The email from the FDA landed at Emulate's offices on September 24, 2024, and it carried more weight than the typical regulatory correspondence. The agency had accepted the company's human Liver-Chip into its ISTAND pilot program—a milestone that might sound incremental but signaled something more fundamental. For the first time, an organ-on-chip technology designed to predict human drug toxicity had crossed that threshold. More significantly, regulators were no longer just observing from the sidelines. They were actively constructing the scaffolding for a future where virtual patients might answer questions that once demanded years of animal studies and billions in traditional clinical trials.
The market, as markets do, had already picked up the scent.
Multiple independent forecasts converge on the same aggressive trajectory. DataM Intelligence projects the in-silico clinical trials market will swell from $3.95 billion in 2024 to $6.39 billion by 2033. Digital twins in healthcare show even steeper expansion—Research & Markets estimates the sector at $2.5 billion in 2025, exploding to $33.4 billion by 2035. That's a compound annual growth rate approaching 30%. And this isn't the kind of hype-driven forecasting that cycles through Silicon Valley every few years. This is what happens when regulatory barriers crumble and economic pressure—mounting costs, slower timelines, ethical scrutiny—leaves the industry searching for exits.
The Regulatory Ground Shifts
December 29, 2022, might not register as a watershed moment outside policy circles, but the FDA Modernization Act 2.0 signed that day quietly rewrote the rules. The law removed the statutory animal testing mandate. Not suggested alternatives. Not pilot programs. It fundamentally altered what drug sponsors could submit in preclinical packages, explicitly blessing microphysiological systems, in-silico models, and advanced in-vitro approaches.
Around the same time, though perhaps with less fanfare, the agency formalized its Model-Informed Drug Development (MIDD) framework. The ICH M15 draft guidance, posted on November 29, 2024, established general principles for MIDD that would apply across both the FDA and the European Medicines Agency. By 2017, the FDA had noted that "almost all NDAs" already contained some modeling and simulation. But formalization matters in ways that aren't always obvious. It creates a common vocabulary. It standardizes how evidence gets documented. It gives sponsors something resembling a roadmap through what had been murky terrain.
The FDA's MIDD Paired Meeting Program—running through fiscal year 2027 under PDUFA VII—now offers sponsors pre-submission dialogues on dose selection, trial simulation, and mechanistic safety modeling. Rajanikanth Madabushi, who has become something of an agency spokesperson on the topic, frames the value proposition with characteristic regulatory understatement: "When successfully applied, MIDD approaches can improve trial efficiency, increase probability of success, and optimize dosing."
For medical devices, the calculus shifted even earlier. The FDA's Center for Devices and Radiological Health issued final guidance on November 21, 2023, for assessing the credibility of computational modeling in device submissions. It aligns to ASME V&V 40, the internationally recognized framework for verification and validation of biomedical computational models. The committee behind that standard received an FDA award in 2019. These aren't academic exercises collecting dust on conference panels. They're the plumbing that makes routine regulatory acceptance possible.
Building the Infrastructure No One Sees
Underneath the modeling applications sits a problem that rarely makes headlines but determines whether any of this works at scale: life sciences data is fragmented, unversioned, and stubbornly resistant to interoperability. Electronic data capture systems don't talk to lab information systems. Imaging platforms operate in their own silos. Omics data lives in separate clouds, jealously guarded by different academic fiefdoms or corporate divisions.
Building digital twins that can actually simulate patient responses requires something more fundamental—canonical, versioned datasets that can feed both simulations and regulatory submissions without requiring an army of data engineers to translate between formats.
That's where a new generation of infrastructure companies is quietly building. Palantir introduced a Quality Management System for life sciences with GxP-qualified instances of Foundry, consolidating clinical and operational data with 21 CFR Part 11 compliance and pipelines supporting OMOP common data models. Flywheel launched its Validated imaging solution in December 2025, a 21 CFR Part 11-aligned platform for regulated imaging workflows spanning trials, device development, and AI training.
NVIDIA's BioNeMo framework and microservices have been adopted across pharma and techbio faster than many anticipated, with integrations into lab informatics platforms and molecular design tools. The GPU and cloud substrate is maturing quickly. Interoperability is accelerating, too. Near-universal EHR penetration in the U.S. now supports FHIR-based APIs, with academic researchers demonstrating FHIR/openEHR mapping tools that can transform unstructured EHRs into FHIR-compliant patient digital twins.
Mantis Biotechnology, which emerged from Y Combinator's Winter 2026 batch, is betting the infrastructure gap represents the real opportunity. Founded in 2025 by Georgia Witchel, the company describes itself as building "the infrastructure powering human-in-computer models." SEC filings from December 3, 2025, show Mantis raised approximately $4.825 million. The pitch: unify fragmented data across motion capture, biometrics, imaging, and training logs to simulate anatomy and physiology, then validate against real-world outcomes. Early messaging positions it as "Palantir for biomedical," promising to link simulated patient anatomy, test data, and requirements to auto-generate verification reports and FDA Q-Submission packages.
It's early—very early. But the thesis aligns with structural needs emerging across the entire stack.
Where the Rubber Meets the Road
At the PBPK modeling layer, Certara's Simcyp Simulator became the first and only software platform to receive an EMA Qualification Opinion on August 4, 2025, for physiologically-based pharmacokinetic modeling in specified contexts. The qualification means sponsors can use Simcyp for drug-drug interaction risk assessments in European Union submissions without re-establishing platform credibility each time—a subtle but significant efficiency gain. The FDA itself renewed over 400 Simcyp and Phoenix licenses between 2020 and 2021. Simulations Plus, another PBPK vendor, holds multiple FDA-funded grants to validate virtual bioequivalence and inhaled product models.
For clinical trial design, Unlearn has built digital twins for control arms in what CEO Charles Fisher calls "Twintelligent RCTs." The company signed a multi-year collaboration with Merck KGaA in February 2022 for immunology trials and maintains active work in Alzheimer's and ALS. Unlearn claims control-arm size reductions exceeding 30% by generating individualized forecasts for patients who would have been randomized to placebo, trained on high-quality historical trial datasets. Fisher is adamant about pre-specifying model weights before trial start to satisfy regulators—a detail that matters when you're asking agencies to accept synthetic patients as substitutes for real ones.
It's a narrow application, substituting some control patients with their digital shadows. But it directly addresses sponsor pain around enrollment timelines and costs, which is why it's gaining traction.
QuantHealth takes a different tack entirely, simulating entire trials using real-world data and knowledge graphs. The platform claims access to over 350 million patient records and retrospective validation across more than 350 studies. Accenture invested in January 2024, followed by Sanofi Ventures in October 2025. That pairing—a consulting giant and a legacy pharma corporate venture arm—tells you something about the buyer profile: companies that need trial design optimization at scale and have the budgets to integrate complex simulation platforms.
On the organ-on-chip side, Emulate's human Liver-Chip acceptance into ISTAND creates a regulatory pathway for qualifying organ chips as Drug Development Tools. Emulate leadership, in statements throughout 2025, has framed the chips as human-relevant alternatives aligned with the FDA's roadmap to reduce animal testing. As of February 2026, the company indicated it was in the "final stage" of ISTAND qualification for drug-induced liver injury (DILI) risk assessment. The NIH's National Center for Advancing Translational Sciences also awarded $31 million in July 2024 to TraCe MPS Centers specifically to push microphysiological systems toward FDA readiness.
The money is following the regulatory signal.
Where It's Already Working
Computational modeling has found earlier traction in medical devices and clinical decision support, perhaps because the decisions are more anatomically constrained and the evidence packages less sprawling than in drug development.
Dassault Systèmes' Living Heart project operates under multi-year FDA research agreements. In February 2025, Dassault unveiled the next phase: AI-powered virtual twins and an ENRICHMENT playbook for assembling virtual evidence packages that can support device approvals and in-silico trials.
HeartFlow's computational fractional flow reserve and plaque analysis platform received FDA 510(k) clearance for its Next-Gen Plaque Analysis on September 22, 2025, with concurrent Cigna coverage expansion. That's the trifecta: regulatory clearance, payer acceptance, and clinical utilization. FEops' HEARTguide, which received FDA De Novo authorization on September 8, 2021, for left atrial appendage occlusion planning, published additional validation studies in 2025 showing improved device size selection and clinician confidence in prospective data.
These aren't pilot projects collecting dust in academic journals. They're revenue-generating products deployed in clinical workflows. The success establishes proof points that computational evidence can clear both regulatory and reimbursement hurdles—two gates that have to open before any technology reaches meaningful scale.
The Obstacles That Won't Disappear Quietly
Standards and regulatory frameworks are improving, but they're far from uniform. Model credibility and validation remain case-by-case exercises in many therapeutic areas. The ASME V&V 40 standard and the Avicenna Alliance's Good Simulation Practice provide frameworks, but applying them requires deep domain expertise and substantial upfront validation work. It's not plug-and-play.
Data harmonization and lineage are operational nightmares. GxP validation of SaaS platforms is expensive and time-consuming, measured in quarters not weeks. Privacy and security add layers of governance, especially under emerging frameworks like the EU AI Act, which entered force on August 1, 2024. Prohibited AI practices became effective February 2, 2025, and high-risk rules—including many AI-enabled medical devices—apply on August 2, 2026, and August 2, 2027. Conformity assessments will largely embed within MDR/IVDR processes for devices, creating yet another documentation layer that must be navigated.
Bias and representativeness remain stubbornly unresolved research challenges. A January 17, 2025, perspective in npj Digital Medicine outlined the core challenges for precision-medicine digital twins: data integration, validation under uncertainty, and sociotechnical barriers around clinical adoption. Building a digital twin is one problem, a hard one. Trusting it enough to make billion-dollar development decisions is something else entirely.
What Happens Next
The trajectory is set, even if the timeline remains uncertain—and uncertain it will likely remain for longer than the breathless market forecasts suggest.
ICH M15 is progressing toward finalization with alignment between the FDA and EMA on review templates. FDA ISTAND qualifications beyond DILI will establish precedent for additional drug development tools involving microphysiological systems and digital twins. The EU AI Act's harmonized standards for AI-enabled devices will shape global compliance strategies, with Notified Body guidance still evolving in real time.
On the commercial side, watch for payer coverage and guideline references for digital-twin-derived evidence expanding beyond cardiology. Legacy pharma partnerships with trial simulation platforms—Unlearn, QuantHealth, Novadiscovery—will indicate whether virtual control arms and trial design optimization become standard practice or remain niche applications confined to a few adventurous sponsors.
The real opportunity might lie in what Deloitte's 2025 Life Sciences Outlook calls the "digital transformation" priority among executives. The report notes that life sciences leaders expect generative AI investments to generate up to 11% value relative to revenue. That's not about point solutions or clever applications at the margins. That's about reconceiving how evidence gets generated across the entire product lifecycle.
If in-silico becomes the default rather than the exception, the companies that win won't necessarily be the ones with the most sophisticated models or the flashiest demos. They'll be the ones that make computational evidence as easy to generate, validate, and submit as wet-lab data is today. That requires infrastructure—data unification, validation pipelines, audit trails, auto-assembled regulatory packages. It requires regulatory fluency and standards expertise that can't be easily acquired. And it requires solving for the 80% of use cases where model credibility is well-understood rather than chasing the 20% where it's still being debated in academic conferences.
The $6 billion market forecast for in-silico trials by 2033 might prove conservative. When FDA inspectors start asking why sponsors didn't use computational modeling for decisions that could have been addressed virtually—and that day will come—the question won't be whether to adopt these tools.
It will be whether your infrastructure is ready. And whether you started building it soon enough.
