Decentralized Network io.net Utilizes GPU Computing Power to Meet Growing Demand for AI and Machine Learning Services

Reducing Costs and Improving Accessibility

A project originally designed as a quantitative trading system has evolved into a decentralized network that sources GPU computing power to cater to the increasing demand for AI and machine learning services. io.net has developed a test network that aggregates computational power from various data centers, cryptocurrency miners, and decentralized storage providers. By pooling together these resources, io.net aims to significantly reduce the cost of renting GPU computing power, which has become more expensive as AI and machine learning technologies advance.

A Platform for Renting Computing Power

Ceo and co-founder Ahmad Shadid spoke exclusively to Cointelegraph about the network, which aims to provide a decentralized platform for renting computing power at a fraction of the cost of current centralized alternatives. The project was conceived during a Solana hackathon in late 2022 when io.net was developing a quantitative trading platform that relied on GPU computing power. However, the soaring costs of renting GPU computing capacity posed a significant obstacle.

Addressing the GPU Rental Challenge

io.net encountered the challenge of renting high-performance GPU hardware, with the average cost of renting a single NVIDIA A100 card amounting to around $80 per day. To operate for 25 days a month, more than 50 of these cards would cost over $100,000. A solution was found with Ray.io, an open-source library used by OpenAI to distribute ChatGPT training across CPUs and GPUs. By leveraging this library, io.net streamlined its infrastructure and developed its backend within two months.

Unlimited Computing Power at a Lower Cost

Shadid showcased io.net's working testnet at the AI-focused Ray Summit in September 2023, demonstrating how the project aggregates computing power and serves it to GPU consumers as clusters tailored to specific AI or machine learning use cases. This model not only enables io.net to provide GPU compute up to 90% cheaper than existing suppliers but also offers virtually unlimited computing power.

Leveraging Solana's Blockchain for Payments

The decentralized network plans to leverage Solana's blockchain to facilitate SOL and USD Coin (USDC) payments to machine learning engineers and miners who rent or provide computing power. When ML engineers pay for their clusters, the funds go directly to the miners who contributed their GPUs to the cluster, with a small network fee allocated to the io.net protocol.

Introducing a Dual Native Token System

The project's roadmap includes the launch of a dual native token system featuring $IO and $IOSD. Miners will be rewarded with the IO token for executing machine learning workloads and maintaining network uptime, taking into account the dollar cost of electricity consumption. The IO coin will be freely traded in the crypto market and serve as the gateway to accessing compute power, while the IOSD token will act as a stable credit token algorithmically pegged to 1 USD.

Challenging Centralized Cloud Services

Shadid emphasizes that io.net fundamentally differs from centralized cloud services like Amazon Web Services (AWS). He likens AWS to United Airlines and io.net to Kayak, stating that while AWS owns the planes, io.net helps people book flights. He notes that businesses in need of AI computation typically rely on third-party providers due to a lack of in-house GPUs. However, with GPU demand projected to increase tenfold every 18 months, there is often insufficient capacity to meet demand, resulting in long wait times and high prices.

Unlocking the Potential for Miners

According to Shadid, the average cryptocurrency miner stands to benefit by renting out their hardware to compete with AWS. Miners using a 40GB A100 card can earn $0.52 per day, while AWS charges $59.78 per day for AI computing with the same card. Cointelegraph estimates that miners with GPU resources could earn 1500% more by renting their hardware for various AI and machine learning tasks.