Cloudflare opens Quicksilver-backed KV Instant — fast global reads with $100-per-MB storage

On October 1, 2026, Cloudflare introduced Workers KV Instant in private beta. New namespaces can use the Quicksilver-backed 'instant' mode through most existing Workers KV API calls. Cloudflare reports 1.62ms p99 read latency and approximately 256ms p99 replication to its edge locations, compared with 287ms p99 overall reads and 4.38s p99 write replication in its comparison for classic KV. These are Cloudflare's measurements, not independent production benchmarks.

Feature flags, small configuration maps and routing switches can be read on every request with lower latency and faster worldwide update propagation. But KV Instant is economically and operationally unsuitable for ordinary caches, large records or frequently updated data. Its low read price hides unusually high write and storage charges, creating a concrete architecture and billing decision for edge developers.

Classic Workers KV relies on a different read/replication architecture. Instant mode uses Quicksilver, which Cloudflare has long used for internal edge configuration. The headline 100-times-faster claim compares vendor-measured p99 timings; it should not be interpreted as a general 100x application-speed improvement.

For a small namespace read millions of times and rarely changed, the lower per-read cost can make sense. But one thousand write/delete/list operations cost $100, before storage and reads, and a full 1MB stored for a month has a nominal $100 storage charge. Engineers should calculate real write rates, retained bytes and namespace count before opting in.

Namespaces must be created in instant mode, metadata support is absent, list pagination is absent and writes are limited to one per namespace per second. Code relying on getWithMetadata, paged key listing or frequent configuration updates needs changes or should remain on classic Workers KV.