A 10-MINUTE PRIMER
Today, roughly 80% of all internet bandwidth is consumed by video.
It’s already the dominant share of all internet traffic — and AI is making it even more dominant. A growing wave of applications now use AI models to create, transform, and interpret video in real-time, requiring continuous GPU inference on every frame.
But running AI on video is compute-intensive — and expensive. Real-time video AI requires continuous GPU inference — processing every frame through AI models as it arrives, with no room for delay. Centralized GPU clouds aren’t built for this. They’re optimized for batch processing — not the continuous, low-latency inference that real-time video demands.
AI-generated worlds, real-time video analysis, AI-driven avatars and agents — demand for these applications is accelerating. The developers building them — creative technologists, startups, independent builders — need GPU providers that are fast, affordable, and always available.
What if there were an open network purpose-built for this?
What is Livepeer?
Livepeer is a specialized GPU network for real-time video AI. It coordinates a global pool of GPU providers to deliver low-latency AI inference on video streams — frame-by-frame processing, not batch jobs.
The network is open-source, built on Ethereum for coordination and permissionless access, and 60–85% cheaper than centralized cloud alternatives.
Who is Livepeer for?
Developers
building real-time AI video applications — generative worlds, live video analysis, AI avatars, interactive streaming — can use Livepeer to access GPU compute via simple API calls.
GPU Providers
contribute GPU compute to the network by running nodes. Anyone with capable hardware can join and earn fees for running AI inference workloads.
How does Livepeer work?
Meet: Alice
Alice is a developer building an interactive world model — an AI-generated environment that responds to user input in real-time, rendering 30–60 frames per second of generated video, continuously.
Meet: Bob
Bob is a gamer. He’s exploring Alice’s world model — an AI-generated environment that reacts to his every move in real-time.
When Bob plays, every input he makes triggers a new frame of AI-generated video. Alice’s app sends these requests to Livepeer, which routes the GPU-intensive work across the network and returns generated frames in real-time — inference, encoding, and delivery in one pipeline.
Bob’s experience is seamless. But behind the scenes, Alice needs massive GPU compute to keep up — and her app is gaining users fast. How does Livepeer make this affordable?
There are two key roles in the Livepeer protocol that make this work: Orchestrators (GPU providers who run the compute) and Delegators (token holders who help secure the network).
Let’s start with Orchestrators.
Orchestrators
In the Livepeer protocol, GPU providers are called orchestrators. They run the AI models — image generation, video analysis, style transfer, depth estimation — and earn fees in ETH or stablecoins for the work they perform. Anyone with capable GPU hardware can join the network and become an orchestrator.
Sounds good, right? But wait, there’s a catch! In order to earn the right to do this type of work on the network, you must first earn or acquire Livepeer Token, also known as LPT.
Livepeer Token
The purpose of the Livepeer token (LPT) is to coordinate, bootstrap, and incentivize participants to make sure the Livepeer network is as cheap, effective, secure, reliable and useful as possible. In the Livepeer protocol, LPT is required to perform AI inference work on the network. The more LPT you own, the more work you’re able to perform on the network in exchange for fees.
As the network’s usage grows, so does the demand for orchestrators and thus LPT.
Of course, not everyone has the expertise required to perform the job of an Orchestrator. It requires serious technical knowledge and can be a full-time job. What if you’re a Livepeer tokenholder but don’t have the time or expertise to run the necessary infrastructure 24x7?
There’s another set of actors in the Livepeer protocol who play a less active albeit equally important role within the protocol — Livepeer Delegators.
Delegators
Delegators are Livepeer tokenholders who participate in the network by staking their tokens towards orchestrators who they believe are doing good and honest work. You can think about staking like putting a deposit down. When you stake, your tokens become locked up for a period of time and then you can take them back or stake them to a different orchestrator. By delegating stake to high-performing orchestrators, delegators signal to the network which ones are most reliable — the network uses this signal to route video streams to the best orchestrators, ensuring high-quality inference across the network.
You may be wondering, why would a Livepeer tokenholder choose stake their tokens? What’s in it for them?
Rewarding Participation
When a developer pays fees for AI inference on the network, both orchestrators and delegators earn a portion of those fees as a reward for ensuring a high-quality and secure network.
In addition to earning fees, Livepeer mints new token over time, much like Bitcoin and Ethereum block rewards, which are split amongst delegators and orchestrators in proportion to their total stake relative to others in the network.
This has the effect of growing network ownership amongst those who participate and shrinking it amongst those who do not.
It also gives orchestrators a powerful economic advantage over centralized GPU cloud providers since the value of the token offsets what they need to charge developers to break even. This is a key reason Livepeer can offer GPU compute at 60–85% lower cost than centralized alternatives.
52.24%
0.0620% inflation
Neat right? Next, let’s go over how often new tokens are minted.
Rounds & Inflation
Rounds
In Livepeer, new tokens are minted every so-called round. Rounds are measured in Ethereum blocks, where one round is equal to 5,760 Ethereum blocks. In Ethereum, one block is mined on average every 12.69 seconds, which means one Livepeer round lasts roughly 20.31 hours. Assuming the Orchestrator you’re staked to is doing its job, this is how often you can expect to receive reward tokens.
Next, let’s go over the Livepeer inflation rate, or in other words, the way by which the Livepeer protocol determines how many new tokens to mint each round.
Inflation
The current rate of inflation as of today’s round is 0.0620% and there are currently a total of 50,948,531.72 Livepeer tokens in supply. So, if you do the math, a total of 31,568.58 newly minted Livepeer tokens will be rewarded to all participants during the next round.
The cool thing about Livepeer is the inflation rate adjusts automatically depending on how many tokens are staked out of the total circulating supply. Currently, the total supply of Livepeer tokens stands at 50,948,531.72 and of those, 26,613,846.27 are staked. Livepeer refers to this ratio (52.24%) as its ‘participation rate’.
Livepeer presupposes that a target rate of 50% is a healthy trade-off between network security and token liquidity, so in order to hit this target, the protocol incentivizes participation by increasing the inflation rate by 0.00005% for every round the participation rate is below 50% and decreasing it 0.00005% for every round the participation rate is above 50%.
LIVEPEER IS GROWING!
Today, there are 2,485 delegators securing the network, with more and more participants joining every day.
Livepeer is open GPU infrastructure for real-time video AI — specialized, affordable, and built to scale. Here’s how to get involved.
Want to build with Livepeer?
Build real-time AI video applications with affordable GPU compute. Get an API key and start building with Livepeer.
Get Early Access
Want to provide GPU compute?
Run an orchestrator node and earn fees for performing AI inference on video streams. Contribute your GPU hardware to the network.
Provide compute
Want to participate in the network?
Get Livepeer token
Get tokenDelegate stake towards an Orchestrator
Delegate Stake