Concurrency, runtimes, and execution sandboxes

The fundamental difference is where and how execution environments run and how many invocations each environment can handle at once.

Note

For quota- and region-sensitive behaviour, use Lambda Managed Instances as the product reference.

Concurrency model

• Lambda (default): single concurrency — at most one invocation per execution environment at a time.
• LMI: multi-concurrent invocations — one environment can process multiple invocations simultaneously, improving utilisation of underlying EC2 — especially for IO-heavy apps (AWS calls out web services and batch-style work).

Multi-concurrency changes how you think about thread safety, global state, and context isolation (Concurrency model).

Supported runtimes

LMI does not support every default-Lambda runtime. The supported language floors and multi-concurrency behaviour are defined on the product pages; the table below matches the Supported languages list as of that document.

Language	Minimum (per AWS)	Example runtime values	Multi-concurrency (per AWS)
Java	Java 21 and later	java21, … (newer managed identifiers as AWS adds them)	One process, OS threads — handler must be thread-safe
Python	Python 3.13 and later	python3.14, python3.13, …	One process per concurrent request — watch shared resources such as /tmp
Node.js	Node.js 22 and later	nodejs22.x, nodejs24.x, …	Worker threads + async; safe shared state
.NET	.NET 8 and later	dotnet8, …	Tasks / async; safe shared state
Rust	OS-only provided.al2023 and later	provided.al2023, …	One process, Tokio tasks — handler Clone + Send

Exact runtime strings, region availability, and new language versions live in the language topics and API reference. Use the LMI runtimes hub (and Next steps there) before you pin a runtime value in config.

Caution

Multi-concurrency changes isolation semantics versus default Lambda. Whether invocations share a process, threads, or separate processes depends on the runtime (see the table above and the language topics linked from the hub). Treat global state, caches, connections, and /tmp accordingly before you migrate or raise concurrency.

See also

Tenancy & isolation Shared Lambda infrastructure and Firecracker versus Nitro-backed dedicated capacity and trust boundaries.

Elasticity & CPU Managed instances versus execution environments at idle and under CPU-driven scaling.

Decision guide & reference Full comparison table, guardrails, and reference material for default Lambda versus LMI.