PhoenixNode Economics Litepaper
I. What is PhoenixNode?
PhoenixNode is a proprietary lightweight AI compute unit developed by DePIN hardware and investment group JDI Group and its subsidiary Bobber.com (previously main DePIN hardware partner of Helium) in partnership with Phoenix. Its positioning in the AI compute space is a next generation, low-cost AI edge computing device for specific AI, machine learning, and data-related tasks, in which we will outline in more detail below. PhoenixNode is specifically designed for efficiently running particular applications and AI models deployed on Phoenix’s AI Elastic Compute Layer, SkyNet.
A PhoenixNode consists of 2 components:
The Hardware
Hardware specs of the initial version of the PhoenixNode is optimized for vertical-focused machine learning and AI, and utilizes a novel non-Nvidia AI processor provided by a fast-growing Israeli chip manufacturer Hailo:
CPU: 4xA76+4xA55
AI Processor: 2x Hailo-8TM
GPU: ARM Mali-G640 MC4
RAM: 8GB LPDDR4x
Internal Storage: 64GB
The Software
The software aspect of PhoenixNode is equally important and is developed by Phoenix Core Devs in conjunction with Bobber.com. In particular, the software pre-installed on every node insures the management of the node and the delegation of AI and computational tasks on the system. In particular, software modules bundled with the node are as follows:
◼ Terminal for Node System Settings
In this module node operators are able to set and adjust core settings such as connectivity, time zone, and SkyNet binding.
◼ AI & Computational Task Router
This module syncs AI, data science, and computationally-intensive tasks with SkyNet, which determines which particular tasks are allocated to any node. For tasks distributed across multiple nodes, this module also syncs information between nodes.
◼ AI Frameworks Package
The AI model frameworks used are bundled with every node.
◼ Software Updater
Auto updater that upgrades various software and AI framework packages.
◼ Compute Benchmark Analysis
A proprietary software module developed by Phoenix that assesses the computational capability of any device, used to measure node output and for reward calculations.
II. PhoenixNode Applications & Use Cases
As mentioned, PhoenixNode was specifically designed to fulfill a subset of particular vertical use case AI tasks deployed on SkyNet, either used by native applications such as AlphaNet and PhoenixLLM or by SkyNet users who create custom AI models and tasks. Note that we say “subset” as there are particular models that cannot be used on a lightweight compute device like PhoenixNode – one example of which is Phoenix GenAI’s Text-to-Motion.
Types of Tasks Supported (Categories)
Below are the types of tasks supported by PhoenixNode:
● Deep Learning Inference
● LLM Inference
● Image Generation
● Computer Vision Models
● Unsupervised Learning
● Data Processing, Search Algorithms
● Predictive Analytics, Data Science
Specific Apps and AI Models Supported:
Below is a list (non-exhaustive) apps and AI models supported from SkyNet and the Phoenix Ecosystem:
Deep Learning & Deep Neural Networks |
● Transformer ● LSTM (Long short-term memory) ● CNN (Convolutional neural network) ● MLP (Multilayer perceptron) ● KAN (Kolmogorov-Arnold Network) ● GRU (Gated recurrent units) ● Deep Reinforcement Learning ● Gradient Boosting
|
AlphaNet |
● Hyperparameter Search Algorithm (proprietary developed by Tensor ● Signal optimization module ● Minima/Maxima ● WaveML ● ViperAI ● KnightML ● OptimaShort ● AI MeanRev ● AI Directional Risk
|
Unsupervised Learning & Predictive Analytics |
● HMM (Hidden Markov) ● K-Means ● SVM (Support Vector Machines)
|
Generative AI |
● LLM Inference (particular models) ● Text-to-Image ● Image-to-Image |
Computer Vision |
● ImageNet ● CNN (Convolutional Neural Network) |
III. PhoenixNode Compute Mining Mechanism
PhoenixNodes provide compute for various AI and computation tasks delegated from SkyNet via its task allocation and routing algorithm. For serving as a compute resource, PhoenixNodes earn rewards in unit of PHB or CCD (see more below). Nodes receive rewards in two main components: 1) baseline rewards and 2) task-based compute rewards.
Baseline Rewards
Baseline rewards, also known as “idle rewards”, are given for a node’s participation in SkyNet, regardless of whether any tasks are delegated to and completed by it. In order to receive baseline rewards, the node must have normal, stable, and adequate connectivity to the network. Baseline rewards exist to encourage nodes to be on standby for any influx of delegated tasks.
Nodes are equipped with a software module that computes an algorithm which assesses the rate of the computational capacity. These results are constantly synced with SkyNet to insure normal computing performance and connectivity to the network.
Task-Based Compute Rewards
Task-based compute rewards are dependent upon the tasks delegated to every node. Specific tasks are routed from SkyNet to global PhoenixNodes based on a priority system based on the following factors*:
● Random Factor (70% weight)
In the case where conditions for nodes are above a particular threshold considered to be “normal” (computational performance + connectivity), SkyNet randomly assigns tasks to nodes that meet this criteria.
● Node Stability (30% weight)
The stability metric of a node depends on 2 main factors: a) historical task completion metrics b) historical connectivity stability.
* Factors above are specifically designed for PhoenixNodes, not the entirety of Phoenix’s AI node network.
Reward Claiming
PhoenixNodes can be binded to your SkyNet account – node operators will be able to manage all their nodes via SkyNet Control Panel.
Rewards from nodes linked to a particular SkyNet account will also be claimed via the SkyNet Compute Mining Contract using a claim functionality within SkyNet Control Panel.
PhoenixNode Node Management Flowchart
Putting the pieces together to better see the bigger picture, refer to the flowchart below that shows all the main junctures involved in rewards, claiming, and node management.
IV. Compute Mining Calculation & APR Estimates
For nodes in the PhoenixNode network, mining rewards are calculated using a system that is designed to balance 1) the actual value provided to the network 2) fair value for operating a node, and 3) dynamic factors such as token price, and size of network.
For sake of simplicity and easy of understanding – we will focus on the baseline rewards component of PhoenixNode rewards, as this is the easiest to calculate and serves as a base reference rate for node operators. Baseline rewards are obtained from the same base inflation pool as staking rewards, until CCD mining becomes activated (see more regarding CCD below). Hence no additional inflation would incur from the PhoenixNode Network. This would suggest a slight decrease in staking rewards over time to create space for compute mining in the Phoenix ecosystem.
On the other hand, task-based compute rewards are determined by the task allocated to and complete by the node dependent on the SkyNet routing system, and will not be a part of this APR estimate.
We will first review the baseline rate calculation mechanism, and then give certain estimates for scenarios where various factors fluctuate (ie. token price, number of nodes). Unit reward will be denominated in PHB or PHB equivalent (the market value equivalent of PHB+CCD). More information on the specific PHB and CCD components of node rewards will be outlined in the next section.
Below is the basic calculation mechanism for computing node baseline rewards, denominated in equivalent of PHB:
● P: token market value (price)
● F: baseline factor, initially set to 1.45
● N: current number of total connected nodes in the network (assuming full connectivity)
● M: Node surplus threshold for the network
First, we obtain the USD equivalent of tokens mined per node-day by multiplying the latter by P. F is the baseline factor and may be adjusted in cases where there are significant network growth or changes. M is the determined node threshold in that any in excess would be considered surplus, this is currently set to around 2500. We can consider M/N as a dampening factor that will reduce average rewards once exceeded.
The negative exponential on P indicates that real-time token rewards will not be scaled linearly with market price, which is often geometric and exponential in nature. There will be a chart below that demonstrates USD denominated returns as market price P increases linearly.
However, as a part of the SkyNet philosophy and AI elastic compute approach, we will limit the number of nodes based on the extent of compute demand, we do not see the purpose, unless in special circumstances, the need for nodes above the threshold of M.
Below is a chart estimating the baseline rewards assuming N<M, and F=1.45, where Y axis is tokens mined per node-day and X axis is price P. The blue line is the threshold for the lowest value of P for purposes of node reward calculation, set at P=1.
Below is a chart showing the USD value of tokens mined per node-day with the same parameters, with price P on the X-axis and Y axis shows the USD equivalent of tokens mined.
Note that the USD chart is to show the daily and real-time changes in dollar value of tokens mined per node-day. The daily rewards are added to the Compute Mining Contract – ultimately the node operator is claiming a sum of unclaimed node-day rewards added together.
With the PhoenixNodes’ per node cost at around $1300, the chart below estimates the APR (return per annum) based on the mentioned parameters and the default case of M<N:
P (Price) | APR |
$1 | 41% |
$2 | 57% |
$3 | 70% |
$5 | 91% |
$10 | 129% |
$15 | 157% |
$20 | 182% |
$30 | 220% |
$40 | 250% |
Note that the APR estimates are based off of a constant token asset price P throughout a year, but in practice P fluctuates by the day. The exact calculation will be based off of the daily return rate calculated from P and other parameters. A weighted average of price ranges is able to roughly approximate the actual APR.
Based on our estimates, the APR from Q3 2024 to Q3 2025 may average anywhere from 50% to above 150% per node.
As the Phoenix DAO will limit the number of nodes based on network demand, the increasing APR of operating a node will also increase the resale value of the node as well. The current baseline reward calculation mechanism insures a fair level of upside & passive income for node operators as well as a healthy economic model for token value and the ecosystem.
Operating a node is a good alternative to staking & utility staking for participating in the Phoenix AI Ecosystem.
V. PHB & CCD
In section IV we estimated the baseline node APRs in “PHB equivalent”, meaning that eventually rewards will be in a combination of PHB and CCD, and ultimately CCD. However, this will not occur in the near future. Currently rewards in the foreseeable future will be in PHB and shared with the staking pool.
What is CCD?
Computation Credits (CCD) is the native token on Phoenix Layer 1 that is for computational resource use cases on the Phoenix Network. CCD however differs from PHB in that it can be only used for paying for the utilization compute resources or in certain cases fees relating to providing compute.
In staking, utility staking, special holder privileges and governance cases, only PHB can be used. CCD can be thought of as "GAS" on the Phoenix AI Network & Ecosystem.
However, as CCD is not specifically tied to the BNB Chain ecosystem in any manner, it can be deployed across multiple chains other than Phoenix Mainnet including BSC, Solana, and Ethereum, and will likely have DEX options on these chains.
Current swap rate for PHB to CCD is 1:5 – however this will change as CCD starts getting released in later phases. The CCD liquidity plan and schedule is dependent on ecosystem growth, market conditions, and strategic considerations of Phoenix DAO.
As for approximate stages of liquidity, we refer to Phoenix Token Economics update from late 2022:
Given the start of Phase 2 outlined above, in which CCD will start flowing into circulation via staking rewards and other mechanisms (including compute mining), we will see Phase 3 very shortly after as liquidity would be deemed necessary.
However, the Phoenix DAO may determine based on circumstances to move the original Phase 4 Ecosystem Airdrop to before CCD becomes available on various swaps (Phase 3).
VI. Elastic Compute – Matching Supply with Demand
Phoenix differs from most DePIN AI compute infrastructure projects in that we do not encourage nodes and computational capacity in excess (ie. as many nodes, and GPUs as possible). Our view is that networks built on top of the concept of supply and not actual user demand are unsustainable and will shrink over time due to resource providers getting diminishing returns as well as severe token supply inflation and devaluation.
Phoenix takes an approach that focuses on driving demand by value creation – our AI Elastic Compute Layer SkyNet bundles native AI applications and ready to use and deploy AI models for users and our ecosystem partners. We place AI applications and models first, and the DePIN node network layer is merely the supporting infrastructure.
SkyNet is built utilizing an ad-hoc task-based architecture, allowing for AI tasks to be broken down to minute subtasks, unlike the traditional GPU-hour rental DePIN model. This enables further eliminating idle resource waste, minimizing total required nodes to scale AI applications and models.
Hence, Phoenix and JDI Group will limit the production of nodes based on network demand. In the short term, setting the initial total nodes threshold to 2500-3000 is more than ample for all uses cases covered in Section II.
VII. Node Longevity
PhoenixNode is designed to create a lightweight solution to AI and machine learning inference at scale. Due to the growth of the Phoenix ecosystem and evolving AI landscape, PhoenixNode may update future versions with enhanced technology and specs (with limited production). Current versions will continue to operate and be allocated tasks and will not be affected.
SkyNet’s smart routing system enables nodes to continue to execute tasks suitable for them, regardless of the short and long-term changes in AI task and model types. Baseline rewards for nodes with higher computational capacity are expected to be higher. This is calculated in real-time using the computational performance software module mentioned in Section I.
VIII. Delivery & Colocation
For node operators who reside in countries and jurisdictions in which import tariffs and restrictions for the PhoenixNode, or where stable connectivity is an issue, we suggest considering our colocation option, which allows operators to utilize one of our global datacenter partners for managed node operations. The benefit of the colocation option is better connectivity and less time and effort required on the node operator’s behalf.
More information on colocation to follow.
Last updated
