The AB side of penetrating io.net: An underestimated AI computing power productivity revolution?

Written by: LFG Labs

What do you think if the core essence of io.net is "grassroots"?

With a funding of $30 million and the attention of top-tier capital such as Hack VC, Multicoin Capital, Delphi Digital, and Solana Lab, it doesn't seem very "down-to-earth," especially with the added labels of GPU computing power/AI revolution, which are synonymous with high-end sophistication.

However, amidst the noisy community discussions, key clues are often overlooked, especially the profound changes that io.net may bring to the global computing power network------unlike the "elitist" positioning of AWS, Azure, and GCP, io.net is essentially taking a democratization approach:

It aims to address the overlooked "mid-tier + long-tail" computing power demands, gather idle GPU resources, and create an enterprise-level, decentralized distributed computing network, empowering a broader range of small and medium-sized users' AI innovations with more incremental and existing computing power resources, achieving "re-liberation of productivity" for global AI innovation at low cost and high flexibility.

The Overlooked Undercurrents of Computing Power Production Relations Behind the AI Wave

What is the core productive resource of this round of AI wave and the future digital economy era?

Undoubtedly, it is computing power.

According to data from Precedence Research, the global AI hardware market is expected to grow at a compound annual growth rate (CAGR) of 24.3%, exceeding $473.53 billion by 2033.

Even without considering predictive data, from the perspective of incremental and existing logic, we can clearly see that in the future development of the computing power market, two main contradictions will inevitably exist for a long time:

• On the incremental dimension, the exponentially growing demand for computing power will far exceed the linearly growing supply side;

• On the existing dimension, the head effect causes computing power to be "squeezed," leaving mid-tier and long-tail players without resources, while a large amount of distributed GPU resources remains idle, leading to a serious mismatch between supply and demand;

Incremental Dimension: Demand for Computing Power Far Exceeds Supply

First, on the incremental dimension, aside from the rapid expansion of AIGC large models, countless AI scenarios in the early stages of explosion, such as healthcare, education, and intelligent driving, are rapidly unfolding, all of which require massive computing resources. Therefore, the current market gap for GPU computing power resources will not only continue to exist but will also continue to widen.

In other words, from the perspective of supply and demand, in the foreseeable future, the market demand for computing power will definitely far exceed supply, and the demand curve will show an exponential upward trend in the short term.

On the supply side, due to limitations imposed by physical laws and real production factors, whether it is improving process technology or expanding production capacity through large-scale factory construction, it can at most achieve linear growth, which means that the bottleneck of computing power in AI development will persist for a long time.

Existing Dimension: Serious Mismatch Between Supply and Demand for Mid-Tier and Long-Tail Players

At the same time, in the context of limited computing power resources facing severe growth bottlenecks, Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP) together occupy more than 60% of the cloud computing market, creating a clear seller's market.

They hoard high-performance GPU chips and monopolize a large amount of computing power resources, while mid-tier and long-tail small computing power demanders not only lack bargaining power but also face high capital costs, KYC entry barriers, and various other issues. Additionally, traditional cloud service giants, due to revenue considerations, are also likely to overlook the differentiated business demands of "mid-tier + long-tail" users (such as shorter, more immediate, and smaller-scale leasing needs, etc.).

However, in reality, outside the computing power networks of cloud service giants, a large amount of GPU computing power remains idle. For example, there are hundreds of thousands of third-party independent Internet data centers (IDCs) globally where training tasks waste resources, including crypto mining farms and massive idle computing power from projects like Filecoin, Render, and Aethir.

According to estimates from io.net, the idle rate of IDC graphics cards in the U.S. alone exceeds 60%, creating a rather ironic supply-demand mismatch paradox: tens of thousands of mid-sized data centers, mining farms, and other operators waste over half of their computing power resources daily, failing to generate effective revenue, while mid-tier and long-tail AI entrepreneurs endure high costs and high barriers to entry for cloud giants' computing power services, with even more diverse innovative demands left unmet.

Droughts kill the dry, floods kill the wet; once we clarify these two basic premises, we can see the core contradiction in the current global AI development and computing power market------on one hand, AI innovation is everywhere, and computing power demand is constantly expanding; on the other hand, a multitude of "mid-tier + long-tail" computing power demands and idle GPU resources cannot be effectively met, remaining outside the current computing power market.

This issue is not only a contradiction between the growing computing power demand of AI entrepreneurs and the lagging growth of computing power but also a contradiction between the vast "mid-tier + long-tail" AI entrepreneurs, computing power operators, and the unbalanced and insufficient supply-demand mismatch, far exceeding the capabilities of centralized cloud service providers to solve.

For this reason, the market is calling for new solutions. Imagine if these operators holding computing power could flexibly choose to rent out their computing power during idle times; wouldn't that allow them to obtain a computing cluster similar to AWS at a low cost?

Building such a large computing power data network is extremely expensive, which has led to the emergence of specialized computing power matching platforms targeting idle computing power resources in the mid and long tail, to mobilize these scattered idle computing power resources and match them with mid-sized model training and large models in specific scenarios such as healthcare, law, and finance.

This not only meets the diversified computing power needs of the mid and long tail but also serves as a complementary solution to the existing centralized cloud giants' computing power service landscape:

• Cloud service giants with massive computing power resources are responsible for large model training, high-performance computing, and other "urgent and heavy demands";

• Decentralized cloud computing markets like io.net are responsible for mid-sized model computing, large model fine-tuning, inference deployment, and other more diversified "flexible low-cost demands";

In essence, it provides a more inclusive dynamic balance supply-demand curve between cost-effectiveness and computing power quality, which aligns better with the economic logic of optimizing resource allocation in the market.

Thus, distributed computing networks like io.net are fundamentally a solution that integrates "AI + Crypto," using a distributed collaborative framework combined with token incentives as a basic economic means to meet the enormous but currently marginalized mid and long-tail AI market demands, allowing small AI teams to customize and purchase GPU computing services that large clouds cannot provide as needed, achieving "re-liberation of productivity" for the global computing power market and AI development.

In short, io.net is not a direct competitor to AWS, Azure, or GCP; rather, it is a "complementary ally" that optimizes global computing resource allocation alongside them, jointly expanding the market pie, just managing different levels of "cost-effectiveness & computing power quality" demand.

It is even possible that io.net, by aggregating the supply and demand from the "mid-tier + long-tail" players, could recreate a market share comparable to the existing top three cloud giants.

io.net: A Matching Trading Platform for Global GPU Computing Power

Because io.net is based on Web3 distributed collaboration + token incentives to reshape the production relations of the mid and long-tail computing power market, we can actually glimpse the shadow of shared economies like Uber and Didi, i.e., a matching trading platform for GPU computing power similar to Uber and Didi.

As we all know, before Uber and Didi, the broadly defined user experience of "instant availability" for ride-hailing did not exist, as numerous private cars formed a vast and chaotic network of idle vehicles. If you wanted to hail a ride, you could only wave by the roadside or request dispatch from the corresponding taxi center company in each city, which was time-consuming, highly uncertain, and belonged to a seller's strong market, making it unfriendly for most ordinary people.

This is also a true reflection of the current supply and demand situation in the entire computing power market. As mentioned earlier, mid-tier and long-tail small computing power demanders not only lack bargaining power but also face high capital costs, KYC entry barriers, and various other issues.

So, specifically, how does io.net achieve its positioning as a "global GPU computing power hub + matching market," or what kind of system architecture and functional services are needed to help mid and long-tail users obtain computing power resources?

A Flexible and Low-Cost Matching Platform

The biggest attribute of io.net is that it is a light-asset computing power matching platform.

Similar to Uber and Didi, it does not engage in the actual operation of high-risk GPU hardware and other heavy assets but connects the retail computing power of mid and long-tail suppliers (many of which are considered second-tier computing power at AWS and other large clouds) through matching, activating previously idle computing power resources (private cars) and the urgent computing needs of mid and long-tail AI demanders (passengers).

On one end, io.net connects thousands of idle GPUs (private cars) from small and medium-sized IDCs, mining farms, and crypto projects, while on the other end, it links the computing power demands of hundreds of millions of small and medium-sized companies (passengers), and then io.net acts as a matching platform for intermediate scheduling, just like a broker matching countless buy and sell orders.

This helps to mobilize idle computing power at a low cost and with more flexible deployment configurations, allowing entrepreneurs to train more personalized mid-sized AI models, greatly improving resource utilization. The advantages are clear: regardless of whether the market is too cold or too hot, as long as there is resource mismatch, the demand for a matching platform will always be the strongest:

• On the supply side, small and medium-sized IDCs, mining farms, and crypto projects with idle computing power resources only need to connect with io.net, without needing to establish a dedicated BD department, nor being forced to sell at a discount to AWS due to small computing power scale; instead, they can directly match idle computing power to suitable mid-sized computing power clients at very low friction costs, even at market prices or higher, thus generating revenue;

• On the demand side, mid-sized computing power demanders who originally had no bargaining power in front of AWS and other large clouds can also connect through io.net's resource pipeline, accessing smaller-scale, permissionless, no-waiting, no-KYC, and more flexible deployment computing power, freely selecting and combining the chips they need to form a "cluster" for personalized computing tasks;

Both mid and long-tail computing power suppliers and demanders face similar pain points of weak bargaining power and low autonomy in front of AWS and other large clouds. io.net revitalizes the supply and demand of the mid and long-tail, providing such a matching platform that allows both sides to complete transactions at better prices and more flexible configurations than AWS and other large clouds.

From this perspective, drawing a parallel with platforms like Taobao, the early emergence of inferior computing power is also an inevitable development pattern of platform economy, and io.net has set up a reputation system for both suppliers and demanders, accumulating scores based on computing performance and network participation to earn rewards or discounts.

Decentralized GPU Clusters

Secondly, although io.net is a matching platform for retail supply and demand, current computing scenarios for large models require multiple GPUs to work together------it is not only about how many idle GPU resources this matching platform can aggregate but also about how closely the dispersed computing power on the platform is connected.

In other words, this distributed network encompassing mid-sized computing power from different regions and scales needs to achieve a "decentralized yet centralized" computing architecture: it should be able to flexibly place several distributed GPUs under the same framework for training based on different scenario computing needs, ensuring that communication and collaboration between different GPUs are very rapid, achieving at least low-latency characteristics that are sufficient for use.

This is completely different from some decentralized cloud computing projects that can only be limited to the use of GPUs in the same data center. The underlying technical implementation involves the "three-horse carriage" of io.net's product combination: IO Cloud, IO Worker, and IO Explorer.

• The basic business module of IO Cloud is clusters, which are groups of GPUs that can self-coordinate to complete computing tasks. AI engineers can customize the clusters they want based on their needs, and it seamlessly integrates with IO-SDK, providing comprehensive solutions for expanding AI and Python applications;

• IO Worker provides a user-friendly UI interface, allowing both supply and demand sides to effectively manage their supply operations on the web application, covering functions related to user account management, computing activity monitoring, real-time data display, temperature and power consumption tracking, installation assistance, wallet management, security measures, and profit calculation;

• IO Explorer primarily provides users with comprehensive statistical data and visualizations of various aspects of GPU cloud. It enables users to easily monitor, analyze, and understand the details of io.net network data by providing complete visibility into network activities, key statistics, data points, and reward transactions;

Thanks to the above functional architecture, io.net allows computing power suppliers to easily share idle computing resources, significantly lowering the entry barriers. Demanders can quickly assemble clusters with the required GPUs and obtain powerful computing power and optimized server response services without signing long-term contracts or enduring the long waiting times common in traditional cloud services.

Lightweight Elastic Demand Scenarios

To be more specific, discussing the misaligned service scenarios of io.net and large clouds like AWS mainly focuses on lightweight elastic demands that large clouds do not offer cost-effectively, including various scenarios such as model training for mid-sized AI startup projects and large model fine-tuning.

Additionally, there is a commonly overlooked universally applicable scenario: model inference.

It is well known that the early training of large models like GPT requires thousands of high-performance GPUs to perform high-quality calculations over long periods with massive data, which is an absolute advantage of AWS, GCP, and other large clouds.

However, once trained, the main computing power demand shifts to the steady flow of model inference, which requires far more computing power than the training phase------the interactions we have with models like GPT in our daily conversations account for 80%-90% of AI computing share.

Interestingly, the overall computing power required for inference is much more stable, possibly needing only dozens of GPUs for a few minutes to arrive at an answer, with lower requirements for network latency and concurrency; furthermore, most AI companies may not train their own large models independently but choose to optimize and fine-tune around a few leading large models like GPT, making these scenarios naturally suitable for io.net's distributed idle computing power resources.

Beyond the high-intensity, high-standard application scenarios of the minority, the broader, everyday lightweight scenarios are also a virgin land waiting to be developed. They may seem fragmented but represent an even larger market share------according to a recent report from Bank of America, high-performance computing accounts for only a small portion of the total addressable market (TAM) in data centers, with only about 5% share.

In short, it's not that AWS, GCP, etc., cannot afford it, but that io.net offers better cost-effectiveness.

The Key to Web2 BD

Of course, the core competitiveness of platforms like io.net that focus on distributed computing power resources lies in their BD capabilities, which is the key to victory.

Aside from the phenomenon of GPU brokers emerging due to NVIDIA's high-performance chips, the biggest problem troubling many small and medium-sized IDCs and computing power operators is that "good wine fears no alley."

From this perspective, io.net actually possesses a unique competitive advantage that is difficult for other projects in the same track to replicate------it has a Web2 BD team directly stationed in Silicon Valley, consisting of veterans who have immersed themselves in the business field of the computing power market for many years. They not only understand the diverse scenarios of small and medium-sized clients but also grasp the terminal demands of numerous Web2 clients.

According to official disclosures from io.net, currently, two to three dozen Web2 companies have expressed their willingness to purchase/rent computing power, willing to try or experiment due to lower costs and more flexible computing power services (some may not even be able to obtain computing power from AWS), with each client needing at least hundreds to thousands of GPUs (equivalent to tens of thousands of dollars in monthly computing power orders).

This genuine willingness to pay on the demand side will essentially attract more idle computing power resources to actively flow in on the supply side, making it easier to achieve a breakthrough between Web2 and Web3, forming a first-mover network scale effect.

The Possibility of a Tokenized Computing Power Ecosystem Settlement Layer

As mentioned above, io.net is based on Web3 distributed collaboration + token incentives to reshape the mid and long-tail computing power market, with the latter primarily adopting a dual-token model of IO and IOSD:

1. Token IO, with utilities including paying for computing power rental fees, providing allocation incentives to IO Workers, rewarding AI and ML deployment teams for continuous use of the network, balancing some supply and demand, pricing IO Worker computing units, and community governance;

2. Stablecoin IOSD, pegged to the US dollar, can only be obtained by burning IO, aimed at providing a stable value storage and trading medium for the io.net platform;

Additionally, io.net also considers supporting suppliers to increase their chances of being rented by collateralizing IO, while demanders can also collateralize IO to prioritize the use of high-performance GPUs, thus allowing the development of a complete ecosystem around the collateralization function to capture the incremental value generated by the entire computing power ecosystem.

This actually raises another question: since io.net aggregates a large amount of idle computing power resources, can it go further and directly combine the tokenization of computing power with Crypto, providing greater on-chain financial possibilities for GPUs?

For example, io.net could potentially build a dedicated computing power chain based on its vast computing power network, providing permissionless, no-entry-barrier tokenized infrastructure services, allowing anyone and any device to directly tokenize computing power (e.g., converting A100, H100 into standardized tokens or NFTs), thus enabling trading, staking, lending, borrowing, and leveraging.

This would create a broad on-chain market for GPU computing power, allowing users and funds from around the world to enter freely and efficiently. We can simply envision two scenarios to glimpse the imaginative space that an on-chain financial market around computing power could have.

1. Securities-type Computing Power Token

For instance, a computing power operator on io.net owns several A100 or H100 GPUs, but they currently need funds or want to cash out early; they can package the computing power value corresponding to this portion of GPUs into an NFT or FT on io.net------where this Token represents the discounted cash flow of the corresponding GPUs' computing power for the next year (or a certain period), priced in IO.

Since the vast majority of ordinary investors do not have the opportunity to directly purchase, hold, or operate AI computing power, these Tokens provide market traders with an opportunity to speculate on the future price fluctuations of computing power. Operators with a large amount of computing power but tight cash flow can also gain financial leverage, flexibly liquidating based on actual needs anytime and anywhere.

During this period, the GPUs behind the Token are operated by io.net, and the subsequent cash flow earned from the corresponding computing power is shared proportionally (Token holders receive 0.9, and operating nodes receive 0.1).

Moreover, because it is a standardized token, it can also be freely traded on CEX or DEX like other Tokens, further forming real-time pricing for computing power that allows for free entry and exit, truly transforming GPU computing power into a globally liquid resource.

2. Bond-type Computing Power Token

Additionally, bond-type Tokens can be issued to raise funds for purchasing high-performance GPUs, increasing network computing power, where the bond principal corresponds to the value of the GPU devices themselves, and the bond interest represents the cash flow income from renting out the future computing power of the GPUs, meaning the potential rental value and future earnings of the GPUs become the market value of the Token, allowing Token holders to obtain real RWA returns.

This effectively creates a vast GPU computing power market for global users, allowing users and funds from around the world to freely and efficiently enter the GPU computing power market without worrying about high barriers and high capital, further fully integrating real GPUs with various decentralized financial products, laying the foundation for more and richer supporting services for users in the future.

More critically, the entire process uses IO as the main transaction/circulation currency, making io.net/IO likely to become the settlement layer/settlement currency for the entire global computing power ecosystem, and the vision of this on-chain financial market around computing power could almost recreate a valuation space similar to the narrative of the io.net decentralized computing power network.

Conclusion

Overall, Web3, as a new type of production relationship, naturally adapts to represent the new productive forces of AI. This is also a simultaneous advancement of technology and production relationship capabilities. From this perspective, the core logic of io.net is to change the production relations between traditional cloud service giants, mid and long-tail computing power users, and global idle network computing resources by adopting the economic infrastructure of "Web3 + token economy":

Providing solutions to the real pain points of AI computing power supply and demand, building a bilateral market that encompasses and serves "mid-tier + long-tail" GPU computing power resources & user demands, optimizing the supply and allocation of computing power resources, and bringing about a great liberation of productivity for global AI development, especially for small and medium-sized AI innovations.

The envisioned goal is undoubtedly grand. If successful, it is highly likely to become the core matching infrastructure and value settlement layer of the global GPU computing power ecosystem, expected to achieve the best valuation premium and immense imaginative space, but it is also undoubtedly full of challenges.