DePIN x AI - An Overview of Four Decentralized Compute Network

• Akash, Render Network, and io.net are three of the largest decentralized compute networks in the market. Although they all offer decentralized compute services, each has a distinct business focus.
• Bittensor is a decentralized AI project that utilizes distributed compute resources to perform machine learning. Its objective is to directly compete with centralized AI services like OpenAI.
• On the supply side, Akash has a diverse hardware network including CPU, GPU, and storage, while Render has a large number of GPU capabilities. io.net accumulate impressive amount of GPUs from its own network as well as sourced from other platforms.
• Decentralized compute network is a two-sided market, token for every project is used as the medium for exchange in their respective system. Render network and Bittensor implement token burning mechanism to enhance the value accrual.

Different Flavors of Decentralized Computing

Akash v/s Render Network

Both Akash and Render Network are decenntralzied computing networks, they provide a platform where users can buy and sell computing resources for various tasks. 

Akash operates as an open marketplace, allowing users to access CPUs, GPUs, and storage. It offers compute resources that can be utilized for various purposes, such as hosting game servers or running blockchain nodes. On the Akash marketplace, Tenants (those deploying applications) set the price and terms for their desired deployment, and computing resource providers bid on these deployments, the lowest bidder (provider) wins the deployment. This reverse auction auction model empowers users to set price and terms, while Render’s dynamic pricing algorithm adapts to market conditions. 

On the other hand, Render Network is specifically focused on GPU-based 3D rendering. Render specializes in rendering 3D images and operates as a physically distributed GPU network. In this model, hardware providers offer their resources, and the network uses a multi-tier pricing algorithm to determine prices and match users with buyers of services. Render does not function as an open marketplace where users can independently set prices or terms.

Io.net – A new competitor focused on AI and Machine Learning

io.net is a new decentralized network that sources GPU computing power from geographically diverse data centers, crypto miners, and decentralized storage providers to power machine learning and AI computing. It also partnered with existing decentralized computing networks such as Render to leverage the underutilized computing resources available for AI/ML tasks. 

The main differentiated factors for io.net are 1) its focus on AI workloads; and 2) its emphasis on GPU clusters. A GPU cluster refers to a group of multiple GPUs working together as a unified system to handle computationally intensive tasks, such as AI training and scientific simulations. 

Bittensor - A AI-focused Blockchain

Different from other decnetralized computing networks, Bittensor is the leading decentralized AI project. It aims to create a decentralized machine learning marketplace where decentralized AI applications can be built and directly compete with centralized AI project such as OpenAI’s ChatGPT. The network consists of nodes (miners) that contribute computational resources for training and running AI models. 

Bittensor utilizes a subnet structure, a subnet is like an application-specific chain. It currently has 32 subnets and each focuses on specific AI-related tasks, including text prompting networks like ChatGPT in a decentralized manner, image generation AI that can transform text prompts into images, and AI-powered search engine.

Miners play a crucial role in the Bittensor ecosystem, they provide computational resources and host machine learning models to perform AI-related tasks off-chain and generate the outputs. Anyone can join the network as a miner by meeting the minimum hardware requirement. Miners compete to provide the best results for users' enquiry. 

Network Capacity and Performance

Akash originally focused on CPU and has a large number of CPU network capacity. With the rise in AI which leads to huge demand for GPUs, Akash has been adding GPU capacity to its network since Q3 last year. Nevertheless, it still has a relatively smaller number of high-performance GPUs compared to other projects that solely focus on GPU capabilities.

On the other hand, Render Network specializes in providing decentralized GPU-based rendering solutions, which has allowed them to accumulate a significant number of GPUs within their network.  

Both Render network and Akash, being more established projects, have seen consistent increases in usage over time. Akash, in particular, witnessed a notable surge in quarterly active leases after shifting its focus to include GPUs. 

io.net is a new decentralized computing network that launched its beta public net in Nov. 2023. Despite its short history, io.net has accumulated an impressively large number of GPUs by integrating resources from Render, Filecoin, and its own network. io.net has recently announced the support of Apple silicon chips clustering, allowing Apple users to distribute their unutilized computing power to the network, further boost number of hardware in its network. In addition, io.net hasn’t launched its protocol token, the high amount of hardware providers that joined the network recently may probably caused by people who are hunting for potential token airdrops. 

Bittensor is a decentralized AI network and miners are responsible for contributing computing resources to the network. Miners can invest in hardware setup themselves or simply use computing resources provided by cloud providers. It is not directly comparable with a typical decentralized computing network in terms of the number of hardware, Bittensor has now over 7000 miners.

Tokenomics

Decentralized compute platforms function as a two-sided market, where users pay compute resource providers. Akash, Render Network, and Bittensor have all introduced their respective tokens, which serve as a medium of exchange within their ecosystems. Both Render and Bittensor have implemented burning mechanisms to enhance the value accrual of their tokens. Further details regarding the utility of these tokens are discussed below.

Akash

Akash is a standalone PoS blockchain, and $AKT is the native token used for staking to secure the network and pay the network fee. The token also functions as a medium of exchange in the ecosystem, when users settle transactions or lease on Akash, $AKT is the primary unit of measure for pricing. As a PoS chain, Akash needs to produce block rewards for validators using $AKT emission, the current inflation rate is about 14%.

Akash currently charges 4% for payment in AKT and 20% if the payment is in USDC, the take rate goes to community pool. The specific use of the community pool capital is not decided and potential use could include public funding, incentives, or simply burn these tokens. 

Render Network

Render Network has migrated to Solana from Ethereum, the protocol token RNDR is used for exchange value within the Render ecosystem, where creators and users pay for rendering jobs using the token. 

To balance the dynamic between the supply of computing resources and the demand for them, Render has implemented a Burn and Mint Equilibrium (BME) mechanism. 

When the demand (the rendering jobs) is larger than supply of computing resources, RNDR tokens are burned, resulting in a deflationary effect. Conversely, if the supply is larger than demand, more RNDR will be minted, making the token inflationary. Currently, there is not enough demand, causing RNDR to be inflationary at the moment. 

Bittensor

The native token $TAO for Bittensor served as a medium for accessing network services and as the core reward mechanism. $TAO has a maximum supply of 21 million, and 7,200 tokens are generated daily to reward miners and validators. Bittensor implements a token issuance halving mechanism where the issuance rate halves once half of the total supply has been distributed. After the first halving, the subsequent halving will happen once the half of the remaining token supply is distributed, until reach the 21 million maximum supply. 

While the daily issuance rate of 7,200 TAO is fixed at the current period, the timing of next halving is not predetermined due to the token recycle mechanism in place. This recycle mechanism burns the already issued token, which effectively delays the point of time at which the half of the total supply is reached. 

To register into the network, miners and validators are required to recycle (i.e. burn) TAO token. These burned tokens are taken out of the circulating supply and can be mined again. The network periodically de-register miners and validators who are not competitive enough to provide AI tasks, making the registrationa repeated cost. This dynamic creates a consistent demand for TAO and continues to burn the token. 

The originally planned first halving date is in January 2025, but the current halving date has been postponed to Oct 2025. This suggests that a significant amount of TAO tokens has already been burned.