# Why not Kubernetes?
Kubernetes is ubiquitous, but it's not the best fit for all use cases.
__GPUs are scarce and expensive__
This breaks Kubernetes:
* If you need to scale up and there are no GPUs in your region, too bad.
* Running a control plane in _every_ region is expensive.
* The GPU you need can't be found reliably in a single cloud provider, so now you are runnning a control plane in every cloud and every region :(
* If you need to connect regular services to GPU workloads in different regions, Kubernetes is not made for this.
* All open source multi-cluster management solutions have failed.
* Easy to spin up on a single node? Eh, kinda in a limited way.
* Need to secure communication to a GPU on a disparate platform or on prem? Very challenging without complex service meshes.
* Need multi-tenancy in Kubernetes? Good luck.
* Load balance traffic across clouds? Very hard.
=> Kubernetes was simply not __designed for this use case__.
Nebulous is built from the ground up to be __cross cloud__. It keeps the good parts of Kubernetes (declarative, reconciliation, composition) while making a set of primitives which fundamentally assume that resources can be running in any cloud and any region.
Nebulous hedges against the complexity of Kubernetes by making things lighter and more "batteries included".
We aim to be the __best__ solution for running accelerated workloads in the cloud.