A decade back, "digital twin" was jargon for aerospace engineers tinkering with jet engines and factory managers optimizing assembly lines. Now? Regulators in Brussels and Washington are signing off on virtual patients—computational constructs that have never drawn breath, never felt a needle prick, never filled out a consent form. These simulated humans, calibrated from mountains of real-world data, can test drugs, flag side effects, and potentially shave years off development timelines.
The shift happened quietly. In December 2022, the FDA Modernization Act 2.0 struck down the legal mandate for animal testing in drug development, cracking open a door to in-silico alternatives. By 2032, forecasters say, the market for these virtual trials will hit $7.25 billion.
Those numbers suggest more than a research curiosity. But the foundational question—can a model genuinely replace flesh and blood?—remains one the industry is still figuring out, study by study, submission by submission.
Where the Market Stands
The in-silico clinical trials market clocked in at $3.60 billion in 2024, according to Grand View Research. Projections point to $7.25 billion by 2032—a compound annual growth rate of 9.5%. Medical device applications led last year's tally; pharma, characteristically slower to move, is accelerating hard behind it. The broader healthcare digital twins market, which folds in personalized medicine, continuous patient monitoring, and device design, sat at $0.90 billion in 2024. By 2030, analysts expect $3.55 billion, a CAGR approaching 26%.
North America still dominates. Europe, though, is closing the distance fast, propelled by regulatory tailwinds and public funding that American researchers often eye with envy. The European Commission's Life Sciences Strategy earmarks funds for "virtual human twin" incubators. The EU Biotech Act proposal—floated in December 2025—explicitly recognizes new approach methodologies (NAMs) as legitimate alternatives to animal testing. Across the Atlantic, the FDA's Model-Informed Drug Development (MIDD) Paired Meeting Program, active through June 2027, signals institutional buy-in. The program offers quarterly windows for sponsors to discuss dose selection, trial simulation, mechanistic safety evaluation. In plain terms, the agency is inviting companies to walk in with computational evidence instead of just preclinical readouts.
Yet regulatory acceptance doesn't guarantee widespread adoption. A 2025 scoping review published in Nature Digital Medicine found that only 12.1% of 149 studies claiming to use "human digital twins" actually met the definition aligned with National Academies of Sciences, Engineering, and Medicine (NASEM) standards. The term gets thrown around casually. The standards remain under construction. And the gap between industry hype and scientific rigor? Still wide.
The Forces Driving Change
Three currents are converging: regulatory harmonization, mounting trial inefficiencies that no one can ignore any longer, and computational infrastructure that has finally caught up to ambition.
Start with regulation. The FDA finalized guidance in November 2023 on assessing the credibility of computational modeling and simulation for device submissions—a risk-informed framework aligned with ASME V&V 40 standards. On the drug side, the International Council for Harmonisation's M15 guideline on model-informed drug development gains legal force in the EU on July 23, 2026. The European Medicines Agency went a step further in August 2025, issuing a qualification opinion for Certara's Simcyp PBPK platform—the first software platform to secure such endorsement for specific drug-drug interaction contexts. That precedent matters. It lightens the validation burden for sponsors using pre-qualified platforms, potentially paving the way for similar endorsements across other modeling domains.
The inefficiencies fueling demand are documented exhaustively, if not always with perfect precision. Industry surveys report that roughly 85% of clinical trials experience delays. The Tufts Center for the Study of Drug Development cites a 32% uptick in certain cycle-time metrics tied to external data burdens. An Oracle survey from 2018 found 57% of researchers blaming data issues for holdup. Deloitte's 2025 update pegged the average cost per approved asset at $2.23 billion—a figure academics quibble over but one that captures the financial stakes involved.
Georgia Witchel, founder of Y Combinator-backed Mantis Biotechnology, claims that 80% of clinical trials are delayed by data inaccuracies, with an average $15 million lost per trial due to data quality problems. Whether those numbers hold under forensic scrutiny, the underlying challenge is undeniable: trial data fragments across electronic data capture systems, clinical trial management platforms, lab instruments, and omics repositories. Lineage is often murky. Queries run slow. Audits turn painful. The chasm between messy data silos and computational models demanding clean, traceable inputs has become a chokepoint.
Meanwhile, computational power caught up. NVIDIA's Omniverse platform positions itself as infrastructure for real-time physics simulations across industries; healthcare vendors are adopting it for device and surgical digital twins. High-performance computing consortia like EuroHPC are running virtual trials to assess pro-arrhythmic drug risk at scale. Open-source tools such as Stanford's OpenSim for musculoskeletal modeling coexist with commercial platforms—AnyBody, Dassault Systèmes' Living Heart Project—that have collaborated with the FDA for five years on using cardiac simulations to reduce animal and patient testing for device approvals.
Who's Building What
Unlearn.ai has perhaps the clearest regulatory validation story to date. The company's PROCOVA method, which uses AI to generate "digital twins" of control patients, secured a qualification opinion from the EMA for use in Phase 2/3 trials with continuous outcomes. The FDA offered positive feedback for covariate-adjusted analyses using prognostic scores and digital twins. Unlearn's blog points to scenarios where control arm sizes in Alzheimer's trials could shrink by up to 40%—potentially meaningful when recruitment drags and cohorts prove hard to assemble.
Virtonomy, a German startup that raised €5 million in Series A funding last December, focuses on device-centric virtual trials. The company generates digital patient populations from CT scans, then runs simulations on NVIDIA-accelerated pipelines to predict device performance before any human testing begins. It's a niche play at the intersection of imaging, statistical shape modeling, and regulatory science—a niche that grows more relevant as device regulators warm to computational evidence.
Dassault Systèmes' Living Heart Project sits further along the maturity curve. The multi-stakeholder initiative, spanning academic and industry partners, extended a five-year FDA collaboration focused on using in-silico heart models to evaluate pacemaker leads and valve devices. The project sketches out a pathway from research tool to regulatory evidence: models validated in tightly defined contexts, with clear documentation of assumptions, uncertainties, and limits.
Novadiscovery's Jinkō platform takes a different angle, simulating protocol designs and virtual populations to explore oncology and infectious disease scenarios. The platform emphasizes SBML compatibility and provenance—recognizing that traceability isn't optional when computational outputs start informing regulatory submissions.
Then there's Mantis Biotechnology, launched from Y Combinator's Winter 2026 batch. The company bills itself as "Databricks for biomedical and clinical data," building a domain-aware platform that unifies motion capture, sensors, imaging, and trial system data into versioned, queryable datasets. Founder Georgia Witchel—who previously started Louiza Labs, a surgical digital twins physics engine—positions Mantis as infrastructure for validated digital humans. The pitch: transform fragmented inputs into physics-based or hybrid simulations, validate against real-world outcomes, generate submission-ready packages with full lineage. The team of three is hiring for roles bridging simulated patient anatomy, test data, and requirements. It's an early-stage play at the intersection of data engineering, simulation, and regulatory compliance.
The contrast between these companies is instructive. Unlearn leans into statistical and AI methods with regulatory traction already in hand. Virtonomy and Dassault are physics-heavy, device-focused. Novadiscovery occupies systems biology and protocol optimization territory. Mantis is aiming at the data layer beneath them all—the canonical datasets and lineage infrastructure that credible models require. Whether that positioning holds depends on execution and whether the market values infrastructure separately from the models themselves.
What Comes Next
The next eighteen months are pivotal. ICH M15 takes legal effect in the EU in July 2026, codifying global expectations for model-informed drug development. The FDA's MIDD program runs quarterly submission windows through mid-2027, providing a visible timeline for sponsors testing the waters. The EU AI Act's high-risk obligations for medical AI—data governance, logging, transparency, human oversight—phase in between 2026 and 2028, intersecting with existing device regulations (MDR/IVDR). The HIPAA Security Rule modernization, proposed in December 2024, could tighten requirements for platforms handling electronic protected health information, raising the compliance bar for data infrastructure providers.
Regulatory momentum is one thing. Technical maturity? Another entirely. The computational biology community has produced frameworks—ASME V&V 40 for verification and validation, FDA CDRH's risk-informed credibility guidance, EMA's PBPK reporting standards—but applying them at scale remains non-trivial. Models depend on assumptions. Real patients are messier, less predictable than virtual populations. A 2025 consensus statement in npj Digital Medicine emphasized that digital twins should augment clinicians, not replace them, and that adoption conditions and concerns still need addressing.
The terminology problem persists, too. The Nature Digital Medicine scoping review's finding that only 12.1% of studies correctly used "human digital twin" underscores the risk of dilution. Standards bodies and consortia are grinding away at definitions, but the field moves faster than consensus processes.
For founders and investors, the opportunity sits squarely at the intersection of three trends: regulators increasingly accepting computational evidence, trial sponsors desperate to compress timelines and slash costs, and advances in AI and HPC making previously intractable simulations suddenly feasible. The question is where value actually accumulates—in the models themselves, in the platforms running them, or in the data infrastructure making them credible.
Mantis and companies like it are betting on infrastructure. If they're right, the next generation of digital humans won't just be better simulations—they'll be better documented simulations, with lineage traceable from raw sensor feeds to submission packages, queryable across studies, validated against outcomes that regulators actually care about.
The $7 billion market forecast assumes in-silico trials move from pilot programs to standard practice. That shift hinges less on whether virtual patients can work in theory—they probably can—and more on whether the industry can build the unglamorous plumbing required to make them trustworthy in practice. Data pipelines. Validation frameworks. Audit trails. The infrastructure that never makes headlines but determines whether breakthroughs scale or stall.
In other words: the science may already be there. The question now is whether the scaffolding can support it.
