Vercel cuts Sandbox domain lookup latency 18x with regional replicas

Faster routing for sandbox domains

Vercel has made routing to Vercel Sandbox public domains 18 times faster worldwide, according to a post on the company's blog. The improvement comes from changing where domain resolution happens: domains created through the sandbox.domain() SDK call are now resolved from the closest regional replica rather than a single centralized store. The result is that incoming requests reach the process running inside the sandbox with less delay.

The change applies automatically to every request made to a sandbox domain, and Vercel notes that it comes with no pricing changes. Developers do not need to modify code or configuration to benefit.

From one central store to regional replicas

Previously, resolving a sandbox domain meant consulting one centralized store, no matter where in the world the request originated. That design added a round trip whose cost grew with distance: a request arriving far from the store could spend a large share of its time on the lookup before the sandbox process ever received it.

Distributing resolution to regional replicas shortens that path. The lookup now happens close to wherever the request arrives, so the distance penalty largely disappears. Because every request to a sandbox domain passes through this resolution step, the saving applies on each hit, and repeated access patterns accumulate the benefit.

Median lookup latency down from 62ms to 3.4ms

According to Vercel, median domain lookup latency dropped from 62ms to 3.4ms, which produces the headline 18x figure. The largest gains appeared in the regions farthest from the old centralized store: at the 99th percentile, lookups are now up to 112x faster in Sydney (syd1) and up to 146x faster in Cape Town (cpt1).

The gap between the median and tail numbers points to the nature of the fix. It reads less like a general optimization and more like the removal of a geographic bottleneck: regions that the centralized design served worst saw the biggest relative improvements, bringing their lookup times closer to what users near the old store already experienced.

Nothing for developers to change

Vercel says the improvement ships automatically for every request to a sandbox domain, with no pricing changes. For teams already using the SDK, the only visible difference should be faster responses, particularly from regions that previously sat far from the central resolution point.

Why it matters

Latency at the resolution layer is a fixed tax on every request, so a 62ms median lookup was a real overhead for any workload that hits a sandbox domain repeatedly. Cutting it to 3.4ms makes processes running in sandboxes noticeably more responsive.

The change also illustrates a common failure mode in global infrastructure: a centralized component that performs fine near its home region but degrades sharply at the edges. Vercel's regional-replica approach is a standard remedy, and the outsized p99 gains in Sydney and Cape Town show how much room there was to recover.

And because the upgrade requires no code changes and no new pricing, it is a pure developer-experience win, the kind of quiet infrastructure work that rarely draws attention but directly shapes how fast tools built on the platform feel in everyday use.