Audio & Video 1:1 Live Streaming Cost Per Minute: Cloud vs Self-Hosted

How Much Does 1:1 Audio or Video Calling Cost Per Minute?

If you are building a video calling, consultation, tutoring, dating, healthcare, social networking, customer-support or live-streaming application, one of the first questions eventually becomes:

“How much will every minute of an audio or video call actually cost us?”

The answer is more complicated than simply looking at an API provider’s advertised price.

For a 1:1 call, you may have two participants, but many communication platforms charge based on participant-minutes, not simply “call minutes.”

For example, if two people remain connected for 10 minutes, a provider charging $0.004 per participant-minute effectively bills:

2 participants × 10 minutes × $0.004 = $0.08

Twilio currently lists Video Group Rooms at $0.004 per participant-minute.

Agora lists its RTC pricing separately for audio and video, with current list pricing shown as $0.99 per 1,000 audio minutes, $3.99 per 1,000 HD video minutes and $8.99 per 1,000 Full HD video minutes, with pricing calculated per participant minute.

AWS Chime SDK uses a similar attendee-minute model for WebRTC media and currently lists WebRTC media at $0.0017 per attendee-minute in its pricing examples.

But there is another option:

Build and Operate Your Own WebRTC Infrastructure

This can reduce the variable cost at scale, but it transfers the responsibility to your engineering and DevOps team.

You may need to operate:

WebRTC signaling, SFU/media server, STUN/TURN, load balancing, servers, bandwidth, monitoring, logging, security, scaling, recording infrastructure and storage.

So the real question isn’t:

“Which provider has the cheapest per-minute price?”

It is:

“At our expected call volume, is managed RTC or self-hosted infrastructure economically and operationally better?”

First: What Does a 1:1 Call Actually Require?

A production-quality 1:1 audio/video system normally has several components.

Client Applications

Android, iOS, web or desktop clients capture microphone and camera data and establish WebRTC sessions.

Signaling Server

Signaling exchanges information required to establish the connection, including session descriptions and ICE candidates.

It does not necessarily carry the actual audio/video media.

STUN

STUN helps clients discover their public network address and determine whether a direct connection is possible.

TURN

When direct peer-to-peer connectivity fails because of NAT, firewall or network restrictions, TURN can relay media traffic.

SFU

An SFU, or Selective Forwarding Unit, becomes important when you want more control over media routing, recording, moderation, scaling or future group-call functionality.

Popular self-hosted approaches include WebRTC-based SFUs such as LiveKit, mediasoup and Janus.

Infrastructure

You also need servers, networking, monitoring, deployment automation, certificates, DNS and security controls.

This is why:

WebRTC SDK ≠ complete production infrastructure.

1:1 Audio Call Cost Per Minute

Let’s first consider audio-only calling.

Managed RTC providers can be extremely inexpensive for audio.

For example, Agora currently lists:

Audio: $0.99 / 1,000 participant minutes.

That works out to approximately:

$0.00099 per participant-minute

For a 10-minute 1:1 call:

2 participants × 10 minutes × $0.00099

= $0.0198

So the RTC provider’s media cost is approximately 2 cents for a 10-minute call, before other application-related costs.

Pricing varies by provider, geography, product configuration and volume, so published list prices should be treated as a starting point rather than a final enterprise quote.

1:1 Video Call Cost Per Minute

Video is more expensive because substantially more data must be transmitted.

Agora currently lists:

HD video: $3.99 / 1,000 participant minutes

Full HD video: $8.99 / 1,000 participant minutes

The pricing is calculated per participant minute.

For a 1:1 HD call:

2 participants × $0.00399

= $0.00798 per call minute

So a 60-minute 1:1 HD call would be approximately:

$0.4788

before other application costs.

At Full HD:

2 × $0.00899

= $0.01798 per call minute

A 60-minute call would therefore be approximately:

$1.0788

again before other services.

What About Twilio?

Twilio Video currently lists $0.004 per participant-minute for Video Group Rooms.

For a 1:1 call:

2 participants × $0.004

= $0.008 per call minute

Therefore:

Call duration	Approx. RTC charge
1 minute	$0.008
10 minutes	$0.08
30 minutes	$0.24
60 minutes	$0.48
1,000 minutes	$8
10,000 minutes	$80
100,000 minutes	$800

These are illustrative calculations based on the published participant-minute rate, not a complete application operating cost. Twilio also charges separately for optional capabilities such as recording, composition and transcription.

AWS Chime SDK: Another Model

AWS Chime SDK uses attendee-minute pricing for WebRTC media.

AWS’s published example uses:

$0.0017 per attendee-minute

For a 1:1 call:

2 × $0.0017

= $0.0034 per call minute

A 60-minute call:

60 × $0.0034

= $0.204

AWS also offers additional capabilities such as media capture, streaming and concatenation, which can introduce additional costs.

This illustrates an important point:

There is no universal “video call price.”

Your final cost depends on the provider, media quality, participant count and additional features.

Managed RTC vs Self-Hosted WebRTC

Now comes the more interesting question.

What If We Build Everything Ourselves?

Suppose Ethereal builds a platform using:

WebRTC + SFU + Coturn + signaling + PostgreSQL + Redis + cloud infrastructure + monitoring

instead of paying a managed RTC provider for every participant-minute.

There is no fixed “self-hosted cost per minute.”

Why?

Because your infrastructure is usually billed by:

server time + bandwidth + storage + load balancing + monitoring + recording + engineering/DevOps

rather than directly by call minute.

Therefore, the correct formula is:

Self-hosted cost per minute = Total infrastructure cost ÷ Actual billable call minutes

This distinction is extremely important.

Example: Self-Hosted 1:1 Audio Infrastructure

Suppose you operate:

1 signaling server
1 small application server
1 TURN server
Monitoring/logging
Database
Basic load balancing

The servers run continuously even when nobody is calling.

Assume the complete infrastructure costs approximately:

$150/month

and your platform handles:

500,000 audio call minutes/month

Then:

$150 ÷ 500,000

= $0.0003 per minute

or approximately:

$0.30 per 1,000 minutes

That looks dramatically cheaper than many managed solutions.

But there is a catch.

The infrastructure is not free when traffic is low.

If you only have:

10,000 minutes/month

then:

$150 ÷ 10,000

= $0.015 per minute

The same infrastructure suddenly looks expensive.

This is why utilisation matters more than the server’s monthly price.

Self-Hosted Video Is Different

Video consumes substantially more network bandwidth.

Suppose a 1:1 HD call averages approximately:

1 Mbps per participant

This is only an illustrative engineering assumption because actual bitrate changes dynamically depending on codec, resolution, network conditions, motion and adaptive bitrate settings.

Two participants could therefore produce approximately:

2 Mbps of aggregate media traffic through an SFU

for a 1:1 session.

At 2 Mbps, the outbound traffic is roughly:

15 MB per minute

for the media stream as a simplified estimate.

Therefore:

60 minutes ≈ 900 MB

per 1:1 session of aggregate media traffic under this simplified model.

Actual traffic can be materially different because WebRTC dynamically changes bitrate and because architecture determines whether media is directly peer-to-peer, relayed through TURN or routed through an SFU.

The Most Important Self-Hosted Cost: Bandwidth

Many teams initially focus on:

“Which EC2 instance should we buy?”

But for video applications, network traffic can become more important than CPU cost.

AWS notes that data transfer out to the internet is a separate pricing dimension, and AWS networking charges depend on the service, region and traffic path.

That means a self-hosted architecture should be designed around:

bandwidth efficiency + media routing + server utilisation

rather than simply selecting the cheapest VM.

A More Realistic Self-Hosted Cost Model

For an early-stage application, consider this as an illustrative planning range, not a guaranteed price:

Infrastructure	Approximate monthly range
Signaling/API server	$20–$60
SFU/media server	$50–$200+
TURN servers	$20–$100+
Database/cache	$20–$100+
Load balancer/networking	$20–$100+
Monitoring/logging	$10–$100+
Storage/recordings	Usage based
Bandwidth	Usage based
DevOps/maintenance	Depends on team

A small production deployment might therefore begin around:

$150–$500+/month

but a serious production platform with redundancy, multiple regions, recording, high availability and substantial video traffic can quickly move into:

$1,000s/month and beyond.

The infrastructure cost should therefore be calculated from your expected concurrency and traffic rather than from a generic “server cost.”

So What Is the Self-Hosted Cost Per Minute?

There is no honest universal number.

But for planning purposes, a company can create a model like this:

Low utilisation

50,000 call minutes/month

Infrastructure: $250/month

Effective infrastructure cost:

$0.005/minute

Medium utilisation

500,000 call minutes/month

Infrastructure: $500/month

Effective infrastructure cost:

$0.001/minute

High utilisation

5,000,000 call minutes/month

Infrastructure: $2,500/month

Effective infrastructure cost:

$0.0005/minute

These numbers are illustrative, not provider quotations.

The actual cost can be significantly higher if video bandwidth, TURN relay, recording or multi-region traffic becomes the dominant expense.

Why TURN Can Change Your Economics

A common misconception is:

“We’re using WebRTC, so our server doesn’t carry the video.”

Not necessarily.

When peer-to-peer connectivity succeeds, the media can travel directly between users.

But when direct connectivity fails, TURN can relay the traffic.

For a TURN-relayed video call, your infrastructure may effectively become part of the media path.

AWS documentation explicitly describes self-hosted TURN using open-source software such as Coturn on EC2 or ECS.

This means a production system should monitor:

P2P success rate

TURN usage percentage

SFU bandwidth

average bitrate

concurrent sessions

packet loss

jitter

CPU utilisation

network throughput

These metrics are much more useful than simply monitoring “number of calls.”

The Break-Even Question

This is the question Ethereal recommends asking before selecting the architecture:

At what volume does self-hosting become cheaper than managed RTC?

Suppose a managed provider effectively costs:

$0.008 per 1:1 video call minute

and your self-hosted platform costs:

$500/month

Your theoretical break-even point is:

$500 ÷ $0.008

= 62,500 call minutes/month

But this calculation is incomplete.

You also need to consider:

bandwidth, TURN traffic, recording, monitoring, backups, engineering salaries, DevOps time, incident response, redundancy and scaling.

Therefore, the real break-even point can be considerably higher.

Don’t Ignore Engineering Cost

This is where many “self-hosted is cheaper” calculations fail.

If you use a managed service, your engineering team doesn’t have to build and operate the entire media infrastructure.

With self-hosting, your team becomes responsible for:

WebRTC configuration, SFU deployment, TURN, NAT traversal, scaling, security, certificates, observability, capacity planning, regional deployment, upgrades, failure recovery and performance troubleshooting.

A $300/month infrastructure bill isn’t necessarily cheaper if your engineering team spends dozens of hours every month maintaining it.

Infrastructure cost and total cost of ownership are different numbers.

When Should You Use Managed RTC?

Managed RTC is generally attractive when:

You are building an MVP

Call volume is uncertain

You want to launch quickly

Your team doesn’t specialise in WebRTC

You need global connectivity

You want managed scaling

Your application requires recording/transcription

Your engineering team should focus on the product rather than media infrastructure

For early-stage products, paying a little more per minute can be rational if it dramatically reduces development and operational risk.

When Does Self-Hosting Make Sense?

Self-hosted WebRTC becomes more interesting when:

Call volume is predictable

Monthly minutes are high

You need deep control over media infrastructure

You need custom routing

You need specialised media processing

You have an experienced backend/DevOps/WebRTC team

Data residency or infrastructure control is important

Managed RTC costs are becoming a significant percentage of revenue

At this stage, owning the media infrastructure can potentially improve unit economics.

The Architecture Ethereal Would Recommend

For a serious 1:1 calling product, don’t immediately build 20 microservices.

A practical architecture could start with:

1. Mobile/Web Client
2. API + Authentication
3. Signaling Service
4. WebRTC
5. SFU
6. TURN/STUN
7. Monitoring

with:

PostgreSQL + Redis

for application state and supporting services.

As traffic increases, introduce:

multiple SFU nodes, regional media routing, autoscaling, dedicated TURN pools, load balancing, QoS monitoring and capacity-based scheduling.

This allows the architecture to evolve with actual usage.

Audio vs Video: Which Is Cheaper?

Obviously:

Audio

Very low bandwidth and infrastructure requirements.

HD Video

Significantly higher bandwidth and media-processing requirements.

Full HD Video

Higher bandwidth requirements again.

Recording

Adds storage, processing and potentially composition/transcoding costs.

Transcription

Adds another per-minute processing cost.

TURN Relay

Can materially increase network traffic costs.

Therefore:

The “RTC price per minute” is only one component of the total cost of a communication product.

What About Recording?

Recording should be treated as a separate cost layer.

For example, Twilio currently charges separately for participant recordings and recording compositions.

A self-hosted platform can instead capture media using its own recording pipeline and store the output in object storage.

But that introduces:

media capture infrastructure + encoding + storage + data transfer + retention + security

into the architecture.

So if your application requires:

1:1 video + recording + transcription + moderation

the cost model needs to be calculated separately from the base video-call cost.

Example: 100,000 Minutes Per Month

Let’s compare conceptually.

Assume:

100,000 1:1 video call minutes/month

Managed RTC

At an illustrative effective cost of:

$0.008 per call minute

100,000 × $0.008

= $800/month

Self-Hosted

Suppose infrastructure + bandwidth + monitoring averages:

$600/month

Then:

$600 ÷ 100,000

= $0.006/minute

Self-hosting appears cheaper.

But now suppose your actual infrastructure reaches:

$1,200/month

Then:

$1,200 ÷ 100,000

= $0.012/minute

The managed service becomes cheaper.

This is why architecture should be selected using real traffic projections, not assumptions.

The Better Strategy: Hybrid RTC

There is also a third option.

Don’t choose between 100% managed and 100% self-hosted immediately.

Build your application with a media abstraction layer.

For example:

1. Application
2. RTC Interface
3. Provider A / Provider B / Self-Hosted SFU

This allows you to change the media infrastructure without rewriting the entire product.

You can start with a managed provider.

Once your traffic reaches a meaningful level, evaluate:

managed cost vs self-hosted cost

using real production data.

This can be much safer than building an expensive media infrastructure before you know whether your product will generate enough usage.

What Should a Startup Budget?

For an MVP with 1:1 audio/video calling, a reasonable planning approach is:

Option 1 — Managed RTC

Lower infrastructure complexity, usage-based RTC charges and faster launch.

Option 2 — Self-Hosted WebRTC

Higher engineering responsibility, potentially lower variable cost at scale.

Option 3 — Hybrid

Managed infrastructure initially, with an architecture designed to support self-hosting later.

For most startups, Option 3 is worth serious consideration.

The Real Formula You Should Use

Instead of asking:

“How much does video calling cost per minute?”

Use this:

Managed RTC

Total Cost = Participant Minutes × Provider Rate + Recording + Transcription + Other Usage

Self-Hosted

Total Cost = Compute + Bandwidth + TURN + SFU + Storage + Monitoring + Networking + Operations

Then:

Effective Cost Per Minute

Total Monthly RTC Infrastructure Cost ÷ Total Monthly Call Minutes

This is the number that should go into your product’s unit economics.

In Summery

A 1:1 audio or video call can be surprisingly inexpensive at the media layer.

But building a reliable communication platform is much more than paying for WebRTC minutes.

For a startup, managed RTC services can provide a fast and predictable way to launch.

For a high-volume platform, self-hosted WebRTC infrastructure can potentially reduce the variable cost per minute—but only when utilisation is high enough to justify the infrastructure and engineering overhead.

The best architecture depends on:

monthly call minutes, concurrent calls, audio/video quality, geographic distribution, TURN percentage, recording requirements, expected growth and engineering capabilities.

At Ethereal Softech, we don’t recommend choosing a technology simply because its advertised per-minute price looks cheaper.

We first calculate the expected cost per call, cost per 1,000 minutes, monthly infrastructure cost, peak concurrency and five-year scalability requirements.

Then we decide whether the product should use Agora, Twilio, AWS, LiveKit, a self-hosted WebRTC/SFU architecture or a hybrid approach.

Because the cheapest technology on paper isn’t always the cheapest technology to operate.

Planning to build a 1:1 video calling app, consultation platform, social app, tutoring platform, telehealth product or live-streaming application? Ethereal Softech can help design the RTC architecture and calculate the expected infrastructure cost before development begins.