Introduction
As artificial intelligence (AI) gains prominence, AI derivative markets are also experiencing growth. Among these, the GPU (Graphics Processing Units) market stands out as the most dynamic and rapidly expanding sector.
GPUs play a pivotal role in the realm of the AI industry, and they are widely adopted in several key processes of AI training and inference. GPUs excel at parallel processing, which is crucial for AI workloads, where neural networks involve numerous matrix multiplications and activations. GPUs accelerate these operations, making them ideal for training deep learning models.
In addition, GPU systems can scale up to supercomputing heights. By combining multiple GPUs, researchers can create powerful clusters capable of handling massive datasets and complex AI tasks. The scalability feature of GPUs enables faster model training and inference.
In terms of efficiency, GPUs deliver technical calculations faster and more efficiently than CPUs. Their high memory bandwidth allows rapid data movement between the CPU and GPU during model training. Hence, GPUs are more cost-effective for AI workloads, making them more popular in AI tasks.
These features make GPUs the perfect workhorses of AI enabling faster training, efficient inference, and breakthroughs across various applications. Thus, the GPUs demand rises drastically as AI is widely studied and adopted in various industries.
Market Pain Points
The AI revolution has accelerated the move from CPUs to parallel processing with GPUs and contributed to the rise of the GPUs market. However, high-performance GPUs like Nvidia’s H100s and A100s are primarily reserved for large cloud providers and tech giants. This leaves a long tail of underserved AI start-ups, developers, and researchers without access to these critical resources. And it also brings vulnerability to a single point of failure and censorship risks when using these giants’ services.
Sources: https://www.stateof.ai/compute
Moreover, established industry giants such as Meta, Tesla, and Microsoft wield significant GPU power and impose hefty fees on end users. Conversely, GPU consumers find themselves locked into inflexible pricing models, unable to transfer or monetize their dormant GPU resources, all while continuing to incur charges from these dominant providers.
GPU Renting Market
The GPU rental market provides individuals and organizations with access to powerful GPUs without the burden of ownership and maintenance. However, the traditional GPU rental market faces limitations in terms of incentives and flexibility. In response, the decentralized GPU rental market is emerging to address these challenges within the conventional GPU market.
The decentralized GPU renting projects aim to democratize computational power by pooling together GPU resources from various contributors, making them accessible to a broader user base. Unlike traditional setups where computational resources are hoarded by a few, decentralized GPU renting projects promote a more equitable distribution of resources with token incentives to both providers and consumers.
The decentralized GPU renting projects have a variety of advantages over the traditional GPU market, such as:
-
Cost efficiency: GPU prices are more affordable and attractive compared with their traditional peers, especially for AI start-ups.
-
Idle compute capacity: Decentralized compute networks connect entities with idle computing power. These networks aim to mitigate the GPU shortage by efficiently utilizing existing resources
-
Democratization of computational power and global access: Decentralized GPU projects aim to democratize computational power by crowdfunding GPU resources from various contributors and global accessibility allows AI models to be trained on a more diverse and extensive range of data, reducing the risk of bias.
-
Blockchain integration and token incentives: Blockchain allows for transparent resource allocation, revenue-sharing and efficient coordination, as well as flexible payment. It revolutionizes the sales model and creates massive networks. Moreover, tokens are the key factor to bootstrap network growth and user adoption.
By virtue of these inherent advantages, decentralized GPU rental projects offer the potential for a more accessible, efficient, and collaborative GPU resource ecosystem. More importantly, they also play a pivotal role in fostering innovation and driving the adoption of AI computing.
Notable Projects
In the burgeoning landscape of decentralized GPU renting projects, several notable players have emerged. Among them, io.net, Akash Network, and Nosana stand out as the most frequently discussed projects in recent times.
io.net
io.net is a platform that lets you rent out or use global GPU resources for AI applications. It allows you to deploy clusters, earn rewards, monitor performance, and access a decentralized cloud network at low costs and high speed.
Compared with its traditional competitors, io.net offers up to 90% less on compute costs, instant cluster deployment and flexible pricing. Researchers or AI start-ups can easily access idle GPUs with a flexible and affordable solution.
To date, io.net has onboarded over 728k GPUs from all over the world on its own platform and connected to over 14k GPUs from partner platforms like render network and filecoin. Around 431k GPUs are currently online, achieving an impressive over 59% uptime. A variety of GPUs are supported on io.net, ranging from Apple M chips to Nvidia and AMD graphics. Both consumer and enterprise GPUs are available for renting at affordable prices.
In the meantime, the total cluster payment has surpassed $668k with over 80k compute hours served. A total of 470k workers have earned $638k for contributing their idle compute power.
As the major part of the AI industry, AI inference is one of the most computing intensive processes. So far, there are over 259k inferences finished on io.net via 1.29M onchain transactions.
The ignition program of io.net is still ongoing and the $IO token will be announced to launch at the end of April. It's not bold to expect more GPUs and customers to be onboarded in the following weeks and unlock more potential of decentralized GPU markets.
Akash
Launched in 2018, Akash is a well renowned decentralized computing platform built with the Cosmos SDK. Akash Network launched its mainnet as early as 2020, initially focusing on CPU and storage for distributed services. In June 2023, it introduced a GPU service testnet, followed by the mainnet release for GPU distributed computing in the same year. With Akash, you can now access powerful GPUs for various applications. The network’s recent upgrades, such as Mainnet 8, have increased GPU visibility and streamlined the deployment process, making it even more attractive for developers and innovators.
On the supply side, there are around 19 unique providers on Akash network across 7 countries supplying over 15 types of chips. The largest provider is Foundry, which is a DCG-backed company that also does crypto mining & staking. Since the GPU service launch, Akash has scaled to 370+ GPUs reaching 25%+ utilization.
Sources: https://deploy.cloudmos.io/providers
Enterprise grade chips, like A100s, are focused on Akash network, as they are widely adopted in AI workloads. It turns out that these high-performance graphics are more popular on the platform.
Sources: https://deploy.cloudmos.io/providers
The GPU service bootstraps the revenue of Akash network. Total USD spent on Akash rallies from $100k to over $400k after GPU service being available on Akash. With the rapid growth of graphic demand, Akash may step on a new stage in the GPU market and expand its business organically.
Sources: https://deploy.cloudmos.io/providers
Nosana
Nosana is a new platform that stands at the forefront of decentralized computing, particularly catering to the AI and development sectors. Leveraging the latent power of idle GPUs across the globe, Nosana introduces a cost-effective, efficient, and eco-friendly alternative to traditional cloud computing services.
Due to Nosana's phased onboarding of GPU nodes, there are 106 GPU nodes online with 56 nodes still in queue. So far, there are 208,541 inferences finished on Nosana. End of 2023 and the start of 2024 is an explosive period for Nosana business, where tens of thousands of inferences are finished.
Sources: https://explorer.nosana.io/
Interestingly, consumer grade graphics are popular and dominate the workloads on Nosana.
Sources: https://explorer.nosana.io/
In a nutshell, Nosana provides affordable GPUs for AI users and offers GPU owners an income stream. It may be small on the current scale, but it offers a new method for people to cash out their idle computing powers.
Obstacles and challenges
While the decentralized GPU rental market holds long-term promise, it also faces several challenges at the moment, including:
-
Interoperability issues: Integrating decentralized GPU networks with existing infrastructure and protocols can be complex. The collaboration in different GPUs may be limited by the discrepancy of GPU specs. Latency is another problem with interoperability of GPU networks.
-
Scalability: As demand grows, ensuring efficient scaling of decentralized GPU networks becomes crucial. Especially when AI workloads increase, more enterprise grade GPUs are needed.
-
User onboarding threshold: Current GPU provider and consumer onboarding experience is pretty complex and fussy. Simplifying user adoption and providing intuitive interfaces are essential for widespread use.
-
Security concerns: Protecting decentralized GPU resources from attacks and vulnerabilities is a constant challenge. But still, not enough attention has been paid as the market is in its infancy.
Despite these challenges, decentralized GPU initiatives possess significant potential for democratizing computational power and driving innovation in AI, gaming, and other domains. As the market expands and more competitors enter the fray, it’s not unrealistic to expect concerted efforts will be made to address these obstacles.
Thoughts and outlooks
The demand for GPUs has surged due to the fast growth of AI, gaming, and augmented reality sectors. However, purchasing GPUs can be prohibitively expensive for researchers and small start-ups. Consequently, renting computing power has become an economical solution for those in the long tail.
In response to the backdrop, the decentralized GPU rental market has emerged with one of decentralized physical infrastructures (DEPIN). It harnesses GPU resources from individuals all over the world and redistributes them on a global scale. By integrating token incentives and cryptography, decentralized GPU rental projects encourage both providers and consumers to participate in this burgeoning market and benefit from blockchain technology.
Despite several challenges such as interoperability and scalability, remaining unsolved, the decentralized GPU rental market holds immense promise for fostering innovation, collaboration, and equitable resource distribution. It aims to make GPU-intensive industries more accessible to long tail players. The market is still in its early stages, but it has demonstrated boundless potential and will play a significant role in reshaping the distribution of computational power.
Let’s observe how it unfolds!