A digital twin shouldn't live in someone else's cloud
Escherbridge
A digital twin is supposed to be a working model of a place you can stand in. A surprising number of them are a login. The building is yours. The account is not. When the contract ends, or the region blips, or the price card changes, the twin stays where the invoice goes.
That is not a metaphor about “the cloud.” It is a portability problem with a street address.
The platforms admit the shape of the trap
Researchers who designed an open digital-twin platform called KTWIN say it plainly: most existing platforms are tied to one vendor, and moving to another takes "a considerable amount of rework." Because those platforms live almost entirely in the vendor's data centers, there is also little room to run the model on or near the site itself. Their answer is dull in the useful way: open standards, and a setup you can run on your own premises or with a provider you picked.
Twin2Clouds, a 2025 cost model for twins spread across providers, says the quiet part at the start: putting the whole twin inside one provider “further risks vendor lock-in, constraining flexibility and negotiation power.” The authors’ own scenarios — a large smart building down to a smart home — are where they report lower monthly cost against single-cloud baselines. The number belongs to their evaluation, not to a brochure. The structural point does not need their percentage: a single provider is a pricing conversation you can only have with one party. DOI record.
Security people are not asking for a prettier dashboard
NIST IR 8356 (February 2025) treats a digital twin as a system with ordinary security problems and a few new ones: confidentiality, integrity, availability, and the chance that an operator is looking at a facsimile while the controls are being lied to. The report’s advice is not “buy the suite.” It says any serious effort should follow the NIST Risk Management Framework, the Cybersecurity Framework, and the Privacy Framework, and it warns that proprietary encryption “can be weak and lack thorough vetting.” Hashes and error detection, it adds, still have a job even when the encryption is real — including on the path from devices to “the central digital twin definition repository.” A twin with one repository and one vendor’s crypto is a trust decision, whether or not the sales deck called it one. NIST IR 8356.
McKinsey’s note on when a twin is worth having is similarly unromantic. The prerequisites it names are a viable business case, high-quality data, and ongoing operational demand. Miss those and the model does not become wise. It becomes stale, which is a polite word for wrong. McKinsey Tech Forward.
What “someone else’s cloud” actually costs
- Moving the twin means rework, not an export button. KTWIN’s survey of PaaS offerings is explicit about that.
- The edge is where the site is. A cloud-only twin is a long way from a feeder, a floor, or a field that has lost its uplink.
- The security boundary includes the vendor’s repository, the vendor’s crypto, and whoever can edit the facsimile the operator trusts.
- Price is a feature of the lock-in. Twin2Clouds exists because single-provider pricing is hard to argue with from inside the account.
None of this says “never use a cloud.” It says the twin should still be yours when the cloud is a bad day, a bad contract, or a region you are not allowed to use. Ground truth with a receipt is more than most digital twins can say. A receipt you cannot take with you is a screenshot.
The habit, and a better one
The habit is pilot purgatory with a cloud login. A twin gets stood up for a demo, the data never quite leaves the vendor account, and the “pilot” becomes the product because moving it would mean rebuilding it. That is not a moral failure. It is what you get when the export path was never part of the design. The better path is dull: a model you can run beside the site, a hash for every blob, and a contract that assumes you will leave. The cloud can still do the heavy rendering. It should not be the only place the building exists.