AWS Lambda Managed Instances: Scaling with EC2 Efficiency

AWS has long offered Lambda as a fully managed compute service where each invocation triggers a fresh execution environment. This approach excels for event-driven workloads but faces limitations with memory-intensive or multi-concurrent applications. Enter Lambda Managed Instances (LMI), a game-changing alternative announced at re:Invent 2025 and enhanced in March 2026 with support for up to 32GB memory and 16 vCPUs. Unlike traditional Lambda, LMI runs functions on EC2 instances within your VPC while handling provisioning, scaling, and patching automatically.

Why Lambda Managed Instances Matters for Modern Workloads

Traditional Lambda functions process requests one at a time per execution environment. When handling 1,000 concurrent requests, AWS spins up 1,000 isolated environments—each with its own memory allocation, cold start penalty, and per-GB-second billing. This model works well for sporadic workloads but becomes inefficient for sustained, compute-heavy applications.

Lambda Managed Instances changes this paradigm by running multiple concurrent requests within the same execution environment on EC2 instances. The benefits include:

• - Configurable memory-to-vCPU ratios up to 32GB/16vCPU
• - Access to EC2 instance types like Graviton4 for specialized workloads
• - Reduced operational overhead through automated provisioning
• - Potential cost savings at scale compared to per-invocation pricing

Consider a product similarity engine that loads a catalog of embeddings into memory. Standard Lambda would struggle with the memory footprint and CPU demands of vector comparisons. LMI, however, can allocate sufficient resources while processing multiple requests in parallel within a single environment.

A Practical Implementation: Product Similarity Engine

To demonstrate LMI’s capabilities, I built a product similarity engine using:

• - Embeddings: Nova Multimodal Embeddings via Amazon Bedrock for query processing
• - Vector math: Cosine similarity calculations using ThreadPoolExecutor for parallel processing
• - Infrastructure: Terraform for provisioning and configuration
• - Language: Python 3.14 with AWS Lambda Powertools for observability

The handler loads product catalogs with pre-computed Nova embeddings, processes incoming search queries by generating embeddings on-demand, and performs similarity comparisons across categories. This workload benefits from LMI’s multi-concurrency support and configurable hardware, as it combines I/O-intensive API calls with CPU-heavy vector operations.

Navigating the AWS Compute Continuum

AWS offers a spectrum of compute options, each suited to different workload patterns. Here’s how Lambda Managed Instances fits into this landscape:

Scaling and Concurrency Comparison

| Service | Scaling Approach | Concurrency Model | Pricing Model | Cold Start | |---------------------------|------------------------------------|-----------------------------|--------------------------------------------|----------------------| | Standard Lambda | Per-invocation, instant | 1 per environment | Per-request + GB-second | Milliseconds-seconds | | Lambda Managed Instances | CPU-based, concurrency saturation | Multiple per environment | Per-request + EC2 + 15% management fee | Tens of seconds | | ECS Express Mode | Traffic-based, auto-scaling | Configurable | Fargate + ALB | Minutes | | ECS Fargate | Task-based | Configurable | Per-vCPU-hour | Minutes | | EKS | Pod-based | Configurable | EC2/Fargate + control plane | Minutes |

Key Differences and Use Cases

Lambda Managed Instances shines in scenarios where:

• - Workloads require sustained throughput (hundreds or thousands of requests per second)
• - Memory needs exceed Lambda’s 10GB limit
• - Specific EC2 instance types (e.g., Graviton4) offer performance benefits
• - Cost optimization at scale favors EC2 pricing with Savings Plans over per-GB-second billing
• - Functions load large datasets into memory for reuse across multiple requests

Standard Lambda remains preferable for:

• - Bursty or unpredictable traffic patterns
• - Low-to-moderate throughput workloads
• - Workloads needing instant scaling (LMI scales asynchronously based on CPU utilization)

Understanding Lambda Managed Instances Architecture

LMI’s architecture consists of three core components that work together to deliver its unique capabilities:

1. Capacity Provider

The foundation of your LMI deployment defines:

• - VPC configuration and networking requirements
• - Instance types and hardware specifications
• - Scaling policies and security boundaries

Critical Consideration: All functions sharing a capacity provider run on the same EC2 instances. This means:

• - Functions must be mutually trusted (container-based isolation only)
• - A security breach in one function could affect others on the same instances
• - Workloads should be segmented by trust level (e.g., production vs. non-production)

2. Managed Instances

These are EC2 instances launched and managed entirely by AWS Lambda within your VPC. Key characteristics include:

• - Visible in the EC2 console but untouchable (no SSH access)
• - Automatically patched and rotated every 14 days for compliance
• - Tagged as "managed by Lambda" for easy identification
• - Scale dynamically based on workload demands

3. Execution Environments

Each environment runs your function code in containers on the managed instances. Important details:

• - Handles multiple concurrent requests simultaneously
• - For Python, each concurrency slot uses a separate process with isolated memory
• - Lifespan exceeds individual invocations (long-lived environments)

Networking Requirements

A critical but often overlooked aspect of LMI is networking. Since VPC connectivity is mandatory:

• - Private subnets with NAT Gateway are typically required for outbound telemetry
• - Without proper network configuration, functions may execute while logs and traces are lost
• - This project leverages private subnets with NAT for reliable telemetry transmission

When to Choose Lambda Managed Instances Over Alternatives

The decision to adopt LMI depends on your workload’s characteristics. Evaluate these factors:

Opt for LMI if:

• - Your application requires sustained throughput beyond 1,000 requests per second
• - Memory demands exceed 10GB or need configurable vCPU ratios
• - You need access to specific EC2 instance types for performance optimization
• - Your workload benefits from large in-memory datasets shared across requests
• - You’re processing over 10 million invocations monthly where EC2 Savings Plans offer cost advantages

Stick with Standard Lambda if:

• - Your traffic patterns are highly unpredictable or intermittent
• - You process fewer than 10 million monthly requests where per-invocation pricing is more cost-effective
• - Your workload doesn’t require high memory or multi-concurrency
• - You need instant scaling capabilities that LMI’s asynchronous approach doesn’t provide

For workloads at the boundary between these options, consider testing both approaches. The flexibility of Lambda Managed Instances makes it an excellent choice for evolving applications that may outgrow standard Lambda’s limitations.

The Future of Serverless and Hybrid Compute

Lambda Managed Instances represents a significant evolution in AWS’s serverless offerings, blurring the lines between traditional serverless and containerized compute. As applications grow more complex and resource-intensive, hybrid approaches like LMI will become increasingly valuable.

For developers building modern, scalable applications, understanding this compute continuum is crucial. The ability to mix and match services based on workload requirements—while maintaining operational simplicity—will define the next generation of cloud-native architectures.