How CSPs and telcos can launch a GPU cloud in under 30 days
How can a CSP or telco launch a GPU cloud quickly?
A CSP or telco with existing fibre, power, and colocation infrastructure can launch a commercially operational GPU cloud in 30 days using a neocloud software platform like hosted·ai. The platform provides the full software stack: GPU pooling, multi-tenant provisioning, metering and billing, and a white-label customer portal. The CSP brings the physical infrastructure and the customer relationships. The 30-day timeline covers software deployment, first GPU pool configuration, billing setup, and controlled customer onboarding. It assumes the CSP already has at least one rack of GPU nodes available or has sourced capacity through GPU Mesh. Legal and commercial setup runs in parallel with technical deployment, not sequentially.
Why now is the right moment for CSPs and telcos
CSPs and telcos have been watching the GPU cloud market from the sideline for a few years. Most concluded it was a hyperscaler game. That was true for training workloads. It is not true for inference.
Inference is what happens after a model is trained: every response, every summary, every recommendation. It runs continuously, close to users, at lower latency requirements than training. Gartner projects inference will account for 65% of AI-optimised infrastructure spend by 2029 (up from 55% in 2026). McKinsey puts inference demand growing at 35% CAGR through 2030, faster than training.
That workload profile suits what CSPs and telcos already operate: distributed, latency-sensitive infrastructure, close to end users, in jurisdictions where data residency matters. The EU sovereign cloud market is projected to grow from $154 billion in 2025 to $823 billion by 2032, per IDC. North American neoclouds hold 88% of GPUaaS revenue today. That share drops to 72% by 2030 as regional supply builds out. The redistribution does not have a predetermined winner.
There is also a regulatory pressure point that did not exist two years ago. The EU AI Act, now in full enforcement for high-risk systems, creates documentation, auditability, and transparency requirements that are materially easier to satisfy on infrastructure the buyer can verify. GDPR restricts personal data flows outside the EU. The US CLOUD Act means data held by US-headquartered entities remains subject to US jurisdiction regardless of where the rack sits. For enterprise buyers in finance, healthcare, and the public sector, that distinction is not theoretical. It is the deciding factor.
CSPs and telcos are European entities operating European infrastructure under European law. AWS cannot replicate that with a regional availability zone announcement, however large the investment figure.
What is missing: the software layer
Most CSPs have the physical infrastructure AI needs. What they do not have is the software to turn a rack of GPU nodes into a commercially operable, multi-tenant GPU cloud.
Running GPU cloud is not the same as running a VM platform. GPU workloads have different scheduling requirements. Multi-tenancy on GPU is harder than on CPU: you need to isolate VRAM between tenants, manage overcommit ratios, and handle context-switching between jobs. Billing needs to track actual GPU and VRAM consumption, not just time. Customer self-service needs to work for AI developers who expect cloud-native UX, not a ticket-based provisioning flow.
hosted·ai is the software layer. It handles GPU pooling, temporal and spatial workload scheduling, configurable overcommit (2x to 10x), metering, billing, and the customer portal. The CSP or telco deploys it on their infrastructure and configures their products through an admin panel. The whole stack is white-label: customers see the CSP's brand, not hosted·ai's.
The 30-day launch path, step by step
Days 1 to 7: Foundation. Commercial agreement with hosted·ai. Infrastructure audit: confirm GPU node count, networking configuration, and connectivity. Decision on initial product set: which GPU types, which pricing model (reserved, on-demand, or elastic). Source additional capacity through GPU Mesh if needed to supplement owned hardware.
Days 8 to 14: Platform deployment. Deploy hosted·ai on the CSP's infrastructure. Configure the first GPU pool. Set overcommit ratio based on initial workload profile. Connect billing and metering. Configure the white-label portal with the CSP's brand, domain, and pricing.
Days 15 to 21: Product configuration and internal testing. Define the first two or three GPU products (for example: shared inference pool, dedicated H100 instance, bare metal GPU node). Set pricing and billing parameters. Run internal load tests. Verify multi-tenant isolation. Confirm that billing output matches provisioned usage.
Days 22 to 30: Controlled launch. Onboard two to five pilot customers. Verify the provisioning and billing flow under real conditions. Address any issues before opening to inbound demand. Confirm support workflows. Launch.
The timeline is tight but achievable. The main risk is sequential decision-making: waiting for the platform deployment to finish before starting the commercial configuration, or waiting for legal sign-off before starting technical work. Run tracks in parallel.
What about operators who do not have GPU hardware yet?
CSPs without existing GPU nodes can still launch within the 30-day window using GPU Mesh. GPU Mesh is hosted·ai's wholesale capacity network, connecting operators to available GPU supply across regions. The CSP sources capacity from GPU Mesh at wholesale rates, configures it through the hosted·ai platform, and sells it under their own brand. This is how several operators have launched their first GPU cloud product without a hardware procurement cycle.
Alternatively, a partner like Maerifa can handle hardware sourcing, procurement, and finance as a single engagement alongside the hosted·ai software deployment. That removes the need to manage multiple vendor relationships in parallel.
The sovereign AI argument, made plainly
EU regulators are not the only pressure. Enterprise procurement teams are asking the question directly: where does my data sit, who controls the infrastructure, and under whose law does it operate?
A CSP or telco launching a GPU cloud can answer all three questions clearly and specifically. That answer is worth real margin. The customers who care about it are large, sign long contracts, and churn at lower rates than price-sensitive buyers.
The EU AI Act and GDPR together can expose organisations to penalties up to 11% of global annual turnover in the most serious cases. Board-level risk management is now a driver of infrastructure procurement decisions, not just a compliance checkbox. CSPs who can offer documented, auditable, EU-law-governed AI compute are selling something hyperscalers structurally cannot.
What happens after launch?
The 30-day window gets you to first revenue. What comes after is a product and commercial question, not a software question.
The hosted·ai platform scales as you grow. You can add GPU types, expand to new regions, add bare metal nodes alongside elastic GPU pools, and introduce new billing models. GPU Mesh gives you access to incremental supply without incremental hardware CAPEX. The white-label portal lets you build a product identity that is yours, not a reseller badge on someone else's infrastructure.
The operators who move in the next 12 months will be competing from a structurally different position than those who move in 24. That is not a sales pitch. It is what happened in the VM cloud market between 2010 and 2014, when the distance between early movers and late arrivals became permanent.
Book a demo to see the 30-day launch path applied to your infrastructure.