Breaking Lab srl
Why slow storage kills fast apps
When you build a web application, speed usually comes down to two things: your code and your data. Cloudflare Workers — a platform that runs small pieces of server code at hundreds of locations around the world — has always been fast for the code part. The data part was harder. Workers KV, its key-value store (think of it as a giant, distributed dictionary where your app can look things up), had a hidden penalty: the first time you read a value at a location that hadn’t seen it recently, you’d wait while it fetched a fresh copy. That cold-read delay could add tens or hundreds of milliseconds to a response.
What’s actually changed
The new Workers KV Instant tier is built on something Cloudflare calls Quicksilver, an internal system that keeps data in sync across its 300-plus edge locations. The headline number is a p99 read latency — meaning 99 out of 100 reads — of under two milliseconds, with changes reaching every location within 250 milliseconds. In practical terms: no more cold-read penalty, no more waiting for a distant data centre to catch up. The API stays the same, so existing Workers code needs no changes to benefit.
Who should care about this
If you host a static site or a simple blog on Cloudflare, this probably changes nothing for you today. But if you use Workers to do anything data-driven — feature flags that switch behaviour per user, rate limiting to block abusive traffic, session tokens, personalisation rules — latency at the storage layer has real consequences. A 100ms delay on every authenticated request adds up fast when you have thousands of visitors. Sub-2ms reads mean that lookup is effectively free in human terms.
It also matters for the growing category of AI-powered tools and agents running on edge infrastructure. These often need to check state — has this user already asked this question today? — before doing anything expensive. Fast, cheap state lookups make those checks practical rather than a bottleneck.
What to watch for
Cloudflare hasn’t published the full pricing breakdown yet, and “Instant” is clearly a premium tier on top of the existing KV offering. The real question for anyone building on Workers is whether the cost premium over standard KV is justified for their specific workload. For high-traffic, latency-sensitive applications it almost certainly is. For low-volume hobby projects, the existing tier will probably remain the sensible default. Either way, having a genuine sub-2ms option at global scale without managing your own caching infrastructure is a meaningful shift in what’s practical to build on serverless platforms.
