keepcalculator

06 Numbers and Powers

Zoom Recordings to AWS S3: Best Automated Route

Card No. 06·72935 · Updated · Standard formula, shown below

The best automated route for moving Zoom recordings to AWS S3 is a serverless pipeline using Zoom’s Cloud Recording API and an AWS Lambda function triggered by a webhook, with the recording files transferred directly from Zoom’s temporary URLs to S3. This approach avoids running a server 24/7, scales automatically, and costs only when recordings are actually processed. Below is a detailed look at why this works, how to build it, and what to watch out for.

Background: Why Automate Zoom-to-S3 at All

Zoom’s default cloud storage is convenient but limited. Free and low-tier accounts get a small amount of storage (often 1–5 GB), and even paid plans cap out at a few gigabytes per license. Once you exceed the limit, Zoom stops recording to the cloud until you delete old files. For businesses that record training sessions, webinars, or client meetings, that means constant manual cleanup or losing recordings entirely.

Moving recordings to S3 solves the storage ceiling and gives you control: S3 costs about $0.023 per GB per month for standard storage (as of 2025), and you can set lifecycle rules to move old files to cheaper tiers like Glacier. You also get versioning, cross-region replication, and integration with other AWS services for transcription or analytics. The challenge is that Zoom doesn’t natively push recordings to S3—you have to build a bridge.

How Zoom’s Recording Delivery Works

When a cloud recording finishes, Zoom does two things: it stores the files in its own cloud, and it sends a webhook notification to any endpoint you’ve configured. The webhook contains a payload with a download_url and a download_token. That URL is temporary—it expires after a few hours (typically 24 hours, but it can be shorter). To get the files, you must download them from that URL and then upload them to S3.

The recording itself isn’t a single file. Zoom produces multiple files per meeting: a video (MP4), an audio-only file (M4A), a transcript (VTT), a chat log (TXT), and sometimes separate files for each participant’s video. The webhook payload lists each file with its own download URL and size. Your automation must iterate through that list, download each file, and upload it to S3—ideally preserving the meeting metadata in the S3 object tags or a database.

The Core Architecture: Webhook + Lambda + S3

The most common and cost-effective design is:

  1. Create an API Gateway endpoint that receives Zoom’s webhook POST requests. Zoom requires HTTPS, so API Gateway handles the SSL termination.
  2. Write a Lambda function in Python or Node.js that parses the webhook payload, validates the event type (e.g., recording.completed), and extracts the download URLs.
  3. Download each file from Zoom’s URL using the provided token. Use the requests library in Python or axios in Node.js. Be careful: the download can take time for long recordings, so set a generous Lambda timeout (up to 15 minutes) and consider using S3 Transfer Acceleration if your Lambda runs far from Zoom’s servers.
  4. Upload to S3 using the AWS SDK. Store files under a key structure like zoom-recordings/{meeting_uuid}/{recording_id}/{filename}. You can also add metadata like the meeting topic, host email, and date.
  5. Clean up: delete the local temp file in Lambda after upload to avoid filling the /tmp space (512 MB max).

This design is event-driven: it runs only when a recording is ready, so you pay for Lambda invocations (a few cents per 1,000 calls) plus S3 storage. No EC2 instance to manage, no idle cost.

Choosing Between Webhook and Polling

There are two primary ways to detect new recordings: webhooks (push) or polling the Zoom API (pull). Webhooks are the better choice for most cases because they’re real-time and don’t require continuous API calls. However, webhooks have a caveat: Zoom sends them to a single endpoint, and if your endpoint is down, Zoom retries for up to 24 hours. If you need guaranteed delivery, you can use an SQS queue as a buffer between API Gateway and Lambda—if Lambda fails, the message stays in the queue for a retry.

Polling is simpler to implement (no webhook configuration, just a scheduled Lambda that calls GET /users/{userId}/recordings) but it’s less efficient. You’d be calling the API every few minutes even when there are no new recordings, which can hit rate limits (Zoom’s default is 10 requests per second). Polling also introduces latency—you won’t know about a recording until the next poll. For a low-volume use case (a few recordings a day), polling is acceptable; for high volume or time-sensitive workflows, webhooks are superior.

Common Pitfalls and How to Avoid Them

  • Expired download URLs: If your Lambda is delayed (e.g., due to a cold start or a queue backlog), the token may expire. Mitigate by processing immediately or refreshing the token via the Zoom API (the recording.retention settings can extend the window).
  • Large files and Lambda timeout: A 2-hour 1080p recording can be 1–2 GB. Lambda’s max execution time is 15 minutes, and the /tmp space is 512 MB. For files larger than that, you cannot download and upload in one Lambda invocation. Solutions: use S3 Transfer Acceleration to speed up upload, or use AWS Fargate/ECS for very large files, or split the download into chunks using the Range header and upload each part with the S3 multipart API.
  • Webhook signature verification: Zoom signs webhooks with a secret. Always verify the signature (using the X-Zm-Signature header) to prevent attackers from triggering fake downloads or filling your S3 with junk.
  • Duplicate processing: Zoom may send the same webhook multiple times (e.g., on retries). Use the meeting_uuid and recording_id as a unique key in DynamoDB to make the process idempotent.
  • Permissions: The Lambda execution role needs s3:PutObject and s3:PutObjectAcl (if setting ACLs). Avoid giving broad S3 permissions; scope to the specific bucket.

Cost and Performance Considerations

For a typical small business recording 10 hours of meetings per month, the cost is minimal: S3 storage for 10 GB is about $0.23/month, Lambda invocations (say 30 per month) cost nothing (within the free tier), and data transfer from Zoom to S3 is free (Zoom doesn’t charge for downloads). The main variable is data transfer out of S3 if you later access the files—that costs $0.09/GB for the first 10 TB. If you plan to serve recordings to viewers, consider CloudFront to reduce costs and improve speed.

Performance-wise, the bottleneck is the download from Zoom. Zoom’s download speeds vary; a 1 GB file might take 2–5 minutes. If you need faster processing, run the Lambda in the same AWS region as your S3 bucket (though Zoom’s servers are global, the download is from their CDN). You can also use threading in Python to download multiple files concurrently, but be mindful of Lambda’s CPU allocation.

Alternatives to the Serverless Route

If you don’t want to write code, there are third-party tools like Zapier or Make (formerly Integromat) that offer Zoom-to-S3 integrations with a few clicks. These are fine for low volume but can get expensive at scale (Zapier charges per task) and offer less control over file naming and metadata. Another alternative is to use Zoom’s native integration with some cloud storage providers—but as of 2025, Zoom does not offer a direct S3 integration; you must use the API.

For teams already on AWS, you could also use AWS Step Functions to orchestrate a multi-step workflow (download, transcribe, upload, notify) with better error handling. But for simple transfer, a single Lambda function is sufficient.

Frequently Asked Questions

Q: Can I use a scheduled Lambda to poll Zoom’s API instead of webhooks?
A: Yes, you can. Use a CloudWatch Events rule to trigger a Lambda every 5–10 minutes that calls the Zoom List Recordings API. This is simpler to set up but adds latency and potential rate-limit issues. Webhooks are generally preferred for real-time processing.

Q: How do I handle recordings over 512 MB if Lambda has a 512 MB /tmp limit?
A: For files larger than 512 MB, you cannot download the whole file to /tmp. Instead, stream the download directly to S3 using the S3 SDK’s upload method with a stream, or use a multipart upload where you read chunks from the Zoom URL and upload each part. This avoids storing the entire file locally.

Q: Does Zoom charge extra for downloading recordings?
A: No, Zoom does not charge for downloading your own cloud recordings. The only cost is your AWS usage (S3 storage, Lambda invocations, data transfer out if you later serve the files).

Q: How do I verify the webhook is really from Zoom?
A: Zoom sends a header X-Zm-Signature which is an HMAC-SHA256 of the request body using your webhook secret. Compute the expected signature in your Lambda and compare; if they don’t match, reject the request.

Q: Can I automatically transcribe the recordings after they’re in S3?
A: Yes. After the upload, you can trigger Amazon Transcribe on the audio file. Use an S3 event notification to invoke another Lambda that starts a transcription job, then store the output in S3 or DynamoDB.

Q: What if I need to keep recordings for a long time but want to save costs?
A: Set up an S3 lifecycle policy to transition objects to S3 Standard-IA after 30 days, then to Glacier after 90 days. You can also enable intelligent-tiering. Just be aware that retrieving from Glacier takes minutes to hours, so plan for that if you need access.

Related

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.

More cards in this drawer

The whole Numbers and Powers drawer →