The Revolution of Low-Cost AI Computing

A recent peer-reviewed study conducted by io.net has shed light on a silent revolution in the world of artificial intelligence: the use of consumer GPUs to drastically reduce AI costs. 

Published and accepted by the prestigious 6th International Artificial Intelligence and Blockchain Conference (AIBC 2025), the research “Idle Consumer GPUs as a Complement to Enterprise Hardware for LLM Inference” represents the first open benchmark of heterogeneous GPU clusters, tested directly on the decentralized cloud of io.net.

The Core of the Study: RTX 4090 vs H100

At the core of the analysis, we find a comparison between consumer GPUs, such as the popular Nvidia RTX 4090, and powerful enterprise GPUs, particularly the H100. 

The results are astonishing: configurations with four RTX 4090 achieve between 62% and 78% of the computing power of the H100, but at about half the operational cost. In terms of economic efficiency, the cost per million tokens stands between $0.111 and $0.149, with a reduction of up to 75% for batch workloads or latency-tolerant tasks.

Efficiency and Sustainability: A Possible Balance

Although the H100 remain more energy-efficient—about 3.1 times more for each token processed—the study highlights an often overlooked aspect: utilizing idle consumer GPUs allows for extending hardware lifespan and reducing carbon emissions, especially when tapping into electricity grids rich in renewable sources. In this way, sustainability is no longer a distant dream, but a tangible opportunity for those who develop and manage AI infrastructures.

Hybrid Routing: the key to optimizing costs and performance

According to Aline Almeida, Head of Research at the IOG Foundation and lead author of the study, the ideal solution is not to choose between enterprise or consumer GPUs, but to adopt a heterogeneous infrastructure:

“The hybrid routing between enterprise and consumer GPUs offers a pragmatic balance between performance, costs, and sustainability.”

This strategy allows organizations to adapt to their latency and budget needs while simultaneously reducing environmental impact.

Practical Applications: Where Consumer GPUs Make a Difference

The study highlights how consumer GPUs are particularly suited for:

1. Development and testing of AI models
2. Batch processing and latency-tolerant tasks
3. Overflow capacity to manage traffic spikes
4. Research and development environments
5. Chat streaming and embedding, where latencies between 200 and 500 ms are acceptable

Conversely, enterprise GPUs like the H100 remain unbeatable for real-time applications, maintaining a latency below 55 milliseconds even under heavy load.

The Future of AI Computing is Distributed and Accessible

For Gaurav Sharma, CEO of io.net, this research represents a confirmation of the company’s vision: 

“The peer-reviewed analysis validates our core thesis: the future of computing will be distributed, heterogeneous, and accessible. By leveraging both datacenter and consumer hardware, we can democratize access to advanced AI infrastructure, making it more sustainable as well.”

io.net: a global platform for decentralized AI

With the world’s largest network of distributed GPUs and on-demand high-performance computing, io.net positions itself as the go-to platform for developers and organizations looking to train models, manage agents, and scale LLM infrastructures. The integration between io.cloud’s programmability and io.intelligence’s API toolkit provides a comprehensive ecosystem for AI startups of all sizes.

Key Research Points

1. RTX 4090 Configurations: 4 GPUs achieve 62-78% of H100 power at half the cost, offering the best cost/performance ratio per million tokens.
2. Latency: H100 ensures times below 55 ms even under heavy loads; consumer GPUs are ideal for workloads that tolerate latencies between 200 and 500 ms.
3. Sustainability: The use of inactive consumer GPUs reduces carbon emissions and extends the lifespan of the hardware.
4. Flexibility: The heterogeneous infrastructure allows for cost and performance optimization according to specific needs.

A New Era for AI Development

The research by io.net marks a turning point for those working in the artificial intelligence sector. Thanks to the integration of consumer and enterprise GPUs, it is now possible to build powerful, cost-effective, and sustainable infrastructures without significant compromises on performance. An opportunity that promises to democratize AI, making it accessible to an ever-growing number of developers and organizations worldwide.