Cloudflare puts AI Search, Basin, K2 and Clef into production

Cloudflare's Birthday Week has been running as a daily sequence of announcements, and Day 4 concentrated on data rather than the Workers runtime and management topics that dominated Day 3. The rollout spans a storage mode (KV Instant), an ingestion-and-analysis platform (Basin, formerly split across Pipelines, R2 Data Catalog and R2 SQL), and a streaming layer (K2) that began life as an internal buffer for Basin Pipelines ingestion.

The common thread is that Cloudflare is tying these pieces together rather than selling them as standalone tools. Basin stores data in Apache Iceberg, so external tools like DuckDB and Spark can read the same data, and R2 charges no egress fees — a combination that may make it easier for teams to shift analysis work that currently lives in systems such as BigQuery. AI Search's generally available version extends image search to vectorize images directly via Qwen3-VL-Embedding, and raises text and OCR'd PDF ingestion to 10 MiB, though other formats stay at 4 MiB.

Where this lands depends on cost and maturity. AI Search and Artifacts both move to paid billing in the coming weeks, and K2's beta limits storage to 10 GB per account with a default 7-day retention and an initial write latency around 1 second at P99 — so it is likely better suited to bulk event movement than to latency-sensitive paths. Developers already on Cloudflare Workers appear best positioned to benefit, while Kafka client compatibility and push delivery to Workers consumers remain on the roadmap.