What Is io.net?

Io.net is a decentralized GPU network focused on enabling the development of machine learning (ML) and AI applications built on the Solana blockchain, enabling the monetization of computation power by pooling dormant GPUs.

Key Takeaways

• Io.net’s goal is to encourage the development of advanced ML and AI products by allowing developers access to unlimited compute resources for a fraction of the cost of using resource networks managed by centralized corporations.

• Io.net pools idle GPUs from a range of sources and allows individuals and organizations to buy and access unlimited GPU power to develop ML and AI applications.

• Io.net is powered by the IO token, which can be used for payment for GPU services on the platform, and can also be staked for additional passive income while contributing to the security and decentralization of the network. 

With the growing interest in Artificial Intelligence (AI) and Machine Learning (ML), such applications are becoming more cost-intensive to develop due to the increased competition in the space. A major bottleneck for developers and application users is the availability of hardware and software computing resources: CPU and GPU power.

Most of these resources are supplied by centralized corporations via cloud computing facilities; however, these tend to be more expensive, require long-term contracts, and may also have waitlists for the most popular hardware. Io.net, a Solana-based compute resource project, is focused on unlocking fair access to computing power to empower ML applications by creating a decentralized physical infrastructure network (DePIN) focusing on GPU power.

Introducing io.net

Io.net is a compute resource network that offers GPU power to organizations and individuals developing machine learning (ML) and Artificial Intelligence (AI) models and products. According to the project, it offers developers unlimited GPU power, better pricing, and flexibility. Io.net achieves this by developing a model that allows idle GPUs to be put to good use, tapping into over one million GPUs from independent data centers, crypto miners, and other crypto projects that rent out idle GPU power like Render. 

By leveraging a user-friendly portal for users to access this dormant GPU supply, along with rewarding GPU suppliers, io.net positions itself as a marketplace for idle GPU power. This enables it to garner more GPU power than organizations that rely on centralized self-managed data centers and facilities to provide computing power for their clients.

Io.net cites the tremendous growth in software development, the increasing relevance of Machine Learning (ML) and AI model development, and the challenges faced by builders in this space, especially self-funded projects and projects based in certain regions. These challenges are related to accessibility, security, and affordability. By deploying its solution on the blockchain, io.net lowers the barrier for GPU suppliers and companies via a permissionless portal; and by distributing the financial load of providing GPU power to the network to the different suppliers, it reduces the overall facility cost and in turn the cost of obtaining GPU power from the network. Data from the project suggests that io.net’s pricing is at least 50% cheaper than its centralized competitors.

Team Background

The io.net team was inspired to develop a cheaper and more accessible source of computing resources by the high prices of NVIDIA cards required to deploy their High-Frequency Trading (HFT) algorithm using Ray.io. Since this time, the team has rebranded from Antbit to io.net and transitioned from developing institutional-grade quantitative trading systems to developing a decentralized market for computing resources, mainly for Machine Learning (ML) and Artificial Intelligence (AI) model developers.

The io.net team was founded and previously led by Ahmad Shadid, former quant systems engineer at WhalesTrader and Arabfolio and volunteer contributor at Ethereum Foundation. Co-founder and former COO Tory Green has assumed leadership following Ahmad’s stepping down on June 11, 2024. Members of the project team include:

Tory Green, Co-founder and CEO: Tory Green, a Stanford University graduate, is the current CEO of Io.net and former COO at Tiller Partners and HUM capital.

Basem Oubah, Co-founder and COS: Basem Oubah is an Engineering graduate from Yildiz Teknik University 2022. He is the Chief of Staff of io.net.

Gaurav Sharma, CTO: Gaurav Sharma held positions at Binance, Agoda, Amazon, and eBay. He is the current Chief Technical Officer at io.net.

Garrison Yang, CMO: Garrison Yang is the Chief Marketing Officer at io.net, he held a growth leadership position at Ava Labs and advisory positions at Pulsar Games, 77-bit, and Legendary Foundry Games.

Smiral Rashinkar, VP of Engineering: Smiral Rashinkar is a machine learning engineer. He is the VP of engineering at io.net and has held engineering positions at Koo and Rivi.

Funding

Io.net announced that it has raised $30 million in its Series A funding program at a valuation of $1 billion. The funding round was led by Hack VC, with participation from Multicoin Capital, 6th Man Ventures, Solana Ventures, OKX Ventures, Aptos Labs, Delphi Digital, and The Sandbox. Individual angel investors in the projects include Yat Siu of Animoca Brands, Anatoly Yakovenko of Solana Labs, and Mo Shaik of Aptos Labs.

Now, let’s take a closer look at how io.net works.

How io.net Works

In a nutshell, io.net offers suppliers a marketplace to rent out their idle computing resources to developers who are looking to obtain on-demand computing resources. By using the blockchain, io.net creates a transparent platform for computing power management. As io.net operates on an hourly basis, where users can book clusters for specific periods, io.net utilizes its unique Proof concept: Proof of Time-Lock, which proves that the GPUs were not accessed by any other services that would impact compute power during the time it was rented. 

Work on io.net is priced according to presiding factors like workload and availability of computing resources. The facility also ranks suppliers according to their security compliance and upload/download speed, which also affects the pricing. The platform’s economy is built on the IO token. Suppliers are rewarded for works processed using the computing resources using IO tokens, while developers can also pay for resources obtained from the network with IO tokens.

Io.net Architecture

Io.net is a multi-layered network, each layer plays a role in the network’s operation. These layers include;

The user interface: Suppliers, users, and node operators on the network can manage their operations via their respective user interfaces, which are designed to suit the needs of each type of user.

The security layer: The security safeguards the facility. It consists of firewall protection for network protection, user authentication services, and logging services for tracking network activities.

API layer: The API layer enables connectivity with external platforms and some internal operations. It includes public APIs for websites and private APIs for internal operations.

Backend layer: The backend layer is the core of the network. It manages major operations like cluster/GPU operations, customer interactions, fault monitoring, analytics, billing, and resource usage.

Database layer: Consists of main storage for storing structured data and caching for storing temporary and frequently accessed data.

Message broker/task layer: Manages the task flow on the network via communication portals for suppliers and workers.

Infrastructure layer: The infrastructure layer houses the compute resources on the network. It manages deployments and task operations.

The platform layers are the core of the network and enable io.net’s products to function.

Io.net Products

Now, let’s look at the different products in the io.net lineup. Before you can access any of these products, you’ll first need to create an IO ID. IO ID is the control center for the IO Ecosystem. To access io.net, you’ll need an IO ID. Users will need to sign up with either Google, Apple, X, or Worldcoin and connect either a Solana or Aptos wallet for future payments and withdrawals. 

IO Worker

IO Worker is the software node client for GPU suppliers on the IO network. It enables suppliers to set up their accounts, lend idle GPU power, and manage the resources they contribute to the network. GPU suppliers will also be able to monitor how their GPU is utilized and the earnings from these operations as well.

IO Cloud

IO Cloud is a decentralized cloud computing facility for ML and AI model developers. It creates a connection between compute resource suppliers and developers, allowing developers to access computing resources on-demand. Here, users can monitor their jobs, including the list of workers, type of GPU/CPU used, the remaining compute hours, and more.

IO Explorer

IO Explorer gives the user an overview of the network, ranging from the number of verified GPUs and CPUs, the kind of chips available in the different countries, and the quantity and pricing of the chips available. It also features updates on the blocks computed, and rewards emitted. 

Io.net and AI

The rising demand for AI products and the progression in the abilities of AI algorithms have caused a correspondingl rise in demand for infrastructures required to develop these products. However, io.net notes that traditional setups cannot meet up with these demands or cater to the needs of smaller developers. With its approach to sourcing compute power primarily for AI and ML model developers, it hopes to give developers who use its facility an edge. 

Io.net’s distributed computing allows AI and ML developers to run their applications across multiple cores, in parallel, leveraging diverse computing resources and optimizing the performance of their applications. Using IO-SDK, a fork of Ray.io, developers can easily employ the io.net facility to scale their operations through parallel computing. The IO-SDK is designed to be user-friendly, enabling developers to accurately use the infrastructures developed by the project.

Io.net describes itself as a cheaper, more available, scalable, and more secure option to traditional cloud computing services, especially for AI model developers who require more advanced hardware and software capabilities.

Let’s take a look at how io.net supports different stages of AI and ML model development.

Training

AI model training is a data-intensive procedure. This is because AI models are structured for their specific operations using inferences from huge amounts of data. Io.net enables developers to parallelize this process by exporting the architecture and weights of a trained model across a network of distributed GPUs. Io.net’s parallel computing enables faster and more quality inferences and generally optimal training through the abundance of resources and batched processing across multiple cores. This is in contrast to the sequential computing model used in traditional training procedures.

Tuning

AI models undergo regular modifications to adjust to newer principles or optimize performance. Refining an AI model ensures that it not only maintains peak performance but also that it aligns with the user/developers’ goals. Fine-tuning an AI model involves searching through several hyperparameter settings. Like the training procedure, this is also a data-intensive procedure and benefits from distributed computing. Io.net leverages distributed computing libraries with advanced hyperparameter tuning for this procedure. It enables developers to easily discover and use the best hyperparameter setting for the AI models.

IO Token and Tokenomics

The IO token is an SPL standard token minted on the Solana network, and also serves as the utility token of io.net. According to the project, it is the currency of compute required to keep the facility running. While developers can pay for resources obtained from the network in USDC and other accepted tokens (including IO), GPU suppliers on the network are paid in IO tokens only. Developers paying for GPU usage with the IO token enjoy certain benefits like zero or reduced transaction fees. A 2% transaction fee is applied to payments in any other token. IO holders can also earn extra income by staking their tokens on the network’s nodes. 

A total of 800 million IO tokens will be minted on the Solana network. The circulating supply at launch will be 500 million tokens. 300 million tokens will be reserved for rewards to suppliers for jobs completed using their GPU, which will be distributed hourly over the course of 20 years. About 32 million IO tokens have also been distributed to participants of the ignition rewards program and the project’s social campaigns. IO token currently trades on decentralized exchanges on the Solana blockchain and centralized exchanges like Kucoin, MEXC, and Gate exchange. 

Io.net and Decentralized Computing

Now, let’s look at how io.net compares against other decentralized compute projects in the blockchain and crypto space. 

Io.net vs. Akash Network

Like io.net, Akash network also presents a platform for idle GPU owners to provide computational resources to users. Akash uses a reverse auction system to ensure competitive pricing, where network deployers specify their preferred price, and providers compete to fulfill the request. This lets them offer lower costs and greater choice. 

The Akash ecosystem is supported by validators who maintain network security and integrity by staking the network’s native token, AKT. AKT is also used for governance, as well as for facilitating lease settlements. Other whitelisted tokens can also be used, but AKT is used as a unit of measure for settlements in these currencies. 

Meanwhile, io.net uses a specified pricing model to set the prices of resources on its network. The price paid for GPU obtained from io.net depends on the availability of GPU during that period, the workload, and also the GPU specs (this is determined by their security compliance, device specifications, and upload/download speed). Also io.net is meant specifically for AI and ML model developers. While users can be from other sectors, the facility is built on AI and ML frameworks and optimized for this sector as well.

Io.net and Render Network

Before Render Network announced that it would be supporting works from AI and ML model developers, Render was meant for graphics rendering works as the project’s name suggests. Like io.net, it sources GPU power by allowing device owners to join the network and allow clients to render videos and graphics using their GPU. It also operates via a general pricing model, which is similar to that used by io.net. 

However, Render has a multi-tier pricing structure which is based on speed, security, cost, and node reputation. The higher the tier, the higher the cost, as seen in the table below. 

Lease settlement is only available in RENDER on Render Network, while io.net offers a wider range of cryptocurrencies for payment.

The major difference between io.net and Render networks is the purpose they are built for. Io.net is built with AI and ML model developers in mind, the facility is optimized for their operations. On the other hand, the Render network is a multi-purpose facility, built initially to handle video rendering operations and then adjusted to support AI and ML works. 

As a side note, Render has also partnered with io.net, adding its network of distributed GPU suppliers to io.net. Io.net has allocated 300,000 RENDER to its Early Suppliers Incentive Program.

“Our partnership with the Render Network will give us access to Render’s community of quality consumer-grade GPUs while we expand their nodes’ use cases beyond rendering to ML applications. This partnership will strengthen both of our offerings, and we look forward to work together. We are excited to launch at Breakpoint and expand to meet the needs of the incredible growth of AI and ML.“

— Ahmad Shadid, CEO of io.net

Io.net and Bittensor

Bittensor uses a challenge-based approach to select the best resources for AI and ML developers, which is also known as “Proof of Intelligence”. Participants on the networks run a node on a subnet, and workers on the network are informed of the task while they work towards a solution. Validators review the solutions and generate a challenge to score the different miners’ solutions. This can be applied to diverse tasks like the development of an AI or ML model to solve a problem or even the provision of computing resources for AI and ML model developers. 

Bittensor is a diverse network and not limited to the provision of computing power. In contrast, io.net specializes in sourcing compute resources for AI and ML. The main similarity between io.net and Bittensor is their relevance in AI and ML development. The pricing structure on io.net is defined by the network for the general supplier pool; meanwhile, on Bittensor, subnet owners specify the reward structure for each task and selected participants are rewarded accordingly. However, payment for tasks performed is made using the utility tokens of the networks (IO and TAO).

Final Thoughts

Developers building Machine Learning (ML), AI models, and other advanced computing protocols will definitely benefit from readily available computing resources. This is the vision of io.net and other projects in this sector. Decentralizing such solutions using the blockchain makes for easy access for developers and easy management for providers and the general facility. It presents an avenue for proper utilization of globally available computing power by turning dormant GPUs into functional and revenue-generating assets. Even in its earliest stage, a handful of projects are already adopting io.net. As the project continues to develop, we could see even more adoption of the project or its approach to decentralizing provision and access to computing power.

Having said this, note that this article only reviews io.net for educational purposes and should not be taken as financial advice. Featured projects are not endorsed.