Our primary connection consists of two 10 Gbps fiber circuits, but both were recently cut and the office lost internet for roughly four hours. That was unusual, but it prompted us to investigate a backup connection that would not cost a fortune while sitting mostly unused.
We are considering Starlink or cellular service for at least 30 Windows-based users. Most would be doing normal office work such as web browsing, Excel, and PowerPoint, while a few might be on Teams calls. Supporting 30 users would cover only a small part of our workforce, but it would be much better than having no connectivity at all.
Has anyone used Starlink or cellular as a business-internet backup? Which provider did you choose, how many devices or users have you supported, and what did standby service cost compared with active service?
4 Answers
Cellular can be a good choice if you have strong coverage and the local tower has independent backhaul. Business 5G services from providers such as T-Mobile, Verizon, or AT&T may support external antennas, static addressing, or customer-owned routers. Test more than one carrier at the actual building, because signal strength alone does not tell you how the connection will perform when many customers are using it.
For an office of around 30 people, use a proper 5G gateway or SD-WAN appliance rather than a phone hotspot. Shape traffic during failover, give voice and collaboration applications priority, and disable nonessential synchronization. If possible, testing both a cellular link and Starlink is worthwhile because they fail for different reasons.
Some organizations use a layered setup: a second, physically diverse wired provider first, cellular as the next failover, and Starlink as the last resort. That gives you a fast and predictable backup for ordinary provider problems while preserving an entirely different path for regional fiber or backhaul failures.
Whichever wireless option you choose, confirm the details around data limits, standby billing, public or static IP requirements, antenna placement, weather exposure, and whether your firewall can automatically fail over. Keep the connection installed and test it periodically; a backup that has never been exercised is only a theory.
Before choosing another circuit, verify that the two fiber services are genuinely diverse. They should ideally use different carriers, leave the building at different entry points, and follow physically separate routes. Two circuits can look redundant on paper while sharing the same duct, pole line, upstream node, or backbone. Ask the providers for route documentation rather than relying on the service names alone.
If the outage cost the business more than a backup connection would cost, improving physical diversity may be a better long-term investment than trying to replace 10 Gbps service with a wireless link.
That is a concern here. We discovered that both services use the same upstream infrastructure for part of their route, so we are reviewing the circuit diversity as well as looking at wireless options.
For a true fiber-cut scenario, I would lean toward Starlink rather than cellular. A large terrestrial outage can overload nearby mobile towers, especially if the outage affects power, backhaul, or several providers at once. Starlink avoids that shared local infrastructure and should provide enough capacity for email, web access, collaboration, and selected Teams calls during an emergency. It will not come close to replacing two 10 Gbps circuits, so failover policies should prioritize critical traffic and pause things like backups and large updates.
A standby plan can be relatively inexpensive, and the service can remain available for testing instead of being completely unknown until the next outage.
That has been our experience too. Cellular worked during normal testing but became nearly unusable during regional outages. Starlink has been much more consistent for keeping smaller offices operational.
The modem and antenna matter more than I expected. One carrier's router was extremely sensitive to placement, while another worked reliably with external antennas. I would run a real-world test with several users before committing.