Skip to content

An operations team that happens to sell compute

6block has been running large GPU fleets since 2019. The workloads changed, first cryptographic proofs and then AI, but the hard part never did: keeping thousands of accelerators busy, healthy and accounted for.

Where the discipline came from

Proof-of-work networks are a brutal school for infrastructure teams. Revenue is a direct function of sustained throughput, every idle GPU is measurable loss, and there is no support contract to escalate to. We spent six years there and won competitions decided purely on operational execution: first place in the Filecoin Space Race, second in the Aleo incentivized testnet.

That work was never really about cryptocurrency. It was about GPU kernel optimization, cluster scheduling, thermal and power management, and monitoring systems precise enough to catch a degrading card before it took a job down. All of it transfers directly to AI training and inference.

When we moved the fleet to AI workloads, we open-sourced the control plane rather than keeping it as a moat. LLMFabric, OpenModel and the OpenModel Gateway are public, Apache-licensed, and are the same software we install when a customer needs the deployment inside their own network.

The floor of a running GPU hall with the room lights off, lit only by indicator lamps

Timeline

  1. 2019

    6block founded

    Started as an infrastructure team operating distributed compute fleets, with a focus on squeezing real throughput out of commodity hardware.

  2. 2020

    First place, Filecoin Space Race

    Won a global competition decided purely on sustained storage and sealing throughput: an operations result, not a marketing one.

  3. 2022

    Second place, Aleo incentivized testnet

    Rewrote the GPU proof implementation to multiply proving throughput over the reference client. Our workers remain open source and in use.

  4. 2024

    Pivot to AI compute

    Redirected the GPU fleet and the operations practice from cryptographic proofs to AI training and inference workloads.

  5. 2025

    Open-source inference infrastructure

    Released LLMFabric, OpenModel and the OpenModel Gateway: cluster management, GPU time-sharing and metered routing, in the open.

  6. 2026

    Compute as a Service

    Serving AI training, inference, fine-tuning and on-premise deployment as a product, and contributing capacity to the Gonka decentralized compute network.

The infrastructure is public

You can read how our clusters are managed before you buy anything from us.

Where our capacity shows up

Beyond direct customers, we contribute compute and engineering to open networks.

Gonka
Compute contributor and infrastructure engineering for the decentralized AI compute network, alongside Gcore, Bitfury and Hyperfusion.
Filecoin
OpenModel turns storage-provider GPUs into inference capacity without compromising proof deadlines.
vLLM & SGLang
We run both in production, benchmark them publicly and report findings upstream.

Talk to the people who run the clusters

Email reaches the engineering team, not a sales queue.

[email protected] Read our code
What are you trying to run?

Select all that apply.

Pick one or more above and we will prefill the enquiry for you.