GPU-as-a-Service
Close the gap between
AI promise and AI reality.
Submer Group exists to close the gap between what AI promises and what it actually delivers.
One integrated group owning every layer, from the land and the power underneath to the applications that earn on top.
The only group running GPUs at scale.
Revenue across the full AI stack.
One integrated group, ground to cloud and core to edge,
with assets and revenue across the UK, EU, US, India and the Middle East.
End-to-end AI infrastructure.
From ground to cloud.
Radian Arc
Real-world intelligence. GPU cloud and monetization, core to edge, inside operator networks.
Rubix
Engineering the foundation AI runs on. Land, power and the route to live megawatts.
Submer Technologies · Products
The thermal architecture of AI infrastructure. A decade of owned liquid cooling engineering.
Submer Technologies · Projects
Industrialized infrastructure at scale. Real deployments, not reference designs.
We run on what we sell.
InferX sells the capacity. RadianX streams games on it. Our apps run inference on it. Three engines, one job: your racks never sit idle.
InferX
- A portfolio of hosting capacity
- Delivery of GPUs as GPUaaS (GPU/hr)
- NVIDIA Cloud Partner (NCP) orchestration software
RadianX
- GPU infrastructure for cloud gaming
- Game catalogue partners
- User acquisition (acquiring gamers)
InferX App Ecosystem
- Inferencing application marketplace
- Outcome-focused inferencing across core, edge and frontier
Three problems. One platform.
The AI boom is here and it's running on someone else's cloud. Your megawatts are not earning their full potential.
Not even close.
AI (GPU) capacity
Rapid deployment of an integrated solution with market-aware GPU/hr cost.
Sovereignty
GPUs inside your network make you the seller: in-country, under your control.
Outcomes
Raw power in, recurring revenue out. Every MW maximized.
KPIs- Simplified lifecycle management
- Predictable performance
- Deployment speed
- Compliance
- Zero-CAPEX option and low GPU/hr cost
- Higher margin
- Monetization engine
- Secured revenue streams
You bring the megawatts. We bring everything else.
Move up the stack: idle power, cooling and connectivity become a revenue line, on one contract.
Your infrastructure
- Power, cooling and connectivity, 1MW to 1GW+
- Energized, connected and ready
- Today it earns nothing
The frontier compute
- State-of-the-art GPUs, financed by us
- Deployment, operations and telemetry
- Workloads and a marketplace that fill the racks
The revenue
- A share of revenue on capacity that sat idle
- No hardware bill, no new team to hire
- One contract, one counterparty
GPU, financing, marketplace: the hardware, the money and the demand arrive together. We carry the hardware risk; you keep your network.
From the network edge to a gigawatt.
Same deal, every scale. Tell us what you have and we shape the deal around it.
Edge POPs
- AMD R9600D edge stack, AI-ready
- Latency-critical apps, close to your users
- Sovereign by design: data stays in-country
Entry POPs
- Funded by Radian Arc
- Your team sells; we bring the marketplace
- Start with one rack, grow on demand
1MW to 1GW
- Financed and led by Radian Arc
- Co-investment available
- Joint sales to enterprise and government
- Accelerator campuses for global platforms
Federated core to edge.
Applications create demand. The control plane routes it. A federated network of operators hosts it. One stack, built to keep every hosted GPU earning.
The right workload · The right model · The right location · Every GPU earning.
Maximum intelligence per Watt.
Every request lands on the silicon where it performs best.
routing
The plane reads each request's intent, then weighs cost, latency and data residency in real time, on every workload, every request.
Frontier brains. Edge prices.
For the right workloads, a tiny model at the edge matches frontier-model quality.
Then it wins on price, every time. Six workloads, one equation.
Broadband’s #1 value prop. In your bundle.
Why cloud gaming leads every operator conversation.
A frame-perfect world.
At a self-hosted cost.
Game studios render worlds, bake assets and run simulation on our GPUs.
A fraction of cloud-render cost at the same output quality.
Batch jobs fill the hours gamers sleep, so the racks earn around the clock.
Studio-quality curriculum video.
At tiny-model prices.
Proto Education AI generates curriculum video on models hosted on our GPUs. Self-hosting the stack is a fraction of the price of frontier APIs, minute for minute.
What enterprise customers need.
At a fraction of their token bill.
One harness, many models: tasks route to the cheapest model that clears the bar. Frontier only when the job demands it.
Worked example: 10 developers · 5 AI agents · 4.5B tokens per month
Frontier: published list rates, July 2026, blended 70/30 · Self-hosted: Radian Arc economics.
Every incident resolved.
Cheaper per ticket.
Resolv, by Lumin8: a brand's FAQs become support agents on LINE, web and API. On edge GPUs, every incident costs a fraction of a frontier-model ticket. A product your enterprise team can sell as your own.
More cameras per GPU.
At a lower price.
A range of CCTV applications we are already engaged with, running tiny vision models on edge GPUs. The video never leaves your network.
See the storm.
Before the radar does.
The Meteorological Department of Curaçao guards Caribbean airspace where tropical storms form outside radar range. ALICIA, their AI model, learned to see storms first: trained in-network on our GPU cluster, delivered without a single dollar of hardware.
Apps plug in. Revenue flows out.
Every app in the marketplace lands inside your ecosystem the same way: one integration, one bill, one sales motion, and a share on every sale.
One marketplace
We qualify, package and manage the app vendors. You sign one deal, not fifty.
Single billing
Apps land on the bill your customers already pay. Revenue share settles behind the scenes.
Your sales motion
Playbooks, collateral and joint calls: your enterprise team sells, we support every deal.
Shared upside
You host and sell, we run and settle: a share of every app sold on your racks.
Commercial monetization engine, in motion.
Live deployments across telcos, data centers and GPU clusters: the same engine, already earning in market.
Cloud gaming at national scale
- 250+ GPU servers deployed
- Adding 60k subscribers daily
- Expanding to inference for agriculture
National AI infrastructure
- Started with cloud gaming, grew to national AI
- COE for TM; building COE for the Malaysian government
- Current pipeline of 40+MW in committed AI demand
Cloud gaming and edge AI
- 50 GPU servers deployed
- Adding edge inference and Cloud PC
NVIDIA off-take data center
- Deploying 2MW capacity in India
- Sourcing and deploying applications for CCTV and agriculture
GPU clusters across Europe
- Built and shipped with Boston
- Romania 1,024 · Turkey 2,400
- UK 6,000 (super cluster) · Germany 300 NVIDIA GPUs
GPUs to fill your megawatts, secured.
Boston
Supermicro
NVIDIA
Radian Arc deploys the intelligence that monetizes all of this at the edge.
InferX. Built on standards.
Maximizing MW revenue.
Liquid cooled.
Every watt spent on cooling is a watt not spent on compute. Liquid cooling puts more GPUs on the same megawatt, and the density lands on your revenue per MW.
Sell every layer of the stack.
Bare metal, Kubernetes, GPU instances or tokens: four ways your market pays for the same racks.
Architecture
customer K8s
Instances
Tokens
Physical foundation
Flexible deployment models for edge, regional and distributed workloads.
Fleet operations
Lifecycle management, telemetry, deploy, provision, power, thermals and metering.
K8s
Managed Kubernetes and KubeVirt, or customer-managed via InferX FleetOps APIs.
Consumption layer
Tenant-isolated GPU instances, containers, VPCs, quota and usage controls.
SaaS applications
Turnkey inferencing, training and model serving.
Onboarding, megawatts
to money.
- Complete a site survey
- Register your MW and locations
- Close the capacity and app reseller agreementRevenue share · your capital stays put