What the 100ms Startup Program Gives You
100ms offers up to $2,500 in credits toward its live video and audio SDK, covering the meter that actually runs the bill: minutes of real-time audio and video carried for each participant in a room, plus the recording and live streaming layered on top.
AI Perks tracks it alongside $7.7M in credits across 194 companies.
Know the boundary before planning around it. 100ms sells the live layer - media servers, client SDKs for web and mobile, and the room orchestration keeping them in sync. The credit pays that invoice and stops there. It does not cover the object storage your recordings land in, the CDN egress when you serve them back, or the model tokens behind any AI feature in the call. Current amounts and terms are listed on getaiperks.com.
What a Live Video SDK Actually Replaces
A video SDK is three products sold as one: media infrastructure you would otherwise operate, client SDKs for every platform you ship on, and a prebuilt call interface. At seed stage the third is usually what you are really paying for.
The comparison that matters is not 100ms against a rival vendor. It is hosted video against two cheaper options founders skip past.
Plain WebRTC. Two people, two browsers, good networks: free, built in, and enough for longer than most teams expect. The cliff arrives on four fronts at once - three or more participants, where peer-to-peer makes every device upload to every other one; networks that need a relay; cross-border calls; and mobile, where surviving a wifi-to-cellular handoff and backgrounding is a year of SDK work unrelated to your product.
A self-hosted media server. The open-source options are good and a real choice for an infra-heavy team. You take on capacity planning, relay servers, regional deployment and a pager rotation. Sane at scale, expensive at ten customers.
What you buy instead is mostly time. A production call screen is not a video element. It is device permissions, switching microphones mid-call, active speaker detection, network quality indicators, reconnect, screen share, echo cancellation and a layout that survives one person joining on a phone. Each is small. Together they are a quarter of your frontend engineering for a quarter of the year.
The heuristic: if video is a feature, ship a prebuilt room and spend the saved months elsewhere. If video is the product, you will outgrow prebuilt UI and want the low-level SDK, so pick a vendor offering both. Credit amounts across this category are listed on getaiperks.com.
How 100ms Pricing Behaves at Scale
100ms meters per participant per minute. Cost is therefore linear in headcount rather than quadratic in streams, which makes group video far more predictable to budget here than on per-stream pricing - and moves the real risk onto recording, streaming and rooms nobody closed.
| Meter | What it counts | What makes it spike |
|---|---|---|
| Audio minutes | Each participant, per minute in a room | Rooms left open after everyone leaves |
| Video minutes | Each participant, per minute, by quality tier | HD publishing into a thumbnail-sized layout |
| Recording | A separate composite or per-track render | Recording every session by default |
| Live streaming | Viewer minutes on the broadcast leg | Streaming to an audience that could have been a CDN replay |
| Transcription and AI | Billed by other vendors entirely | Assuming video credits cover the model |
Rates, quality tiers and any free monthly allowance change over time. Verify current figures on the vendor's own pricing page before modelling anything.
The participant-minute is what your invoice counts, and almost never what your product spec says:
| Session shape | People | Wall clock | Participant-minutes billed |
|---|---|---|---|
| 1:1 sales call | 2 | 30 min | 60 |
| Live class | 25 | 60 min | 1,500 |
| Webinar run as a conference | 200 | 60 min | 12,000 |
Every row is one event. The last consumes as much metered usage as 200 one-to-one sales calls, and nobody in it was talking.
The second lever is the quality tier. Audio-only minutes cost a fraction of HD video minutes, and the tier follows what each client publishes, not what your interface renders. A grid showing twelve people at 180 pixels wide while every client publishes HD is paying the top rate to discard pixels.
When to Broadcast Instead of Conference
The most expensive architectural mistake in live video is putting a passive audience inside an interactive room. If most participants will never speak, they belong on a live stream, not in a conference.
Interactive video buys sub-second latency so people can interrupt each other. An audience does not. Broadcast trades seconds of delay for a model priced far lower per viewer, and nobody notices unless they are on stage. Three shapes:
- Everyone can speak, roughly twenty or fewer. Interactive room. This is the case the SDK is built for.
- A few speak, many watch, with promotion to stage. Interactive room for the stage, live stream for the audience, and move people between them. Most vendors in this class, 100ms included, support both.
- Nobody ever speaks back. A stream and a CDN, and worth asking whether it needs to be live at all. A recording plus a chat thread costs a rounding error by comparison.
Getting this wrong does not produce a higher bill. It produces a bill of a different order.
What 100ms Credits Stack With
Video credits stack unusually well because the invoices barely overlap. A live product with AI in it is a four or five vendor pipeline, and each vendor runs its own program.
- Live video credits cover the 100ms leg, the $2,500 on this page
- Cloud credits cover recording storage, egress and the backend orchestrating rooms, and are usually the largest program a startup can hold
- Model credits cover the LLM behind summaries, agents or moderation
- Speech credits cover transcription and synthesis, metered per second of audio
- Observability credits cover the session analytics you need to debug latency complaints you cannot reproduce
Hold three of those five and the live half of your product is funded through year one on grants. Which programs combine cleanly, and which quietly exclude each other, is why AI Perks is a maintained list rather than a folder of bookmarks.
What Founders Get Wrong About Video Credits
The common failure is treating a video credit as a video budget. It funds transport, and transport is usually the smaller half of what a live feature costs to run.
Four patterns, in order of cost:
Rooms that never close. Tabs get closed, phones sleep, reconnect loops leave ghost participants. Without leave handling and an idle timeout, a room bills at full rate with nobody in it. One stuck session can outspend a month of usage.
Recording by default. Recording is a second meter, and the file it produces then costs storage and egress on your own cloud bill for as long as you keep it. Record on an explicit trigger, with a lifecycle policy from day one.
Assuming the credit covers the AI. Transcription, model tokens and synthesis are three separate vendors. Funding transport and not inference solves the cheap part of the problem. Those programs are listed separately on getaiperks.com.
Spending the credit on the prototype. Credit clocks generally run from activation, not from first real user. Building against a free allowance and activating once you have traffic beats the reverse.
One more discipline: decide at 70% of credits consumed what your unsubsidised bill looks like, not at 100%. At seventy percent you can still change publish settings, recording policy and session shape. At a hundred you are negotiating instead of engineering.
How to Get 100ms and Other Communications Credits
Step 1: Start at getaiperks.com and filter to the communications category. 100ms sits there alongside the other real-time video, messaging and telephony programs, each with its current amount and eligibility conditions.
Step 2: Apply for cloud credits the same week. Recording storage and egress land on your cloud bill, not the SDK bill, and cloud programs are far larger.
Step 3: Add model and speech credits if anything in the call is AI. Those meters run independently of minutes and often exceed them.
Step 4: Confirm which programs can be held at the same time. Overlap rules differ by vendor, and AI Perks tracks which combinations hold and which cancel each other out.
Step 5: Set idle timeouts and publish quality before you have traffic. Both compound over the life of the credit.
Frequently Asked Questions
How much is the 100ms startup program worth?
Up to $2,500 in credits toward the 100ms live video and audio SDK. How far it goes depends on session shape: audio-only one-to-one calls stretch it a long way, while large HD group sessions with recording burn it fast. Amounts and eligibility are tracked at getaiperks.com.
How does 100ms pricing work?
Usage is metered per participant per minute in a room, with separate meters for recording and live streaming. Cost rises linearly with headcount, so a 25-person hour bills as roughly 1,500 participant-minutes. Audio-only minutes cost considerably less than video. Verify current rates on the vendor's page.
Do I need a video SDK, or is WebRTC enough?
For two-party browser calls on good networks, WebRTC alone is free and genuinely enough. The cliff hits at three or more participants, on networks needing a relay, on cross-border calls, and on mobile. Those four, plus the prebuilt call interface, are what a hosted SDK sells.
Do 100ms credits cover recording storage?
No. The recording meter is part of the live bill, but the file it produces sits in your own object storage and leaves through your own CDN, both billed by your cloud provider. Pair video credits with cloud credits. Compatible programs are listed at getaiperks.com.
What is the fastest way to cut a live video bill?
Three changes, in order of impact: close rooms on disconnect with an idle timeout, stop recording every session by default, and publish video at the quality your interface renders rather than at camera capture quality. All three are configuration, not architecture.
Can I combine 100ms credits with other startup credits?
Yes, and communications credits stack well because the invoices do not overlap. Cloud credits cover storage and egress, model credits cover the LLM, speech credits cover transcription and synthesis. AI Perks tracks $7.7M in credits across 194 companies, including which programs conflict.
Ship the room. Let someone else pay for the first million participant-minutes.