Use the same workload for every quote: concurrent rooms, participants, video quality, duration, recording concurrency, transcript service and retention. The requirements guide explains capacity planning; this page explains the bill.
Build the monthly budget
| Cost | Inputs to collect | Common omission |
|---|---|---|
| Meeting compute | Region, instance, hourly rate and running hours | Dedicated versus shared CPU, CPU-credit charges |
| Data transfer | Measured or modeled outbound GB, included allowance and rate tiers | TURN, inter-region and cross-zone paths |
| Disks and addresses | Provisioned disk, snapshots and chargeable IP addresses | Resources left allocated while compute is stopped |
| Recording | Separate Jibri instances, hours, output size and storage retention | Several simultaneous recordings need several instances |
| Transcription | Speech model/service, concurrency and resources or usage | Model-dependent memory and external API charges |
| Operations | Setup, updates, monitoring, backup checks and incident work | Engineering work is part of total cost |
Use your provider’s current regional quote. For AWS, start with EC2 on-demand pricing, EBS, S3 and the AWS Pricing Calculator. Burstable instances can incur surplus CPU-credit charges. These references do not imply a particular instance will handle your meeting load.
Bandwidth: price what leaves the server
For a steady modeled workload:
outbound GB = aggregate outbound Mbps × equivalent hours at that load × 0.45The factor uses decimal GB: 3,600 seconds divided by 8 bits and 1,000 MB per GB. For example, 12 Mbps for 40 equivalent hours is 216 GB. This is arithmetic, not a measured Jitsi bitrate or a provider charge.
Use the calculator to expose video-stream assumptions and download planning notes. Model differing room sizes separately and evaluate audio, screen sharing and relay traffic as well. Successful two-person P2P may reduce bridge traffic, but bridge or TURN fallback still needs a budget. [S1]
Recording and transcription are separate budgets
Plan one Jibri instance per simultaneous recording and size its resources separately. Do not assume that adding RAM to the meeting server makes colocated recording safe. Budget output retention, storage requests, retrieval and transfer as applicable. [S2]
Choose the transcription model before estimating memory. Vosk describes small models with about 300 MB runtime memory and large models requiring up to 16 GB; neither number is a total-system estimate. Other speech services have their own usage charges. [S3]
Worked example with assumed rates
These inputs are illustrative USD assumptions, not AWS prices, a quote or a tested server recommendation. Replace every rate with your provider’s current regional quote.
| Assumed line item | Calculation | Subtotal |
|---|---|---|
| Meeting compute | 730 hours × $0.05/hour | $36.50 |
| Separate recorder compute | 40 hours × $0.10/hour | $4.00 |
| Provisioned disk | 40 GB-month × $0.10 | $4.00 |
| Billable outbound transfer | 200 GB × $0.05 | $10.00 |
| Subtotal of these four lines | $54.50 |
This subtotal excludes recording-object storage, backups, IP charges, taxes, optional software/service fees and engineering work. It assumes the 200 GB is already the chargeable amount after any allowance. It does not establish that the example resources can support your workload. Use the capacity worksheet to record actual measurements before purchasing capacity.
Compare hosted and self-hosted quotes fairly
JaaS uses monthly active users plus applicable optional usage charges. Self-hosting uses infrastructure and operating costs; it is not automatically cheaper as user counts rise. Compare the same unique users, meeting durations, peak concurrency, recording minutes, retention and support requirements. See JaaS vs self-hosted Jitsi and hosting options.
Getting help with your estimate
If you want deployment in your own AWS account, review our hosting approach and platform feature pricing separately from the AWS bill. For an existing server, request an engineering review with the release, peak workload and sanitized measurements. Agree the scope and quote before paid work.
Sources and scope
Reviewed October 20, 2026. This is a budgeting method, not a regional price feed or a completed capacity benchmark. Recheck current provider rates before purchasing.
- [S1] Jitsi configuration, including P2P.
- [S2] Jitsi requirements and recording separation and Jibri.
- [S3] Vosk models and memory.
- Current provider pricing references are linked in the budget section; JaaS pricing uses a different billing unit.