
Every tech brochure now promises a "digital twin" of something. Your factory, your city, your hospital, possibly your left ventricle. The phrase sounds like science fiction and is sold like a software subscription. So what is it, really, and which of these twins actually exist?
The short answer: A digital twin is a software model of a real object or process that stays linked to the real thing through live data. Take away the data link and it is just a simulation. Real twins now run power grids, car plants and hospital estates. Patient twins are still mostly research, and the hard part is never the 3D graphics. It is the plumbing.
What counts as a digital twin
The Digital Twin Consortium revised its definition in October 2024, and the key word is synchronisation. TechTarget's April 2025 explainer boils it down to three parts: a physical thing, a software model of it, and data flowing between the two. The model does not even need a picture. A spreadsheet fed by sensors qualifies.
That rules out a lot of things wearing the label. A CAD model is a drawing. A simulation runs once, on assumed inputs, then stops. A twin keeps updating as the real object ages, breaks or gets rebuilt. The idea is older than the name: engineers trace it to NASA using telemetry to diagnose the damaged Apollo 13 craft in 1969, and the term itself dates from 2010.
| What you are looking at | Live data from the real thing? | Verdict |
|---|---|---|
| 3D model of a building from the architect | No | Model, not a twin |
| One-off simulation of a new production line | No, assumed inputs | Simulation |
| Grid model updated from substation sensors | Yes | Digital twin |
| Hospital floor map with occupancy and survey data | Yes | Digital twin |
| Google Maps with traffic | Yes, from phones | A data twin of the roads, arguably |
Where twins really run in 2026
Factories. BMW planned its Debrecen plant entirely in a virtual model about two years before the real one opened, as NVIDIA described in March 2023. Siemens said in January 2026 that twins at PepsiCo's US plants raised throughput by 20 percent and cut investment costs by 10 to 15 percent. Siemens is also simulating its whole new Amberg factory before building it, with completion due in 2030.
Power grids. Singapore's grid twin, announced in October 2021, has two halves: an asset twin that watches the health of more than 18,000 transformers and 11,000 substations, and a network twin that tests what happens when EV chargers and rooftop solar plug in. SP Group's stated aim is fewer physical inspections.
Cities. Virtual Singapore, the city-state's 3D twin built with Dassault Systèmes, is used for flood analysis, solar roof planning, pedestrian shade and even tracking dengue clusters, according to GovInsider. Dassault, which has an obvious interest, put the budget at $73 million.
Hospitals. Manchester University NHS Foundation Trust switched on a twin of six hospitals, 274,000 square metres, in late 2025, Building Better Healthcare reported. It flags under-used rooms and runs asbestos and RAAC surveys from phones. The trust says asbestos paperwork at one hospital fell by up to ten days a month. Four more hospitals come next.
The twin of you is not ready
Patient twins get the most breathless coverage and the least deployment. The EU's Virtual Human Twins initiative has over 100 million euros behind it, and the Commission started procuring a shared modelling platform in June 2025. That money is going into research models of organs and disease, not a doctor pulling up your twin at the next appointment. Expect decades, not quarters.
What it costs and why it fails
Vendors rarely publish price tags, and the ones that exist are project-sized: a national city model in the tens of millions, a hospital estate twin built on top of software the trust already licensed. Grand View Research sized the global market at $35.8 billion in 2025 and forecasts $328.5 billion by 2033. Forecasts like that sell reports; treat them as a mood, not a budget.
The money mostly goes on unglamorous things. Sensors have to be installed and maintained. Data from decades-old systems has to be cleaned and matched. Someone has to own the twin after the launch party. A October 2025 post-mortem by ViaTechnik lists the usual killers: no measurable goal, a brittle data foundation with no owner, and nobody planning for the people side. Intel's March 2026 brief adds a blunter one: teams buy the pretty visualisation first and the architecture never.
Four questions before you believe a twin pitch
- What data feeds it, how often? If the answer is "we upload a file quarterly", it is a model with a marketing name.
- Which decision changes? Fewer inspections, faster line changes, rooms reassigned. Pick one and measure it.
- Who keeps it true? A twin that nobody updates drifts from reality within months.
- Whose numbers are those? The 20 percent gains above come from vendors and customers, not auditors. They may be real; they are not independent.
Frequently asked questions
Is a digital twin the same as a simulation?
No. A simulation answers "what if" using inputs you typed in, then ends. A twin keeps answering "what is", because real sensors keep feeding it, and it can run simulations on top of that live state. Every twin contains simulations; most simulations are not twins.
Do small businesses need one?
Almost never as a product. If your equipment already reports its condition to a dashboard, you have the useful part. The big twins pay off where physical mistakes are expensive: a plant layout, a substation, a city block. For most firms, the sensible first step is checking vendor claims the way you would fact-check any confident answer: ask for the source.
The useful test is dull and reliable. Ask what feeds the model and what decision it changed last month. If both answers are specific, you are looking at a working twin. If either is vague, you are looking at a very nice render.
Conversation
Have a question or correction? Contact Rvasus.