Hosted by a vendor or running in the plant: where factory software lives

Comparison criteria

Criteria are stated explicitly and neither option is declared a winner: which one fits depends on the constraint that binds hardest in your operation.

Start by asking what the line does when the link drops

This one question separates the functions that can be hosted remotely from the ones that cannot, faster than any architecture discussion. Planning, reporting, document control and analytics generally survive an interruption because nobody is waiting on them minute by minute. Recording an operation, releasing a batch, printing a label, confirming a torque result or authorising a material move are different: if the system is unreachable, the line either stops or starts working on paper, and paper reconciliation afterwards is where traceability breaks. Look at the specific function, establish the tolerable outage, then check whether the offered product actually degrades gracefully. Vendors differ enormously on this and the answer is rarely in the brochure.

Update timing is a governance matter in a qualified plant

Software delivered as a service improves continuously, and continuous improvement is exactly what a validated environment is designed to control. Where a plant must demonstrate that the system used to release product is the version that was qualified, an update applied by a vendor on their schedule creates a genuine problem: either the qualification evidence trails the software, or every release triggers assessment work nobody planned. Serious vendors in regulated sectors address this with staged environments, advance notice and documentation supporting the change. Establish before signing what notice you receive, whether releases can be deferred, and what evidence the vendor supplies. Where those answers are unsatisfactory, local control of the environment is a legitimate requirement rather than conservatism.

Security responsibility moves, it does not disappear

Hosting shifts platform patching, physical security and infrastructure hardening to a supplier whose specialist teams do that work full time. What stays yours is substantial: who has an account, what those accounts can do, how integrations authenticate, which devices reach the service, and how remote access into the production network is controlled. Industrial incidents frequently arrive through remote access and through the bridge between office and production networks, and neither is fixed by where an application is hosted. On-site hosting keeps everything under your control and thereby keeps every one of those duties as well, including the patching that small teams routinely defer. Both routes need segmentation, access discipline and tested recovery.

Frequently asked questions

Can a hosted system control equipment on the shop floor?

Not directly, and reputable architectures do not attempt it. Real-time control stays with programmable controllers and local systems, while a hosted application supervises, records and instructs at a slower cadence through a local gateway or edge component. That gateway is what allows production to continue during an interruption, buffering data until the connection returns. When evaluating a product, ask specifically what runs locally, how much it can buffer, and what happens to the buffer if the outage outlasts it.

Which option costs less over the life of the system?

It depends on assumptions that are rarely stated in a comparison. Hosted costs continue indefinitely and scale with users and sites; local costs front-load and then reappear as hardware refreshes, database licences, infrastructure staff time and the periodic upgrade project. Compare them across a full replacement cycle including a hardware renewal and at least one major version upgrade, and include the internal effort each option consumes. The result often turns on staffing rather than licensing.

How hard is it to move away from a hosted manufacturing system?

Harder than the sales process implies, so establish the exit terms while you still have negotiating leverage. Confirm what export format your data is returned in, whether it includes historical transactions and audit trails rather than only current records, how long the vendor retains it after termination, and what assistance is provided. Production history frequently has to be retained for years under customer or regulatory obligations, so a route to reading it after the relationship ends is a practical requirement rather than a formality.