The numbers tell a familiar story of healthcare inefficiency: oceans of patient data—lab results, imaging scans, genomic sequences, wearable metrics—most of it marooned in isolated silos, formatted differently, governed by incompatible standards. A cardiologist planning an ablation procedure still can't easily reconstruct a patient's full medical history. Pharma companies design trials by rebuilding data pipelines from scratch each time. Personalized medicine, for all its promise, keeps running into the same fragmentation problem.
Digital twins—virtual representations of patients, organs, or entire healthcare systems that sync with real-world data—have been positioned as the infrastructure to fix this mess. The concept isn't new. Dassault Systèmes has worked with the FDA on heart models since 2014. But something has shifted in recent quarters, perhaps more quickly than even proponents expected. Regulatory pathways are opening. Interoperability standards are maturing, if unevenly. Clinical evidence is beginning to accumulate beyond proof-of-concept demos.
The market, depending on which analyst report you trust, was somewhere around $902.6 million in 2024 and could hit $2.24 billion by 2030. This isn't speculative anymore—or at least, it's operational enough that hospitals are using digital twins to plan tower expansions and manage bed capacity, not just conduct research.
Three Layers, One Problem
The landscape has organized itself into three distinct camps: enterprise software incumbents bringing simulation engines to healthcare, specialized startups attacking narrow clinical problems, and infrastructure plays building the connective tissue beneath it all.
Dassault Systèmes showcased AI-enabled virtual twins for dementia care at CES 2026, framing the technology as central to its healthcare strategy going forward. Siemens Healthineers highlighted its ActExcell Operational Twin at RSNA 2025—a tool designed for hospital capacity planning and cardiovascular procedure modeling, not some distant research initiative. Session listings from the Healthcare Information and Management Systems Society conference in 2026 suggest hospitals are actively deploying these systems.
The startup landscape tells a more surgical story. Unlearn raised $50 million in a Series C in February 2024 to deploy digital twins as synthetic control arms in clinical trials, then launched TrialPioneer in January 2026, an AI-powered tool for upstream trial design. The following month came a partnership with VectorY Therapeutics to apply the approach in an ALS study. Twin Health published results in NEJM Catalyst in August 2025 showing clinically significant diabetes remission compared to standard care, using a metabolic digital twin platform. FEops secured U.S. 510(k) clearance in 2025 for HEARTguide, a patient-specific structural heart simulation tool.
And then there's Mantis Biotechnology. Three people, Y Combinator Winter 2026 cohort, founded in 2025. Mantis isn't building a twin for a specific organ or disease. It's building what founder Georgia Witchel calls a "domain-aware data platform"—one that unifies data across electronic data capture systems, clinical trial management platforms, labs, and omics into canonical, reusable datasets. An SEC Form D filing dated December 3, 2025, shows the company raised roughly $4.8 million of a planned $5.5 million round.
The positioning is revealing. Mantis argues that healthcare's digital twin problem is fundamentally a data lineage and semantics problem, not just a modeling problem. It's an infrastructure bet in a market still figuring out what infrastructure it actually needs.
Three Forces Converging
Three regulatory and technical shifts are accelerating adoption, and they're happening at once—an unusual alignment in healthcare technology.
First, data infrastructure is maturing. Slowly. Unevenly. But undeniably. The Trusted Exchange Framework and Common Agreement (TEFCA) designated its first Qualified Health Information Networks (QHINs) in December 2023. FHIR-to-FHIR exchange pilots began staging last year. The Office of the National Coordinator's HTI-1 Final Rule, finalized in December 2023 and effective in 2024, mandated transparency requirements for algorithmic Decision Support Interventions and pushed FHIR-based data exchange forward. Compliance timelines for USCDI v3 are still rolling out into 2026.
These aren't merely technical standards updates. They're forcing vendors toward explainable, traceable datasets—precisely what digital twin systems require to generate predictions regulators might actually accept.
Second, regulators are signaling openness to in silico methods, a shift that was unthinkable a decade ago. The FDA Modernization Act 2.0, signed December 29, 2022, explicitly allows "nonclinical tests"—including computational models—as alternatives to some animal testing. The FDA's Office of Science and Engineering Laboratories has cataloged validated cardiac electrophysiology models as regulatory science tools since June 2024. In January 2025, the agency released draft guidance on AI-enabled device software functions, setting expectations for continuous lifecycle management.
(Industry publications reported the FDA withdrew its 2017 Software as a Medical Device Clinical Evaluation guidance in January 2026, though no successor document has surfaced yet—leaving companies in a partial regulatory limbo.)
The European Medicines Agency published a reflection paper in 2023 encouraging early scientific advice for AI-enabled methods, with ongoing qualification activity noted in 2025. The EU AI Act, which entered into force August 1, 2024, establishes phased timelines for high-risk AI in medical devices—obligations that will demand documented data lineage, risk management, and post-market surveillance.
Third, clinical evidence is accumulating beyond flashy conference presentations. Twin Health's NEJM Catalyst study, published in August 2025, showed substantial diabetes remission in a primary care setting. A Type 1 diabetes digital-twin-enhanced decision support system improved time-in-range in a randomized controlled trial, results published in Scientific Reports in 2024. Johns Hopkins researchers won a 2025 award for heart digital twin work that validated predictions against invasive measurements—the kind of actionable data that could inform arrhythmia treatment planning.
What It Looks Like in Practice
Unlearn's digital twin approach illustrates both the opportunity and the infrastructure headache. In clinical trials, placebo or standard-of-care control arms are scientifically necessary but prohibitively expensive. Unlearn generates AI-powered "prognostic digital twins" of individual patients based on baseline data—effectively creating synthetic controls that let researchers shrink sample sizes or boost statistical power. The European Medicines Agency has qualified the methodology in specific contexts. The company's January 2026 launch of TrialPioneer extends this upstream, helping sponsors design better trials before enrollment even begins.
The model's Achilles' heel? Data quality. If baseline imaging, labs, and clinical assessments are inconsistent or incomplete, predictions collapse. Unlearn has built pipelines to manage this, but each new trial introduces new data formats, new EDC systems, new ontologies. It's a Sisyphean data-wrangling challenge.
Twin Health operates one layer closer to patients. The company's metabolic twin platform combines continuous glucose monitors, activity trackers, and other wearables with clinical data to generate personalized nutrition and lifestyle recommendations. The NEJM Catalyst study reported medication reductions and remission rates that exceeded usual care. It's a closed-loop system: sensors feed the twin, the twin generates guidance, clinicians and patients act on it, outcomes feed back. When it works, it works well.
Mesh Bio, a Southeast Asia-focused startup that raised a $3.5 million Series A in February 2024, applies a similar chronic disease management model. Q Bio positions its Gemini platform as a consumer-facing digital twin. Predictiv uses genomic data for a direct-to-consumer twin offering. The models vary, but the infrastructure problem remains constant.
Mantis Biotechnology is betting the real bottleneck isn't model sophistication—it's the plumbing. Founder Georgia Witchel previously built physics engines for digital twins in surgical robotics and simulated FDA trials at Louiza Labs. Her argument: life sciences companies burn engineering cycles rebuilding data pipelines for every analytics project, every AI model, every compliance audit. Mantis aims to provide a semantic layer that encodes biological and clinical meaning directly into datasets, with full lineage and versioning, so the same canonical data can power multiple downstream uses.
It's infrastructure in search of a category.
FEops HEARTguide, which secured 510(k) clearance in 2025, represents yet another approach: organ-level simulation for procedural planning. The software uses computational fluid dynamics to model how a structural heart device will interact with a patient's specific anatomy before the procedure. Siemens Healthineers offers comparable cardiovascular planning tools. These are deterministic physics models, not machine learning systems, but they still depend on solving the same data unification challenges—imaging protocols, anonymization workflows, integration with hospital PACS systems.
The Next 24 Months
Two infrastructure trends will likely define the near term, though predicting anything in healthcare technology feels like a fool's errand.
First, watch for consolidation around domain-aware semantic layers. The National Academies of Sciences, Engineering, and Medicine published a consensus report in December 2023 identifying verification, validation, and uncertainty quantification (VVUQ) as critical gaps in the field. Multi-scale modeling—linking molecular, cellular, organ, and system-level simulations—remains largely a research frontier. Data fragmentation persists across nearly every vertical.
Right now, every company building digital twins is solving the same data wrangling problems in slightly different ways. That's inefficient, and markets tend to punish inefficiency eventually. Mantis, by positioning itself as infrastructure rather than application, is betting someone needs to own the canonical data layer. Others may emerge with competing visions. Whether this becomes a winner-take-all platform play or fragments into specialized tools remains an open question.
Second, regulatory clarity will arrive unevenly, as it always does in global healthcare. The FDA's January 2025 draft guidance on AI-enabled devices establishes expectations for lifecycle management, but companies are still waiting for clarity following the withdrawal of the 2017 SaMD Clinical Evaluation guidance. The EU AI Act's high-risk obligations phase in through August 2027. Companies targeting both markets need to architect systems that can satisfy diverging requirements—documented provenance and audit trails for Europe, continuous learning pathways for the U.S. Threading that needle won't be straightforward.
Clinical adoption will follow the path of least resistance, which in healthcare often means avoiding regulatory entanglements. Operational twins for hospital capacity planning and staffing don't require clearance. Neither do trial design tools that don't directly touch patient data. Expect those categories to expand faster than patient-facing or device-integrated twins, where clearance timelines add friction and uncertainty.
The Boring, Unglamorous Test
The broader question is whether digital twins can deliver on the original promise: genuinely personalized medicine at scale, not just for wealthy patients at academic medical centers. Dassault Systèmes demonstrated dementia care twins updated with home sensor data at CES 2026—compelling in a controlled demo environment. Whether it works in a fragmented U.S. health system with misaligned incentives, inconsistent data standards, and Byzantine reimbursement hurdles is another matter entirely.
Perhaps the real test isn't technical sophistication. It's whether the infrastructure layer can solve healthcare's chronic, decades-old inability to make data flow where it needs to go. The companies still standing in 2027 will likely be the ones that figured out how to make the boring, unglamorous plumbing work reliably—day after day, across systems, without breaking.
That's not the kind of achievement that wins headlines at CES. But it might be the one that actually matters.
