Venice.aiMost AI platforms charge per token, creating unpredictable costs that scale with usage. Venice introduces a fundamentally different model through its VVV token staking and Venice Compute Units (VCU).
The Venice token (VVV) combines two powerful benefits for stakers: daily AI inference allocation through VCUs and ongoing staking yield through emissions. This dual incentive structure rewards both active users of the infrastructure and long-term holders supporting network growth.
This guide explains how VCU works and why they matter for users of our private, uncensored AI API infrastructure.
The Venice Token (VVV) and Staking Benefits
VVV is the token that powers Venice's AI infrastructure and serves a specific purpose: providing access to AI inference resources through staking.
The total supply is 100 million tokens, with 14 million new tokens created annually through emissions to incentivize staking and fund network expansion.
Network Utilization and Emissions
The distribution of staking rewards through emissions adapts to network utilization:
At 0% utilization: 80% to stakers, 20% to Venice
At 50% utilization: 20% to stakers, 80% to Venice
At 100% utilization: 80% to stakers, 20% to Venice
Importantly, Venice team doesn't receive emissions directly - they go to the company, maximizing when network utilization is at 50%.
This structure incentivizes Venice to maintain optimal network capacity while ensuring room for growth. The 50% target represents healthy demand with sufficient headroom for usage spikes.
Recent optimization has led to a 14x increase in VCU per token by changing how capacity is allocated. Rather than dividing capacity across all stakers, it's now divided among active API users, making the system more efficient for those actually using the infrastructure.
Comparing Venice API to Traditional AI APIs
To understand how Venice's model differs from traditional AI APIs, let's examine additional differences in approach and cost structure:
Understanding Venice Compute Units (VCU)
Venice Compute Units (VCU) are not tradeable tokens, but the standardized measurement system for AI inference capacity across different models and tasks. VCUs are a “currency” used to pay for inference through the Venice API, with pricing established per 1M input/output tokens.
Think of VCU like a big bucket of inference - when you stake VVV tokens, you receive rights to a corresponding share of Venice's total computing power. As you use the Venice API, you “spend” some of your daily VCU allotment. At the start of each epoch (daily), your VCU allotment will reset to the full amount.
Your VCU allocation is calculated by dividing your staked VVV by the sum of the "total active stakers", i.e. those who are staking and have made an API call in the last 7 days. That gives you the VCU allocation ratio that is then multiplied by the current network capacity. Your VCU allocation refreshes every day at midnight UTC, providing consistent access based on your staked amount.
Although VCUs are the best way to access the API, Venice also allows for direct purchases of inference using USD. For users looking to test, Venice offers access to an “Explorer Tier” of API using with significantly reduced rate limits for PRO users who have not yet staked VVV tokens.
Currently, Venice's total capacity is 181,480 VCU per day, visible on the staking dashboard.
Each VCU represents a share of the computing credit that can be used across any supported model. Your daily allocation is determined by your share of total staked VVV tokens - if you stake 1% of all VVV, you get access to 1% of Venice's daily inference capacity.
Different models require different amounts of VCU based on their complexity.
Chat models are priced per million tokens, with separate pricing for input and output tokens. While the price is per million tokens, you will only be charged for the tokens you use. You can estimate the token count of a chat request using this calculator.
Image models are priced per image. For the moment there is no per-model image pricing. This will change in a future price update. These costs were determined based on computational requirements and are calibrated to provide fair access across different types of usage.
VCU: Your Gateway to Private, Uncensored Inference
We are actively developing dynamic tools so you can calculate how much VCU is needed for your specific use-cases and model usage, so keep an eye on our channels for updates.
This is just the beginning. As Venice expands its infrastructure, total VCU capacity increases. This means the same amount of staked VVV provides access to more computing power over time, creating a virtuous cycle:
Infrastructure expansion increases total VCU capacity
More VCU per staked VVV increases VVV utility
Greater VVV utility drives demand for staking
The combination of VVV staking and VCU allocation represents a fundamental shift in how users and agents access AI infrastructure. Instead of unpredictable per-token pricing, users and agents can stake VVV to obtain ongoing inference rights while earning yield. This creates predictable costs for high-volume users while aligning incentives between Venice and its community.
The network's 14 million annual emissions allocation adapts to balance growth of API capacity with rewards to stakers. When utilization is low, 80% flows to stakers. At peak efficiency around 50%, Venice directs 80% toward expanding infrastructure. At high utilization, the split shifts back to favor stakers, ensuring the network grows sustainably while maintaining strong incentives at every stage.
To stake VVV tokens, simply head over to venice.ai/token, and click “Stake”. You will see your staked balance, your APR, your current rewards and your available VCU capacity. You can claim and restake rewards at any time. When unstaking, there is a 7 day waiting period before tokens can be claimed.
Ready to access Venice's growing AI infrastructure? Stake your VVV now.
Disclaimer: The VCU system is actively being refined by the Venice team. Numbers and calculations are preliminary and subject to change as we optimize the network based on real usage patterns.
Back to all posts