AWS Lambda, Azure Flex, and Cloudflare Workers compared. Cloudflare leads in TTFB, while AWS dominates heavy compute and tiered concurrency. Full breakdown.
What each platform optimizes for
Serverless performance is not a single metric. Cold starts, steady-state latency, CPU-bound work, and how concurrency scales all pull in different directions. AWS Lambda, Azure Flex, and Cloudflare Workers sit on three different runtime models, so the right choice depends on whether your workload is edge-facing and short-lived, or region-bound and compute-heavy.
Cloudflare Workers run close to end users on a lightweight isolate model. That design favors request path latency and keeps startup cost low for small handlers. AWS Lambda and Azure Flex run in regional data centers with fuller language runtimes and more generous compute envelopes. They trade a bit of first-byte speed for the ability to hold heavier work, larger dependencies, and longer execution windows.
TTFB and the request path
Time to first byte is where Cloudflare typically leads. Workers sit on a global network, so a request often hits a nearby location without a long cold path into a distant region. For APIs, personalization stubs, auth gates, and thin BFF layers, that edge placement can matter more than raw CPU power. Keep the handler small, avoid large package trees, and push heavy lifting to a backend when the edge path must stay fast.
AWS Lambda and Azure Flex can still deliver solid TTFB when functions are warm and deployed in the same region as clients or upstream services. Provisioned or always-on capacity, connection reuse, and keeping init work out of the request path reduce p99 spikes. They rarely match pure edge TTFB for global traffic, but they win when the response depends on regional data stores, private networks, or multi-step orchestration that does not belong at the edge.
Heavy compute and concurrency models
AWS dominates heavy compute and tiered concurrency. Lambda supports longer-running, memory-heavy jobs, broader runtime choices, and concurrency controls that scale from account-level limits down to reserved and provisioned capacity per function. That tiering lets you protect critical paths, absorb bursts, and keep expensive packages warm without treating every function the same. Use it for image or media transforms, batch-style processing, and services that need predictable parallel execution against regional resources.
Azure Flex targets a similar middle ground: flexible scaling for HTTP and event-driven work with a managed runtime that feels familiar to teams already on Azure networking and identity. It is a strong fit when the rest of the stack—storage, queues, private endpoints—already lives in Azure and you want regional compute without managing servers. Cloudflare Workers remain best for short, highly parallel request handlers; offload multi-second CPU or large memory jobs rather than forcing them into an edge isolate.
- Edge-first, low TTFB: Cloudflare Workers for thin, global request handlers.
- Heavy compute and controlled scale: AWS Lambda with reserved or provisioned concurrency on hot paths.
- Azure-centric apps: Azure Flex when regional integration and platform consistency matter most.
- Hybrid pattern: edge auth or routing on Workers, heavy work on Lambda or Flex in-region.
How to choose without chasing benchmarks
Start from the critical path. Measure where users and data live, how long a request may run, and whether cold starts or concurrency storms hurt you more. Prefer Workers when the product value is global latency and the function stays small. Prefer Lambda when you need tiered concurrency, heavier CPU or memory, and deep AWS service integration. Prefer Azure Flex when your identity, networking, and data plane already sit in Azure and you want managed scale without a second cloud’s operational model.
Design for the failure modes each model exposes: edge CPU limits, regional cold starts, and concurrency ceilings. Keep payloads small at the edge, reuse connections in region, and separate “fast path” handlers from “heavy path” workers. Performance then becomes an architecture choice—not a single leaderboard score—aligned with how AWS, Azure Flex, and Cloudflare Workers actually run code.