Multi-region deployment used to be the customer’s problem. Pick a region, hope it’s the right one, accept that users on the other side of the world get a worse experience.
This week we shipped multi-region by default for every paid plan. Here’s what changed and why we think this is the new baseline for AI infrastructure.
The old story
For a long time, AI workloads ran in a single region — usually US-East. Latency was acceptable for early adopters, who tended to be on the same continent as the workload anyway.
That stopped scaling the moment we picked up customers in Europe and APAC. P50 latency for them was double what we showed in our marketing.
What we built
Aurora now routes every request to the nearest of twelve regions automatically. Customers don’t pick a region; the system picks for them based on the user’s IP. Failover is automatic, the cost is folded into the plan, and the dashboard shows per-region health in real time.
We considered making this a paid add-on. Then we looked at our European and APAC customers and decided it was table stakes, not an upsell.
What’s next
Edge inference is on the roadmap for Q3. Same idea, smaller footprint, even closer to the user. We’ll write more when we have something to show.
/ JOIN THE CONVERSATION
Leave a comment.