AWS Lambda Web Functions Run Ordinary HTTP Servers With Multi-Request Concurrency
AWS introduced Lambda Web Functions, a third Lambda form factor alongside Event Functions and MicroVMs, all sharing the same Firecracker isolation. Web Functions run standard port-listening HTTP servers without adapters or a rewrite into event handlers: Lambda sets AWS_LAMBDA_HTTP_ENDPOINT to the host and port the server must bind, and any Node.js framework — Express, Hono, Fastify or Next.js — works as-is. At launch the runtime is Node.js 24 (nodejs24.x) on Arm64, with VPC connectivity supported.
The execution model differs sharply from classic Lambda. Each execution environment handles up to 128 concurrent requests, so traffic bursts share warm environments instead of triggering a cold start per request. Billing is based on active CPU time, so time spent waiting on database or API calls costs little CPU, while memory is metered for as long as the environment exists.
Deployment uses aws lambda-web deploy; adding --regions replicates the code to several regions and sets up best-effort latency-based routing behind one global URL. update-web-function-endpoint supports weighted revisions for canary rollouts, and CloudFront can use a Web Function as an origin protected by origin access control. Launch regions are us-east-1, us-east-2, us-west-2, ap-northeast-1 and eu-west-1.
Read more — AWS News Blog
AWS Week in Review: Nova 2.5 Sonic, OAuth for Lambda Kafka Triggers, Redshift Iceberg Materialized Views
AWS's October 13 weekly roundup gathered several developer-relevant launches. Amazon Nova 2.5 Sonic is a new speech-to-speech model for real-time voice agents with better reasoning, instruction following and tool-calling accuracy, lower latency, expressive voices in seven languages, controllable turn-taking and a 256K context window. It pairs with Strands Bidi Agents, the bidirectional-streaming voice agent framework, which is now generally available.
AWS Lambda event source mappings for self-managed Apache Kafka now support OAuth, so consumers can authenticate through an identity provider such as Amazon Cognito or Okta instead of only SASL/PLAIN, SASL/SCRAM or mTLS. This works with self-hosted clusters as well as Confluent Cloud, Aiven and Redpanda, and removes the need to distribute long-lived SASL credentials to Lambda.
Amazon Redshift Iceberg materialized views precompute joins and aggregations and store the results as Apache Iceberg tables in S3, registered in the AWS Glue Data Catalog, so Athena, Spark, Trino and Snowflake can query them; refreshes recompute only what changed. On the security side, GuardDuty RDS Protection now analyzes query activity on Aurora PostgreSQL and RDS for PostgreSQL to detect data exfiltration and destruction, with no agents or database configuration changes.
Read more — AWS News Blog
BigQuery Adds GA Token Quotas to Cap Spend on Generative AI SQL Functions
Google Cloud made token-based cost management for BigQuery generative AI functions generally available. Four daily quotas — input tokens and output tokens, each per project and per user — are managed through the standard IAM quota system and are global across regions. Defaults are 200 billion input / 20 billion output tokens per project per day and 150 billion / 15 billion per user; output quotas include thinking tokens. Lowering a quota takes effect immediately, and quotas can be managed through the console, Cloud Quotas API, gcloud or Terraform.
The quotas cover AI.GENERATE, AI.GENERATE_TABLE, AI.GENERATE_TEXT/ML.GENERATE_TEXT, AI.CLASSIFY, AI.IF, AI.SCORE and AI.AGG, but not embedding functions. Enforcement is graceful: a query is rejected up front only if a needed quota is already exhausted; if a quota runs out mid-query, BigQuery stops issuing new model calls, and functions such as AI.GENERATE still succeed with a quota-exceeded status on skipped rows, while functions supporting max_error_ratio return partial results if within the ratio.
Google recommends sizing quotas with AI.COUNT_TOKENS, which returns token counts without a billable generation call. Its worked example targets about $100 per day on Gemini 3.1 Flash-Lite batch pricing — roughly 754 million input and 7.5 million output tokens daily at a 100:1 ratio — and suggests starting with per-user quotas so one experiment cannot drain a project's budget.
Read more — Google Cloud Blog
Links & Sources
You May Also Enjoy
