disease targets when the deal was announced in January 2024. Novartis paid $37.5 million upfront in the original deal and expanded the agreement in February 2025 — typically a sign that early checkpoints triggered the option to widen scope. Johnson & Johnson signed on in January 2026 as a cross-modality, multi-target collaboration. The cumulative committed milestone value across those three partnerships now exceeds $3 billion, and the Series B effectively bridges Isomorphic to a point where it can begin to collect those milestones rather than continue to spend cash to qualify for them.

The 17 active programs across oncology, immunology, and cardiovascular disease are the operational unit-of-progress. Per the Yahoo Finance summary of the round, each program advances through computational design, in vitro validation, in vivo animal testing, and Investigational New Drug filing — the standard preclinical sequence — but compressed by what Isomorphic claims is a factor of two to three on the earliest design-and-iterate loop. The economic value of that compression is the entire investment thesis. The pharmaceutical industry’s average preclinical-to-clinic timeline is 4-7 years, the cost is approximately $2 million per program per year, and the late-stage attrition rate is more than 90%. Cut the front-end loop in half, and you compound the savings across every iteration that would otherwise have failed in vitro — and you potentially raise the probability of clinical success because the molecules entering trials have been optimized against a deeper search space.

The infrastructure footprint underwrites the platform’s reach. Isomorphic is hiring across its London headquarters, a Cambridge, Massachusetts site that opened in 2024, and a Lausanne, Switzerland office that gives the team European-Union-jurisdiction biology talent without the Brexit overhead. Per Tech Funding News’ coverage of the round mechanics, the capital allocation includes both compute spend — much of it likely flowing back to Google Cloud’s TPU fleet under the Alphabet umbrella — and a substantial wet-lab build-out. The wet-lab component is the part of the strategy that most often gets underestimated by AI-only observers. IsoDDE’s predictions are only as good as the experimental data that calibrates them, and Isomorphic’s ability to run its own iterative biology in-house — synthesizing predicted molecules, testing them against real targets, and feeding the results back into the model — is the slowly-built moat that pure-software AI-drug competitors have struggled to match.

The 2026 IND-filing commitment is the operational tripwire that will determine how investors price Isomorphic’s next round. Hassabis publicly committed in 2024 to getting an AI-designed drug into human trials by the end of 2025, and the deadline slipped — a quiet acknowledgment that even the best-resourced platform in the field has had to confront the hard reality of preclinical wet-lab biology. Per the abhs.in deal analysis, the company is now targeting an IND submission for at least one lead candidate before the end of 2026, with first-in-human dosing in 2027. That timeline is aggressive but not heroic by AI-drug-discovery standards, and the Series B provides the cash runway to fund the GLP toxicology studies and chemistry-manufacturing-controls work that an IND requires. If Isomorphic clears the IND in 2026, the next round prices at a meaningfully higher valuation. If it slips again, the narrative damage is non-trivial — the public-equity comparables in the category have already absorbed several missed milestones from peers.

The economic asymmetry that makes the entire thesis defensible is worth stating cleanly. A single FDA-approved blockbuster oncology drug generates between $1 billion and $5 billion in annual peak sales. The pharmaceutical industry’s all-in development cost per approved drug, including failures, is roughly $2.6 billion per estimates from the Tufts Center for the Study of Drug Development. If IsoDDE’s productivity claims hold — even partially — the platform could plausibly run several parallel programs at a fraction of the per-program cost, with a higher per-program success probability. That math doesn’t require a “cure all disease” framing. It just requires that the platform improve the industry’s economics by 20-30%, which is a far more defensible target than the rhetorical maximalism Hassabis sometimes uses in conference remarks. The Series B is the bet that those gains compound across a portfolio of partners and an internal pipeline that, by 2030, looks more like a vertically integrated biotech than a software-licensing company.

The cemetery of AI drug promises is well-tended

The case for skepticism starts with the historical track record, which is unkind. No AI-designed drug has yet received FDA approval — none — as of May 2026, more than a decade into the modern AI-drug-discovery cycle. Per the same axis-intelligence category survey cited above, 173-plus AI-discovered programs are in clinical development, but the cohort of late-stage assets is small and the attrition rates so far appear no better than industry averages. The pattern is consistent with the history of computational chemistry, in silico screening, and structure-based drug design — each successive technology wave produced impressive academic results and disappointing translational outcomes once the molecules met the unforgiving complexity of human physiology. Isomorphic is making a more sophisticated bet than its predecessors, but the bet is structurally similar: that this generation of tooling will finally cross the gap between predictive accuracy and clinical efficacy. That gap has eaten every previous platform.

The bench-to-bedside cliff is the specific failure mode most likely to bite. AlphaFold 3 and IsoDDE excel at static and dynamic structural prediction — predicting how a candidate molecule fits a target. Per the Insilico Medicine $2.75 billion Lilly deal analysis from IntuitionLabs, the practical bottleneck in modern drug development is less about target-engagement prediction and more about ADME-Tox properties: absorption, distribution, metabolism, excretion, and toxicity, plus the messy realities of pharmacokinetics in human patients. Those properties depend on chemistry decisions far downstream of structure prediction, and the published evidence that AI-only platforms confer a major advantage at those stages remains thin. Isomorphic’s response is to build the wet-lab biology in-house and feed experimental ADME-Tox data back into IsoDDE’s training corpus, which is the right strategy — but it dramatically narrows the platform’s differentiation against integrated pharma R&D, because Big Pharma already does exactly that. The question becomes whether AI provides a sustainable edge on top of the same experimental data, or merely accelerates a process that pharma has been running for decades.

The benchmarking issue compounds the problem. IsoDDE’s reported 76% accuracy on FoldBench antibody-antigen interfaces and its outperformance against AlphaFold 3 on protein-ligand prediction come from internal benchmark exercises that the company designed itself. Per the European Biotechnology coverage of the round, Isomorphic has not yet released the full IsoDDE methodology in a peer-reviewed publication, in contrast to AlphaFold 3, which appeared in Nature and was subjected to extensive external evaluation. Internal benchmarks are useful but not dispositive. The history of machine-learning model benchmarks is full of cases where impressive in-house numbers degraded substantially under external scrutiny on out-of-distribution test sets. Until IsoDDE goes through that public crucible, the performance claims should be read as Isomorphic’s view of its own technology — credible but not validated by the broader scientific community.

The partner-dependency risk is the cleanest commercial counterpoint. Isomorphic’s revenue today is concentrated across Lilly, Novartis, and J&J — three pharma giants with sophisticated internal AI-drug capabilities of their own. Per the Fierce Biotech analysis of the partnership stack, each of those partners has the technical sophistication to evaluate IsoDDE rigorously and the strategic incentive to internalize the technology if it works. The optionality embedded in Big Pharma’s structure means Isomorphic is, in effect, providing pilots to its eventual competitors. The partnerships generate cash and validate the technology, but they also expose IsoDDE’s methodology and operating assumptions to teams that have every reason to replicate the relevant parts internally. The risk is not that any single partner walks away in the next 18 months — it is that, by 2030, Lilly and Novartis have built equivalent in-house tooling and the next renewal cycle prices Isomorphic’s technology meaningfully lower.

The Alphabet-parent overhang creates a different problem. Isomorphic remains majority-owned by Alphabet, and the parent’s strategic priorities can shift in ways that hurt the spinout. Per the BioSpace press-release coverage, Alphabet continues to participate in the Series B, which is constructive — but the broader Alphabet portfolio includes Verily, Calico, and a research-stage AI-for-medicine effort inside Google DeepMind itself that has not been fully separated from Isomorphic’s mandate. Internal turf wars between sibling companies inside a $2 trillion parent are a well-known dysfunction in conglomerate biotech, and there is no public indication that Alphabet has cleanly delineated which biology problems belong to Isomorphic versus DeepMind versus Verily. That ambiguity is manageable while every entity is well-resourced and successful. It becomes a problem the moment Alphabet’s cost discipline tightens, or the moment the Isomorphic IND succeeds and DeepMind decides that a follow-on platform should sit elsewhere in the parent’s structure.

The regulatory environment is the wild card no platform can hedge against. The FDA has not yet articulated a clear framework for evaluating AI-designed molecules differently from any other drug candidate, which means IsoDDE’s regulatory pathway is, on paper, the same as a traditional pharmaceutical company’s. Per coverage of the broader 2026 regulatory environment from PMC, the agency has signaled openness to AI-augmented submissions but has not committed to accelerated review pathways for AI-designed candidates. That posture is neutral, which is fine — but if the political environment shifts toward heightened scrutiny of AI in drug development (a plausible outcome given the broader AI-safety debate), Isomorphic could find itself absorbing approval-cycle delays that purely human-designed molecules would not. The risk is asymmetric: there is no foreseeable scenario where AI-designed drugs get a faster regulatory path than incumbents in the next two years, but there are several scenarios where they get a slower one.

The macro context is the final piece of the bear case. Isomorphic’s $2.1 billion raise lands inside an AI funding environment that has begun to show signs of late-cycle behavior. Per the MobiHealthNews summary of the round and category context, 2026’s mega-rounds have concentrated in a small number of companies — Anthropic, OpenAI, xAI, Isomorphic — at valuations that imply trajectories no historical category has sustained. If the AI-funding climate cools in 2027 — particularly if any of the high-profile public AI-drug names (Recursion, Insilico, Schrödinger) deliver disappointing Phase III readouts — Isomorphic’s next round will price against a weaker comparable set. The Series B is the cushion that lets the company push through a 12-24 month window where it doesn’t need to raise. Whether that cushion is sufficient depends entirely on whether the IND filing happens on schedule and whether the first-in-human data lands cleanly in 2027.

What 2026 has to prove before the cure narrative survives

The single most important deliverable is the IND. Isomorphic’s claim to category leadership cannot be sustained on benchmark numbers alone, and the entire pharmaceutical industry has been watching for the AI-drug-discovery field’s first credible clinical entrant from a marquee platform. Per the Greek City Times summary of Hassabis’s commentary, the company is positioning the 2026 IND submission as the moment IsoDDE’s technical case becomes a clinical case. The most plausible scenario is that Isomorphic files between two and four INDs across its 17-program portfolio before the end of the year, with the first-in-human dosing for at least one lead candidate slipping into early 2027. That sequence would be enough to validate the platform and trigger an upround. A second slipped deadline, by contrast, would expose Hassabis’s “decade of biology” framing to the kind of skeptical recasting that materially damages enterprise-narrative credibility.

The competitive response from incumbents is the second-order signal worth watching. OpenAI’s heavy push into pharmaceutical R&D through the Novo Nordisk and Sanofi enterprise deals — covered in my May 5 analysis of the OpenAI and Anthropic Wall Street partnerships — puts a generalist frontier model in the same procurement conversation as IsoDDE for some workflows. Microsoft’s bet on Recursion through Azure compute commitments, Google DeepMind’s continued internal biology work, and the increasingly aggressive AI-build-out inside Big Pharma’s own R&D organizations all narrow the strategic space inside which Isomorphic operates. The most likely 2026-2027 outcome is consolidation: one or two of the public AI-drug-discovery names get acquired by Big Pharma, several of the private players raise mezzanine rounds that prefigure either acquisition or IPO, and the category’s narrative settles around the small handful of platforms that have shipped clinical molecules. Isomorphic’s Series B is the bet that it ends up in that survivor set.

The Operator Checklist below distills what the partnership, capital, and competitive dynamics imply for anyone tracking AI-drug-discovery from the outside, whether as an enterprise procurement decision-maker, a public-equity analyst, or a biotech founder thinking about positioning:

• Watch the IND date, not the benchmark numbers. Isomorphic will publish more impressive IsoDDE benchmark scores in the next 12 months. They are diagnostic, not dispositive. The first IND submission and the subsequent first-in-human readout are the only data points that meaningfully reprice the platform’s narrative. Calibrate enterprise interest and equity exposure to those events, not to the next paper.
• Treat the partnership backlog as the real revenue model. Lilly, Novartis, and J&J are the three commercial counterparties that matter. Any expansion or contraction of those deals in 2026-2027 — particularly milestone hits or option-exercise decisions — is a stronger signal than any conference keynote. Read the pharma 10-Qs for line-item movement around AI collaborations.
• Map the wet-lab buildout as a moat indicator. The AI-only AI-drug-discovery competitors will continue to underperform integrated platforms on translational outcomes. Isomorphic’s Cambridge and Lausanne expansions are the clearest signal that the company understands this. Compare its wet-lab headcount growth against the headcount disclosed by Recursion, Insilico, and Insitro.
• Track sovereign-wealth participation as a category indicator. MGX, Temasek, and the UK Sovereign AI Fund’s presence in this round is a structural signal about who can write checks at AI-drug-discovery scale. Expect the same names in subsequent AI-biotech mega-rounds. The implications for biotech corporate governance, regulatory positioning, and IP localization are non-trivial.
• Discount the “solve all disease” rhetoric to a 20-30% productivity uplift. That is the level of platform improvement that would justify the $2.1 billion check on a present-value basis without requiring heroic clinical-success assumptions. Anyone modeling Isomorphic’s value at higher productivity multipliers is buying the founder’s framing, not the company’s economics.
• Anchor Alphabet-parent risk in 2027, not 2026. The Alphabet relationship remains constructive through this Series B, but the structural ambiguity about which biology mandate sits where inside Alphabet will become a problem the moment Alphabet’s strategic priorities shift. Watch for any change in the parent’s disclosure language about Isomorphic in upcoming 10-K filings.

The broadest implication is that AI drug discovery in 2026 is finally at the inflection point where capital, technology, and partnership economics have all converged on a small set of platforms that can credibly bid for the entire category’s value. Isomorphic Labs is the most strategically advantaged of those platforms — heir to AlphaFold, parented by Alphabet, anchored by Hassabis’s Nobel-grade scientific credibility, and now armed with $2.6 billion of total raised capital. The next 18 months will determine whether that combination produces the AI-designed drug the industry has been promised since 2018, or whether Isomorphic joins the cemetery of well-funded biotech platforms that never quite cleared the bench-to-bedside cliff. The Series B doesn’t resolve that question. It just buys Isomorphic the cleanest shot anyone in the category has ever had at answering it.

In other news

Meta unveils Muse Spark, its first post-Wang flagship model. Meta’s newly formed Superintelligence Labs, under Chief AI Officer Alexandr Wang, debuted Muse Spark as the company’s first major frontier model since the $14 billion deal that brought Wang aboard. Meta paired the launch with disclosed AI capital expenditure plans of $115-135 billion for 2026, per CNBC’s coverage of Meta’s AI push.

OpenAI consolidates ChatGPT, Codex, and the API under one team. OpenAI is merging its consumer chatbot, coding-agent product, and developer API into a single product organization, with a “super app” vision that integrates the Atlas browser. The internal restructure is the most explicit signal yet that OpenAI sees Codex, ChatGPT, and developer tooling as one product surface, per DevOps.com’s analysis of the broader coding-agent landscape.

Google I/O 2026 keynote is 24 hours away. Sundar Pichai opens Google I/O on May 19, with Gemini 4.0, Android XR glasses, and Aluminium OS expected as the headline announcements. Google has pre-trailed “the latest Gemini model updates” and a major push on agentic coding, per the broader AI category briefings tracked by llm-stats.

Anthropic’s Project Glasswing puts Claude Mythos Preview in enterprise hands. Anthropic launched a controlled preview of an unreleased frontier model, dubbed Claude Mythos Preview, with AWS, Apple, Cisco, Google, JPMorgan Chase, and Microsoft participating. The program reportedly produced thousands of zero-day vulnerability findings across operating systems and browsers during testing, per Anthropic’s official Project Glasswing announcement.

Trump administration formalizes pre-deployment AI testing access. Microsoft, Google, and xAI have agreed to provide US regulators with early access to frontier AI models prior to public release, codifying a voluntary pre-deployment review process the administration began floating in late 2025. The arrangement parallels and extends the CAISI frontier-AI testing framework I analyzed on May 7, per CNBC’s reporting on the policy move.