Cloudflare Workers — source summary

The official Workers developer docs, the primary-source counterpart to the Wikipedia-based cloudflare provider overview. Where that page places Cloudflare as the spoke's edge-compute model, this one documents the platform a developer actually ships on: "a serverless platform for building, deploying, and scaling apps across Cloudflare's global network with a single command." T1 vendor docs — authoritative on what's offered, with the usual vendor caveat on the pricing and "affordable at any scale" framing.

The compute model

Workers is serverless, but not the container-or-VM-behind-an-API shape of AWS Lambda. Code runs at the network edge across Cloudflare's global footprint, so latency comes from proximity rather than from a region you picked and a machine you sized. There is no instance to provision and billing is per-request. Languages: JavaScript, TypeScript, Python, Rust, "and more," with first-class support for full-stack frameworks — React, Vue, Svelte, Next.js, Astro, React Router — so a framework build deploys without a server to manage.

The ecosystem (why it's more than functions)

The docs frame Workers as a hub with storage and compute primitives attached, which is what separates it from a bare function runtime:

• Storage / state — Durable Objects (stateful coordination, single-instance consistency), D1 (serverless SQL), KV (eventually-consistent key-value), R2 (object storage, the no-egress-fee pitch that is the sharp foil to fly-io's metered egress), Hyperdrive (connection-pooling / acceleration for external databases).
• Compute / AI — Workers AI (GPU-backed inference at the edge), Vectorize (vector database), Workflows (durable long-running operations), Browser Rendering.
• Delivery — CDN/cache and image optimization, since the same network is already a CDN.

What it costs (re-verified 2026-09-22, unchanged since 2026-08-18)

The overview page above sells the platform; the numbers live on the pricing page (https://developers.cloudflare.com/workers/platform/pricing/), read on 2026-08-18:

• Free: 100,000 requests/day, 10 ms CPU per invocation. The CPU ceiling is the real limit — request count is generous, per-invocation compute is not.
• Paid: $5/month minimum, 10 million requests and 30 million CPU-milliseconds included, then $0.30 per million requests and $0.02 per million CPU-ms. Max CPU per invocation rises to 5 minutes (30 s by default).

One dated change worth holding against this spoke's silent changes to permanent free tiers edge: Cloudflare announced that storage billing for SQLite-backed Durable Objects starts January 2026, target 7 January, and that Workflows step and storage usage will not be billed before an announced start date. Both are paid surfaces arriving where free ones were — but announced in a changelog with a date attached, which is the opposite of the Oracle behaviour that edge was written about. The edge asks for a second vendor quietly halving a permanent free tier; this is a vendor doing the loud version, and it belongs on the page as the contrast case rather than as a second instance.

Why it matters here

This deepens the spoke's third hosting model. The cloudflare page established edge-per-request as a category alongside raw IaaS (hetzner-cloud) and managed PaaS (render/railway/vercel); the docs show the category has grown a full data plane (D1, KV, R2, Durable Objects, Vectorize), so "edge compute" is no longer just stateless functions at the edge — it's a place to run a stateful app end to end. Workers AI continues to answer the spoke's open question on where AI/real-time stacks run. Note this wiki is itself hosted on Cloudflare Pages, a sibling of Workers.

Related

cloudflare · vercel · fly-io · render · railway