Since the surge in AI adoption over the last couple of months, it’s become clear there isn’t enough computational horsepower to go around (something that has become painfully obvious as cloud providers have accrued months-long waitlists for high-end GPU instances). Unlike the brief cryptocurrency-mining GPU craze a few years ago, today’s crunch is driven by real demand from AI research and deployments.
For the sake of perspective, Amazon Web Services has been charging about $98 per hour for an 8-GPU server loaded with Nvidia’s top-tier H100 chips, while some decentralised GPU platforms offer comparable hardware for as little as $3 an hour. Amid this stark 30× price gap, Singularity Compute, the infrastructure arm of decentralised AI pioneer SingularityNET, has announced phase I deployment of its first enterprise-grade NVIDIA GPU cluster at a state-of-the-art data centre in Sweden.
Under a partnership with Swedish operator Conapto, Singularity’s cluster is using cutting-edge NVIDIA hardware (including the next-generation H200 and L40S GPUs) in a Stockholm facility powered entirely by renewable energy.
What’s on offer exactly?
The cluster has is high density by design, and is the foundation for traditional enterprise workloads and the projects of the Artificial Superintelligence (ASI) Alliance, a decentralised AI ecosystem spearheaded by SingularityNET. It offers flexible access modes that mirror the needs of modern AI developers in which companies can rent whole bare metal machines, spin up GPU-powered virtual machines, or tap into dedicated API endpoints for AI inference.
This means an organisation can potentially train entire large machine learning models from scratch, fine-tune existing models on custom datasets, or run heavy-duty inference for applications like generative AI; all using Singularity’s infrastructure.
Operationally, the partnership will be managed by popular cloud provider and NVIDIA partner Cudo Compute, the latter ensuring the cluster’s delivery of enterprise-grade reliability and providing the support mission-important AI projects demand. Speaking of the development, Dr.Ben Goertzel, founder of SingularityNET and co-chair of the ASI Alliance, said: “As AI accelerates toward AGI and beyond, access to high-performance, ethically aligned compute is becoming a defining factor in who shapes the future. We need powerful compute that is configured for interoperation with decentralised networks running a rich variety of AI algorithms carrying out tasks for diverse populations. The new GPU deployment in Sweden is a meaningful milestone on the road to a truly open, global Artificial Superintelligence.”
A similar sentiment was echoed by Singularity Compute CEO Joe Honan who believes the launch is about more than extra compute capacity; rather a step towards a new paradigm in AI infrastructure. He emphasised that the cluster’s NVIDIA GPUs will deliver the performance and reliability modern AI demands, and uphold the principles of openness, security, and sovereignty in how the compute is provisioned.
The Swedish cluster will serve as the backbone for ASI:Cloud, Singularity’s new AI model inference service, developed in collaboration with Cudo. ASI:Cloud provides developers with wallet-based access to an OpenAI-compatible API for model inference, offering a path to scale from serverless functions to dedicated GPU servers.
Early customers are already using the Swedish cluster, with the team behind it hinting that this is only the beginning of additional hardware and new geographic locations to be brought online. For a community that has often been at the bleeding edge of the ongoing AI and blockchain revolution, this deployment seems to be a tangible step toward the goal of decentralised, globally distributed AI infrastructure.
The race for AI compute is under way and heating up
Since the turn of the decade, the tech sector has poured major investments into AI infrastructure, with 2025 alone having witnessed over $1 trillion in new AI-focused data centre projects. Even nation-states seem to be wading in, with France, for example, having unveiled a surprise €100+ billion plan to boost AI infrastructure.
Yet not everyone can spend billions to solve the current compute shortage, and that’s resulted in emergence of alternate approaches like decentralised or distributed GPU networks that use hardware sited across several locations and operators.
If the 2010s rewarded those who accumulated data, the 2020s will reward those who control compute power. In that future, efforts like Singularity Compute’s new GPU cluster embody a growing determination to democratise who gets to shape AI’s next chapter and where the compute might be situated.
