GPU Compute Brokerage
GPU compute, handled.
H100 capacity for AI teams who'd rather ship models than manage cloud accounts. We source across providers, handle the operations, and bill you in one currency on net-30. Currently onboarding pilot customers.
Join the waitlistHow it works
Tell us about your workload.
You describe what you need to run — model size, duration, region, performance requirements. We match you to the right capacity tier.
We source and provision the capacity.
We route across Vast.ai, Lambda Labs, and other trusted providers to get you the right H100 instance at competitive pricing. You don't manage provider accounts.
You run your workload.
SSH access in your inbox within the SLA. If capacity gets reclaimed, providers go down, or anything else goes sideways — that's our problem to handle, not yours.
Available packages
Pricing is indicative. We confirm availability and exact rates when we reach out.
H100 Spot
Batch inference, evaluation jobs, training experiments
24-hour block · 1× H100
€78 total
Provisioning within 2 business hours
H100 Reserved
Sustained training, fine-tuning, longer runs
7-day block · 1× H100
€588 total
Provisioning within 4 business hours
Multi-GPU Training
Distributed training, multi-GPU workloads
48-hour block · 4× H100
€624 total
Provisioning within 8 business hours
Why teams choose Fjorde over going direct
Capacity that doesn't fluctuate
We route across multiple providers, so when capacity tightens or prices spike at one provider, your workload still runs at the price we quoted. You never see the volatility.
Failures are our problem, not yours
When an instance gets reclaimed, a provider goes down, or capacity vanishes mid-run, we handle it. Re-provisioning, failover, and support — all on our side. You stay focused on the work.
One invoice, EUR, net-30
One bill instead of five. EUR with VAT handled correctly. Net-30 terms instead of upfront credit at each provider. Procurement-friendly for teams with finance functions.
Not for everyone.
If your team has dedicated infrastructure engineers and you're comfortable managing multi-cloud directly, Fjorde isn't the right fit — and we'll tell you on the first call. We're built for AI teams where compute is meaningful but infrastructure isn't your core competency.
Fjorde is built by Kasper Tulinius Daae, founder of Oparko (acquired April 2026). Currently solo with AI-augmented operations during Phase 0 — team expansion follows first paying customers. Based in Roskilde, Denmark.
Be among our first 10 pilot customers.
Currently onboarding selected AI teams for our pilot phase. Tell us about your workload — we'll be in touch as we onboard.