In July, Featherless AI, the serverless platform serving open-weight models, sent a capacity request for dedicated GPUs. Era Compute matched the requirement with a dedicated NVIDIA DGX B300 cluster deployed at CRA, which has now been serving production workloads since early August. Deployments of this size can take months to go from paperwork to live infrastructure. Thanks to Era Compute’s pre-vetted partner network and ready-to-deploy capacity, this one took less than a month.
Customer · Serverless inference platform
A serverless inference platform serving 40,000+ open-weight models through a single API, and an official Hugging Face inference provider. Founded by the team behind the RWKV architecture, Featherless AI raised a $20M Series A co-led by AMD Ventures and Airbus Ventures in April 2026.
- Models served
- 40,000+
- Requirement
- Dedicated DGX B300 cluster, reserved capacity
- Workload
- Production inference
Era Compute
Match · Coordination
Era Compute helps AI companies navigate Europe's fragmented GPU market. We connect customers with the right capacity from 20+ pre-vetted, creditworthy European GPU and datacenter partners, backed by market intelligence tracking GPU models, providers, pricing, and availability to deliver the right solution for their needs.
- Qualified demand
- $500M+ in RFQs
- Available capacity
- H100 to GB300 NVL72 across partners
GPU infrastructure provider
One of the most prominent European broadcast and digital infrastructure operators, with a heritage dating to 1963: 655 communication towers and terrestrial television reaching 99.6% of a national audience, classified as part of national critical infrastructure.
- Certifications
- TIA-942, ISO 27001, NBU Secret
- Delivered
- Dedicated DGX B300 cluster
- Service model
- GPU as a Service
The requirement
Featherless AI serves 40,000+ open-weight models through a single API, using model hot-swapping to switch models on a shared GPU fleet in seconds. Its business offering also provides customers with dedicated GPUs, including B300 capacity, which Featherless needs to secure ahead of demand to keep its customers happy.
As the platform grew, so did the need for dedicated capacity. The requirement was time-sensitive: inference customers do not wait through long procurement cycles. The hardware itself only highlights the urgency.
Each DGX B300 node packs 8 Blackwell Ultra GPUs with 288 GB of HBM3e memory per GPU, built for the high-throughput, long-context inference workloads Featherless serves. Capacity at this level is scarce, making speed to deployment a real competitive advantage.
The team needed a provider that could:
- deliver a dedicated DGX B300 cluster, rather than shared capacity
- commit to a timeline aligned with Featherless AI’s growth
- work directly with its engineering team through deployment and handover
Have a similar requirement? Send us the configuration and timeline.
Request capacityThe match
Era Compute searched its partner network and found a strong match for the requirement at CRA, one of the most prominent European broadcast and digital infrastructure operators: 655 communication towers, television reaching 99.6% of a national audience, and a GPU as a Service line running NVIDIA Blackwell systems in a Tier III certified facility. The configuration fit, and the capacity was available on the required timeline.
GPU procurement comes down to specifics: who can deliver the right configuration, in the right location, at the right price, and on the right date. Era Compute built dedicated market intelligence to track exactly that: real capacity, configurations, pricing, and availability across providers. It also builds partnerships before customer demand arrives, so when a requirement comes in, the infrastructure is ready to be matched and deployed.
After qualifying the match on both sides, the teams aligned on the infrastructure, with Era Compute coordinating the commercial, legal, and technical threads through to deployment.
The parties completed the required documents and moved straight into delivery planning.
One of the DGX B300 crates arriving at the datacenter.
In early August, the DGX B300 cluster began serving production workloads, under a month after the capacity request arrived.
What made this timeline possible
CRA was already a contracted Era Compute partner. The commercial framework was in place before this requirement existed, so there was no onboarding and no contracting round: the capacity request arrived, the fit was confirmed, and delivery planning started immediately.
For infrastructure operators, that is the point of the network. Era Compute brings qualified demand, in this case an inference platform committing to a dedicated DGX B300 cluster on a fixed timeline, directly to the partner with matching capacity. CRA turned available datacenter space into contracted offtake without running a sales cycle.
Apply to become an Era Compute partnerResults
Featherless AI: a dedicated DGX B300 cluster serving production inference, live under a month after the capacity request. Era Compute is already sourcing additional capacity for Featherless AI across the partner network.
CRA: available datacenter capacity converted into contracted GPU as a Service offtake, delivered on the customer’s timeline, generating new revenue and helping unlock further capex for additional servers and future deployments.
"We are very satisfied with our collaboration with the Era Compute team. It has been professional, transparent and enjoyable throughout. This is exactly what a strong business relationship should look like."