⬢
The address never moves
Your endpoint is issued once and stays yours through every tier change. Nothing you have put in your own code has to be edited because you grew.
Most platforms let you grow until the day you have to migrate. Here the tier is a setting, not an architecture: the deployment keeps its address, its configuration and its traffic, and the capacity underneath it changes.
No standing charge to begin. A million requests included.
What people put at the edge
Every platform scales. The question is what it costs you to do it — in engineering time, in downtime, and in the meeting where somebody explains why the URL is changing.
Your endpoint is issued once and stays yours through every tier change. Nothing you have put in your own code has to be edited because you grew.
Changing it is an update, not a rebuild. There is no new deployment to cut over to and no window where both exist.
Requests in flight during a change are served. Scaling is not a maintenance window with a nicer name.
What each tier would cost at your current volume is on the screen beside the button, before you press it.
Every tier, what it includes, and what your current volume would cost on each — so the decision to scale is arithmetic rather than a guess followed by an invoice.
On most platforms that day is a project. A new environment, a cutover, a list of places the old hostname is hard-coded, and a weekend.
Your own code at the edge, and model inference at the edge, are the same product with the same tiers, the same console and the same endpoint behaviour. Serving a model near your users should not mean learning a second platform.
Three to begin with, sixty at the top. You choose where, not how many times to ask.
Each tier states its ceiling, so you can tell whether you are near one.
The ladder goes both ways. Quiet quarter, smaller bill, same deployment.
You choose what an AI edge deployment should be able to do. What answers it is our problem, and it can change without anything of yours changing.
We went from the smallest tier to the largest during a product launch, on a Tuesday afternoon, while it was happening. Nobody had to be told.
No. The address is issued when the deployment is created and belongs to it for its whole life, at every tier. Nothing you have put in your own code needs editing because you grew.
The deployment stays as it is and its capacity, region count and concurrency ceiling change. It is not replaced, so there is no cutover and requests in flight are served.
Yes. The ladder is not a sequence you have to climb a rung at a time, and it goes both ways — scaling down is the same single action.
Edge runs your code close to your users. AI edge runs inference close to them. Same tiers, same console, same endpoint behaviour; you pick which when you create the deployment.
The console shows what each tier would cost at your current volume, beside the button. If a different tier would be cheaper for what you are actually doing, it says so and leaves the decision to you.
The startup tier has no standing charge and includes a million requests across three regions. You move up only when it is worth it, and the figure that tells you is on the screen.
Edge computing and inference that scale from startup to enterprise in a single click.
No standing charge to begin. A million requests included.