From GPU mining rigs to AI compute: an early Hivenet provider story

In December 2024, Hivenet interviewed JB, an operator who had moved from a small crypto-mining setup in his apartment to a roughly 30-GPU operation. His story captures a wider shift: hardware assembled for mining could sometimes be repurposed for AI, rendering, and other GPU-compute workloads.

Editor’s note, August 2026: This is a historical interview. “Certified GPU provider” describes the terminology and supply program used when the article was first published. Hivenet’s current public pages describe Compute as Hivenet-operated GPU and CPU capacity built on Policloud-backed infrastructure. The current site does not provide a self-service supplier-enrollment path, so this article should not be used as present-day onboarding documentation.

From crypto mining rigs to AI infrastructure

Crypto mining and AI computing both use GPUs, but the surrounding system requirements differ. Mining often emphasizes sustained throughput from many cards. AI workloads may also depend heavily on VRAM, host memory, CPU capacity, local storage, network performance, software compatibility, workload isolation, and repeatable access to the machine.

That means a mining rig does not become useful AI cloud infrastructure simply by installing a different application. The operator must evaluate the GPUs themselves, balance the rest of the host, validate cooling and power delivery, install a supported driver and runtime stack, and build the monitoring and operational processes needed for customer workloads.

JB’s 2024 interview

The answers below reflect JB’s experience and expectations at the time of the original interview.

How did you get started with GPUs?

JB: I started with crypto mining when it was booming. At first, I had a few GPUs running in my apartment. Over time, I scaled the setup to about 30 GPUs, which occupied a significant part of the space.

That hands-on experience taught me a great deal about cooling, energy use, and hardware management. As mining profitability fell, I started looking at AI and cloud-computing workloads as another use for the equipment and the operational knowledge I had built.

How did you move from mining to supplying GPU capacity for Hivenet?

JB: It was a mix of timing and opportunity. I had known Hivenet founder David Gurlé for about a decade. When Hivenet was developing its Compute platform, David suggested that I join as one of its early GPU suppliers.

The distributed-cloud idea interested me. It offered a way to apply what I had learned from operating mining hardware to workloads used by developers and businesses.

Why did AI workloads appeal to you?

JB: Stability was the main attraction. Mining revenue depended on volatile markets and changing energy costs. I saw businesses looking for GPU capacity for machine learning, deep learning, rendering, and related work, and that demand felt more connected to useful technical projects.

I also liked the idea that the hardware could help startups and developers access compute resources instead of being used only to pursue mining returns.

Why did you operate from Dubai?

JB: Dubai’s location between Asia and Europe was useful for an internationally connected operation. The city was also investing heavily in technology and AI, and power cost remained an important consideration for running many GPUs.

On a personal level, Dubai worked well for a family with connections to Hong Kong and France. It made travel in both directions easier.

Power prices, energy sources, network conditions, and regulatory requirements change. Anyone evaluating a site today should verify the current local situation rather than treat a 2024 operator’s experience as a general rule.

How had GPU demand changed?

JB: GPUs had already moved beyond gaming and mining. Machine learning, deep learning, rendering, and other parallel workloads were creating demand for serious compute capacity. I had also seen other mining operators begin to investigate AI workloads as an alternative use for their hardware.

What did you think other operators needed to understand?

JB: Start with hardware that fits the workload. AI has different requirements from gaming or mining, so the GPU model is only one part of the decision. You must understand what the customer will run and what the complete host needs to support it.

Operating the machines matters as much as owning them. Cooling, power, networking, software, maintenance, and reliability all affect whether the capacity is genuinely useful.

What did you expect from the future of GPU computing?

JB: I expected AI to keep driving GPU and software development. As models became more demanding, developers would continue to need capable compute infrastructure and better ways to use it efficiently.

How was your early experience with Hivenet?

JB: The team was collaborative and focused on connecting available GPU capacity with developers and businesses that needed it. It was rewarding to participate in that early stage and to use equipment and operational experience from mining for a different type of workload.

What a former mining rig needs before it can run AI workloads

 
• Suitable GPUs: The cards need enough VRAM, supported drivers, and the features required by the intended framework or application.
 
• A balanced host: CPU, system RAM, PCIe layout, storage, and network bandwidth can bottleneck a GPU-heavy system.
 
• Power and cooling headroom: Sustained AI workloads can stress a machine differently from its previous mining configuration.
 
• A supported software stack: Drivers, CUDA libraries, operating-system packages, containers, or VM tooling need compatible versions and controlled updates.
 
• Isolation and access controls: Customer workloads require secure credentials, workload separation, and a clear administrative boundary.
 
• Monitoring and maintenance: GPU errors, temperatures, utilization, storage health, connectivity, and failed jobs need operational visibility.
 
• Real workload testing: A system that ran a mining algorithm reliably is not automatically validated for model training, inference, rendering, or multi-GPU communication.

For the customer side of that decision, the GPU virtual-machine guide explains when a full VM is useful, while the current GPU and CPU rental page shows the Compute paths Hivenet offers today.

What this story means today

JB’s experience remains useful as a case study in hardware reuse and operational transition. It shows that mining operators can possess relevant skills in power, cooling, fleet management, and GPU maintenance. It does not mean every mining rig is suitable for AI or that the economics are automatically better.

Hivenet’s current public Compute offering focuses on GPU and CPU instances that customers can launch for AI, rendering, development, and other demanding workloads. If you need compute capacity, explore GPU and CPU rental. If you operate infrastructure and want to discuss a potential partnership, use Hivenet’s contact page rather than the old Discord and console recruitment links from the original article.

‍