How io.net is unlocking the hidden layer of GPU pupply for AI
In the 19th century, the discovery of gold in modern-day California led to a massive influx of manpower, machinery, and capital that completely reshaped the landscape of whole regions.
In the last decade, a new “Gold Rush” for Bitcoin, the digital gold, brought crypto and Web3 to the forefront of the world stage. In today’s technological landscape, if Bitcoin is digital gold, digital oil is most certainly computing power. Also referred to as compute, available computing power has grown orders of magnitude. However, the AI revolution, especially the demand for consumer AI, has led to enormous shortfalls in compute capacity relative to demand.
At the heart of this whirlwind stands NVIDIA. In 2023, the NVDA stock price experienced a remarkable surge of 239%. The company also reported an incredible $60 billion in revenue in its latest fiscal year, with net income reaching nearly $30 billion.
NVIDIA and other GPU companies have capitalized on the growing adoption of AI, benefiting from the soaring demand for graphics processing units (GPUs) capable of handling AI/ML workloads. Originally intended to enhance computer graphics and image processing, GPUs have emerged as the dominant hardware choice for applications like parallel computing, a crucial element for generative AI models.
The sharp increase in GPU demand has sent prices soaring while NVIDIA grapples with production timelines, delaying or even completely sidelining many AI startups from accessing this critical resource as larger tech giants have the buying power to hoard supply. This has led to immense value accrual to centralized cloud providers like Amazon, Microsoft, and Google. Still, even these major providers are unable to meet the demand in the market - leading to high prices, long wait times, and limited choice for builders and innovators.
io.net disruption through decentralization
Today, AWS is the No.1 cloud computing provider, with Azure and Google Cloud taking 2nd and 3rd place respectively.
Cloud computing enables businesses and individuals to access on-demand computing power, storage, and a multitude of other applications. Businesses no longer need to set up their own physical infrastructure, which saves time and money.
The rise of AI has led to shortages and wait times for small and medium AI startups to access the cloud. Wait times can stretch to almost a year, and users have very little choice regarding GPU hardware, location, security level, latency, and other options. Cost is almost always the most important factor for AI startups, and projects can easily spend hundreds of thousands of dollars monthly on training and inferencing.
The demand for computing has doubled roughly every 3-4 months for the past decade, and as competition for GPU resources heats up, there is a significant opportunity for companies that can step in and solve the shortfall.
Why io.net is the internet of GPUs
Back in 2020, while establishing institutional-grade quantitative trading systems, Ahmad Shadid, CEO of io.net, encountered a common issue faced by many ML startups today: the high cost of renting GPUs from existing cloud providers. Out of necessity, he devised a distributed GPU system to mitigate operational expenses.
After the launch of ChatGPT, he realized that the system he had built was similar to OpenAI’s network architecture, and there was an opportunity to bring AI/ML capable computing networks to market for the rush of new companies entering the space. Following a victory at a Solana hackathon in April 2023, io.net was launched a few months later, in November of the same year, with the mission to build an accessible and decentralized network of GPUs that anyone ranging from enterprise clients to individuals can tap into.
The centerpiece of io.net is the IO Cloud, a distributed cloud network that developers can use to quickly deploy decentralized ML clusters. The IO cloud consists of two sides - the Workers, who provide their hardware, and developers willing to purchase on-demand compute. There is one key element that differentiates this product from other cloud providers like AWS, Azure, and Google Cloud - the network is completely permissionless. Anyone can deploy ML clusters or provide hardware to the network and receive rewards.
IO Cloud offers numerous advantages over established competitors, including rapid and straightforward onboarding, selecting hardware and geographic regions, and transparent payment processes without requiring long-term commitments. Pay for what you use, not a penny more.
The network already boasts an impressive supply of hardware, consisting of over 95 thousand GPUs and more than 1 thousand CPUs, situated all around the globe. Numerous crypto projects have partnered with io.net, such as Filecoin and Render, which have dedicated their excess capacity to the network. Other noteworthy partners include the AI network Ritual and a data storage protocol called Banyan.
The IO cloud can be accessed through a unified dashboard. The goal is to provide an interface that is so intuitive and easy to use that even your grandma can deploy a Decentralized Mega-Ray Cluster. Okay, that might be a stretch, but You get the point. Through this portal, AI developers and GPU suppliers (Workers) can view real-time statistics, manage active clusters, check on the health of provided hardware, and even view the explorer to track the overall performance of the whole network.
Orchestration layer for AI
The io.net architecture is modular, consisting of multiple layers, each with its specific function. Central to this system is the IO-SDK, a proprietary, specialized fork of Ray. It offers a suite of open-source templates and tools utilized by industry giants like OpenAI and Uber to drive their AI models.
An overview of io.net infrastructure
Starting from the top, the UI layer serves as the primary interaction point, prioritizing user-friendly design for seamless navigation. Next, the Security Layer ensures that all interactions with the system are secure and reliable, with features such as a Firewall that protects the network and an Authentication Service that validates users.
The API Layer functions as a central communications hub, providing Private, Internal, and Public APIs for third-party connectivity. The Backend Layer is responsible for managing Providers, operations, interactions, monitoring, and more, making it an essential system component. The Database Layer is where data is stored, with Main storage for structured data and Caching for temporary data.
The Message Broker/Task Layer orchestrates asynchronous communications and task management. Last but not least, the Infrastructure Layer is the system's foundation, providing GPU Pool management, deployment management, and data storage solutions. It also handles GPU performance monitoring and manages computations and machine learning operations.
This layered approach is a deliberate design decision by io.net, intended to mitigate single points of failure and facilitate the addition or removal of stack components if better alternatives arise.
How to buy $IO, a mechanism for exponential growth
In recent months, we’ve seen an elevated interest in projects that fall under the category of Decentralised Physical Infrastructure, or DePIN for short. The excitement for these projects has its merit because, for a long time, the biggest criticism of blockchain and crypto has been that this technology has no connection to the real world.
Many believe that for crypto to go mainstream, it must provide real value to everyday people. By fusing blockchain technology with physical infrastructure like mobile phones, vehicles, cell towers, and, in the case of io.net, GPUs, these projects stand out among the crowd with an actual product for users outside of our small eco-chamber.
By building on top of crypto rails, io.net gains a trustless, transparent, and instant settlement layer, leaving out much of the inefficiencies and middlemen that legacy systems are known for.
In the initial years of any startup, growth is the defining focus. It involves investing all generated revenue back into the business and foregoing short-term gains with the anticipation of yielding significantly larger profits in the future.
Remember, it took Amazon almost a decade to turn profitable. That was a deliberate strategy from Bezos to expand the business as fast as possible and gain a significant market share that persists to this day.
This leads us to the biggest advantage io.net and other DePIN networks have over their Web2 competitors: tokens as a crypto-economic shared incentive mechanism to achieve exponential growth while aligning the interests of all participants in the network.
Tokens are a great mechanism that aids businesses in growing and expanding rapidly in their pursuit of finding PMF. By combining strong base unit economics of the AI/ML compute business with token economics, io.net can attract more users and developers to join the network and build out the necessary infrastructure, supercharging growth.
In the case of io.net, the growth and underlying economy will be facilitated through the native token $IO, and the vision is for this to become the currency for AI compute.
The $IO token will function as the primary unit of payment within the network. Developers seeking to utilize the network can use $IO to pay for compute and in the future use $IO to unlock advanced features in the IO Models Store (such as cloud gaming or pixel streaming) and or add-ons like serverless inferences.
On the flip side, Suppliers offering their computing power to the network will receive both IO and USDC as earnings for dedicating their hardware to the network. To foster the growth and sustainability of the available computing power, the network will compensate Suppliers even when their machines are idle, ensuring an on-demand supply of GPUs and CPUs.
On top of that, developers will be able to gain access to the powerful IO-SDK, which allows them to create and deploy AI models. Whenever others utilize your model, you will be rewarded with $IO for your efforts.
All of this implies that as the network gains more users and interactions, the demand for $IO will exceed the emissions, leading to an increase in value over time. Another avenue that could potentially boost the value of the $IO token is the team's consideration of launching a separate L1. Given the already extensive network of nodes worldwide, the chain could have the largest validator set globally.
blocmates' final thoughts on io.net
In the Wild West of Crypto, what we often see are projects that are solutions looking for a problem. io.net is the opposite - it’s a solution for an existing problem, and it uses crypto to supercharge its growth and adoption.
As we strive for public recognition, our focus should be on championing tangible products like io.net that are capable of solving real problems and generating real, sustained revenue.
io.net comes at just the right time and is poised to be a market-disrupting product, delivering significantly cheaper pricing models while maintaining negligible, and sometimes even zero, performance loss when compared to the incumbents.
With the token launch set for April, users have various opportunities to qualify for the airdrop. Every day degens can complete GALXE quests and participate in Discord AMAs. For more experienced individuals, contributing their GPU to the network is the best way to earn rewards.
https://x.com/ionet_official/status/1762904956057649416?s=20
With the whole world about to go absolutely bananas over crypto, we are more prepared than ever to participate in conversations about the significance of blockchain technology, highlighting the advantages that decentralized projects like io.net offer over their Web2 counterparts.
Disclaimer: A handful of the blocmates team have invested in io.net, having begun our research and work with the protocol’s team. However, that does not include the author of this article.
References:
https://www.calcalistech.com/ctechnews/article/rkd0n3v3t
https://www.investopedia.com/terms/m/mooreslaw.asp
https://www.technologyreview.com/2020/02/24/905789/were-not-prepared-for-the-end-of-moores-law/
https://www.geeksforgeeks.org/real-world-applications-of-cloud-computing/?ref=lbp
