Network slicing is described as running several networks on one infrastructure. What it actually provides is the ability to guarantee different service characteristics to different users on shared equipment.
Traditional networks treat traffic broadly alike
Earlier mobile networks offered limited differentiation, mainly prioritising voice over data. Everything else competed for capacity on similar terms.
That works when requirements are similar, and it fails when one application needs consistently low delay while another needs high throughput and tolerates delay easily.
Adding capacity helps both but guarantees neither, because a busy moment still affects everyone sharing the resource.
A slice is an end-to-end allocation
A slice reserves resources across the radio, the transport network and the core, with its own configuration and performance targets.
Devices are assigned to slices based on subscription or application, and the network schedules their traffic according to that slice's parameters rather than treating all traffic identically.
Because the allocation spans the whole path, a guarantee can be made meaningfully rather than only at the radio, where earlier prioritisation schemes stopped.
Virtualisation is what makes it practical
Core network functions now run as software on general-purpose servers rather than as fixed appliances, so additional instances can be created and configured differently without installing new equipment.
Slicing depends entirely on that flexibility, which is why it arrived with the newer core architecture rather than with the radio equipment alone. Networks running an older core cannot offer it.
Management systems then have to track which slices exist, what each was promised and how resources are currently committed, which is a substantial operational addition rather than a configuration change.
Isolation is a matter of degree
Slices share physical infrastructure, so isolation is enforced by scheduling and configuration rather than by separation. A slice cannot exceed its allocation, but the underlying capacity is finite.
Selling guarantees to many slices requires the operator to manage total commitments carefully, since the sum of guaranteed capacity cannot exceed what exists.
Stronger isolation is possible by dedicating spectrum or hardware, which increases cost and reduces the efficiency that made sharing attractive.
The commercial model is the harder part
Technically the capability is available on modern networks. Selling it requires defining service levels, measuring them and handling disputes when they are not met.
It also requires customers whose requirements genuinely differ enough to pay for differentiation, which is a narrower set than the technology's promotion suggests.
This is why deployment has concentrated in specific settings such as industrial sites and public safety rather than appearing broadly in consumer service.