Walk down any data centre aisle and you’ll hear it.

Idle racks. Idle watts. Bursts of activity during working hours. Silence at night. Bills that don’t sleep.

In the first cloud wave, service providers paid for power and compute 24/7, even when workloads ran for just a few hours. Most saw it as overhead.

One group didn’t.

They saw stranded supply and monetised it.

Years before generative AI was in pitch decks, they built a marketplace around underused infrastructure. Turned cost into opportunity. Standardised the messy parts. 

Then, when the GPU flood arrived, they rebuilt the stack for AI, from provisioning to orchestration, and added managed services to keep workloads stable, secure, and production-ready.

And now?

They’re one of five NVIDIA Cloud Partners in the UK.

Who are they? 

CUDO Compute.

Want the GPU breakdown without the reading? The Audio Companion does it for you, but only if you’re subscribed. If you can’t see it below, click here to fix that.

CUDO Compute started with an operator’s headache:

Paying for power and kit around the clock while workloads typically only ran during working hours.

Founder Matt Hawkins had already built C4L into the UK’s largest independent data-centre marketplace (peaking at ~2% of UK internet traffic). Seeing the opportunity, he took the next step: aggregate the idle, standardise it, and monetise it. That marketplace thesis now runs as a GPU-first infrastructure platform with two consumption paths behind a single control plane.

On one side, teams spin up instances, resize volumes, and fail over across regions.

On the other, they commission bespoke high-performance clusters for AI/ML, simulation, VFX, and HPC.

Teams access the platform via web dashboard, CLI, or API. Under the hood, CUDO delivers on-demand VMs for agility, dedicated bare metal for control, and end-to-end clusters (InfiniBand or Ethernet) built to enterprise architecture patterns. On top of this infrastructure sits their managed stack. Orchestration with Slurm and Kubernetes, 24/7 engineering support, and MLOps and SecOps runbooks let customers focus on models rather than machines.

Why build in this way?

Because while the underlying hardware is often consistent across the market, not all clusters are created equal. Where most providers stop at hardware access, CUDO built a service-first platform.

They staff 24/7 engineering. They run managed services across MLOps, DevOps, and SecOps. They bill per second with cost intelligence. They expose live GPU telemetry (utilisation, latency, memory) so ops teams fix issues before SLAs wobble. They deploy in ISO 27001 facilities across North America, Europe, the UK, and MENA with GDPR/NHS/PCI-DSS/HIPAA alignment and a 99.95% SLA with cross-region replication.

What began as a way to utilise spare compute has evolved into a GPU-first infrastructure platform, built for GenAI. 

CUDO Compute have grown far beyond the “spare capacity” play.

Now they run workloads that can’t afford to fail, offering:

From early-stage research teams to regulated enterprises, customers use the platform to run LLMs, fine-tune weights, spin up ephemeral jobs, or build long-lived training pipelines in production-grade environments, complete with orchestration, monitoring, and SLA-backed support.

Advisors include former PlayStation leaders (Chris Deering; Simon Rutter) and senior AMD leadership (Jörg Roskowetz).

As we move through H2 2025, priorities are shifting, and service is more important than silicon.

That means enterprise customers with enterprise expectations. GPUs alone are no longer enough.

CUDO’s platform already ticks this box, so what’s the first target?

Footprint. They’re targeting deployments in new regions with sovereignty from rack one. That means North America and beyond, with data locality controls and compliance built in, not bolted on. 

The blueprints that worked in the UK get stamped, audited, and repeated across jurisdictions, making it easy for sensitive industries like healthcare, finance, and the public sector to find the fastest route to ROI. 

Then, platform. 

Provisioning is now table stakes. So CUDO is pushing up-stack. Orchestration that abstracts away the complexity. Managed MLOps that shrinks time-to-production. Use-case tooling that lets teams operate models, not machines.

Same control plane, simpler outcomes, greater value and healthier margins. 

Finally, partnership development. 

Telcos bring dense edge. GSIs bring programs and procurement. Venture-backed AI teams bring product velocity. All with CUDO supplying the dependable backbone, aligned to NVIDIA reference standards, keeping the whole thing upright under load.

In 2023 and 2024, builders mattered.

In 2025 and beyond, users do.

Users don’t care who raised the biggest round or hired the biggest name. They want infrastructure that works. Instantly, securely, and without surprises.

That’s the new baseline.

And CUDO’s answer is clear:

Same silicon. Different outcomes.

Because ops, sovereignty, and speed are the real differentiators now.

And CUDO Compute is betting that platform trust, not just hardware, will win the market.

We’ll know if the market agrees soon enough.