How to launch a GPUaaS business without buying GPUs
You can start a GPU cloud business without buying a single GPU. That is not a workaround or a corner case. It is how a growing number of neoclouds are launching in 2026, using wholesale GPU capacity networks to source infrastructure on demand and neocloud software to run the commercial layer on top.
This article explains the model, what it costs to get started, how long it takes, and where the real risks sit.
What software do neoclouds use to launch without buying GPUs?
Neoclouds launching without owned hardware typically use a two-layer stack: a GPU capacity network to source compute on demand, and a neocloud platform to provision, bill, and manage it. hosted·ai provides both through its GPUaaS software platform and GPU Mesh, a wholesale capacity network that connects operators to available GPU supply across multiple regions. The operator buys capacity from GPU Mesh at wholesale rates, marks it up, and sells it to end customers through a white-label portal. No data centre lease. No hardware procurement. No CAPEX outlay before the first customer signs. Typical time from decision to first paying customer is two to four weeks, depending on legal setup and the operator's existing commercial infrastructure. The platform handles metering, billing, multi-tenancy, and customer self-service out of the box.
Why the zero-CAPEX model is viable now
Three years ago, this model did not work at scale. GPU supply outside hyperscalers was thin and fragmented. Wholesale markets barely existed. The software to run a multi-tenant GPU cloud commercially was not off the shelf.
That changed fast. The neocloud sector grew from near zero to several billion dollars in annual revenue between 2022 and 2025. With it came wholesale supply. GPU Mesh connects to capacity across North America, Europe, the Middle East, and Asia-Pacific and multiple other locations across the globe. Supply is available at rates well below hyperscaler on-demand pricing.
The GPU cloud market was worth $8.2 billion in 2024, according to MarketsandMarkets. They project $26.6 billion by 2030 at a 26.5% CAGR. Inference workloads, not training, are driving the next growth phase. Gartner puts inference at 55% of AI-optimised infrastructure spend in 2026, rising to 65% by 2029. Inference runs continuously at smaller scale. It does not need hyperscale clusters. It suits the distributed, lower-commitment supply that wholesale GPU networks offer.
How GPU Mesh works as a supply source
GPU Mesh is an alliance of vetted AI Infra capacity partners. On one side are GPU infrastructure owners, neoclouds and data centres with capacity they want to monetise. On the other are operators who need supply without owning hardware.
When you source through GPU Mesh, you are buying reserved or on-demand capacity from vetted providers. You get a quoted rate, a region, the GPU type, and the availability window. You then sell that capacity to your customers at your own pricing, through your own branded portal.
The spread between wholesale buy price and retail sell price is your margin. How much margin depends on what the market will bear in your segment and region, and how efficiently you run the commercial layer. Operators using GPU overcommit through hosted·ai's platform can increase effective capacity sold from a fixed supply base by 2x to 5x, which changes the unit economics materially.
What does it actually cost to start?
Costs split into three buckets.
Software. hosted·ai is licensed as a platform. Pricing is available on request and scales with usage. There is no upfront licensing fee and no per-seat charge. You pay as you grow.
Capacity. GPU Mesh wholesale rates vary by GPU type, region, and term. H100 80GB capacity is available across regions at rates significantly below AWS on-demand ($6.88/hr) and Azure ($12.29/hr). For example, B200 is live on packet.ai through GPU Mesh alliance network. Exact rates depend on the specific supply you source; the matchmaking process typically takes seven days for comparable quotes across providers.
Operations. You need someone who can manage the commercial relationship with GPU Mesh and handle customer support. Many operators launching this way start with one or two people and a reseller agreement with an infrastructure partner like Maerifa, which covers hardware, procurement, and finance for operators who want a single partner for the full stack.
Is this the same as reselling AWS or Azure GPU instances?
No. That distinction matters.
Reselling AWS GPU instances gives you thin margin, a ceiling on differentiation, and no brand equity with customers. You are an affiliate, not a cloud operator. When AWS changes pricing or availability, your business changes with it.
The GPU Mesh model positions you as the operator. You control pricing, packaging, and the customer relationship. Your portal carries your brand. Customers see your company, not hosted·ai, not the underlying capacity provider. You can add your own services on top. You can set your own SLAs. You can compete on something other than being cheaper than AWS, which is a race nobody wins.
What are the real risks?
The model is not without tradeoffs. Being honest about them is more useful than pretending they do not exist.
Supply reliability. You do not own the hardware, which means you depend on third-party availability. GPU Mesh vets its providers, but you should have fallback supply options for critical customer workloads. Do not promise five-nines SLAs on your first week of operation.
Margin compression. Wholesale GPU pricing is competitive and getting more so. Your margin comes from efficient operations, from GPU overcommit on the software side, and from serving customers who value compliance, support, or proximity over raw price. Operators who try to compete purely on price against well-capitalised neoclouds will struggle.
Customer acquisition. The supply side is solved. The demand side is not given to you. You still need a sales motion, a clear ICP, and a reason for customers to pick you over alternatives. The hosted·ai platform and GPU Mesh give you the infrastructure to operate. The commercial story is yours to build.
Who is this model suited to?
It suits operators with existing commercial relationships in a specific vertical or region, who want to add GPU cloud to their product portfolio without a hardware commitment. Managed service providers, regional telcos, enterprise IT resellers, and technology consultancies are the most common entry points.
It also suits entrepreneurs building a focused neocloud for a specific customer segment. If you have a clear ICP and a sales channel, the zero-CAPEX model lets you test the market before committing to owned infrastructure.
It is less suited to operators whose entire value proposition is low cost at high volume. At that end of the market, owned hardware and aggressive overcommit are how you compete. Zero-CAPEX launch can get you there, but it is a starting point, not a long-term strategy for commodity GPU.
The 30-day launch checklist
For operators ready to move, the path looks like this.
Week 1: Commercial and legal setup. Agree terms with hosted·ai. Source initial GPU capacity through GPU Mesh. Define your first product (GPU type, region, pricing model). Register your company if not already done.
Week 2: Platform configuration. Set up your white-label portal. Configure your first GPU pool. Set pricing and billing parameters. Test the provisioning flow end to end.
Week 3: Customer readiness. Onboard your first two or three customers in a controlled pilot. Verify billing accuracy. Confirm support workflows. Fix anything that breaks before you open to inbound.
Week 4: Launch and iterate. Open the portal. Start your sales motion. Monitor utilisation and adjust pricing based on real demand.
Two to four weeks is achievable. It requires decisions, not delays.
Next steps
If you want to run the numbers on your specific situation before committing to anything, the hosted·ai platform overview covers the full commercial model, and the team can walk you through GPU Mesh capacity and pricing for your target region in a 30-minute call.
Book a demo to see how fast you can launch.