hosted·ai for Telecoms AI Grid

Making AI Grid
earn for Telcos.

NVIDIA’s AI Grid vision is clear: inference should run where latency, cost and data rules fit best, and telcos already own the sites and connectivity to put it there.

Just two things missing: a way to remove the GPU cap from your smaller sites, and the business layer needed to commercialize your AI Grid. With hosted·ai you get both.

Your AI Grid without hosted·ai: most GPU capacity sits idle and each site serves few customersHosted·ai AI Grid: GPU pooling at each site, connected into one grid with a token exchange
  • Scale
  • Monetization
  • Token exchange
Unlock each site with GPU pooling

Remove the GPU cap. Serve 2–10x more models per GPU.

hosted·ai’s GPU Hypervisor brings full GPU virtualization to your NVIDIA sites. It pools GPUs, gives every tenant secure, isolated vGPUs, and multiplies the workloads, revenue and ROI each GPU supports.

Today: one client per GPU

Illustrative four-GPU edge site

GPU 1
70B
GPU 2 · spikeRequests queue
8B
GPU 3 · idle
7B
GPU 4 · idle
BGE

In most GPU clouds, each model is pinned to its own GPU. The number of paying tenants at a site is capped by its GPU count, which is capped by power, especially at cable hubs and RAN sites. Models that outgrow their GPU have nowhere to go, and capacity a model doesn’t use is stranded.

  • 4 GPUs, 4 tenants. A fifth needs a fifth GPU.
  • A spike on one model can’t use idle GPUs.
  • Small models strand most of a card.

hosted·ai: many clients per GPU

Same four GPUs, pooled

GPU 1GPU 2GPU 3GPU 4
Llama 70B
DeepSeek 32B
Vision
Voice agent
Mistral 7B
Llama 8B
Qwen 8B
Gemma
BGE
Whisper
Reranker
+ long tail
Compute share, hour by hour Every model keeps a floor
00:0006:0012:0018:0024:00
Voice agentLarge modelsVisionSmall modelsFree for spikes and token exchange

With hosted·ai, every GPU joins a pool. Tenants run on vGPUs drawn from the pool, not on a particular card.

Every model has a guaranteed floor, and compute follows demand: busy models take the cycles idle models aren’t using and hand them back as traffic falls. Memory is partitioned per tenant with hard limits, so sharing stays secure.

  • 4 GPUs, 12 models and more. Add a tenant, not a GPU.
  • A busy model takes idle compute, then hands it back.
  • Capacity left over is sold on the token exchange.

In most GPU clouds, each model is pinned to its own GPU. The number of paying tenants at a site is capped by its GPU count, which is capped by power, especially at cable hubs and RAN sites. Models that outgrow their GPU have nowhere to go, and capacity a model doesn’t use is stranded.

With hosted·ai, every GPU joins a pool. Tenants run on vGPUs drawn from the pool, not on a particular card. Every model has a guaranteed floor, and compute follows demand: busy models take the cycles idle models aren’t using and hand them back as traffic falls. Memory is partitioned per tenant with hard limits, so sharing stays secure.

Fits the reference design

hosted·ai runs on the RTX PRO 6000 and GB300 sites NVIDIA’s AI Grid specifies. No new hardware.

Works with your serving stack

Dynamo, NIM, TensorRT-LLM and partner placement engines keep their jobs. We add utilization, products, billing and demand.

No application changes

Unmodified CUDA. Tenants see a GPU or an API, never the hardware.

Maximize output per GPU

MIG vs time-slicing vs MPS vs the hosted·ai GPU Hypervisor

Multi-Instance GPU, time-slicing and Multi-Process Service each give up flexibility, isolation or fairness. Multi-tenant inference at the edge needs all three. hosted·ai delivers by pooling GPUs and dynamically scheduling multi-tenant workloads to maximize output per GPU.

ApproachModels per GPUIsolationChanging sharesBusy model uses idle compute
Dedicated GPUOneFullBuy a GPUNo
MIGFixed slicesHardwareDestroy and recreateNo
Time-slicingSeveralNone for memoryRound robinYes, unprotected
MPSSeveralAddress space onlyFirst comeYes, can starve others
hosted·ai GPU HypervisorUp to 10 vGPUs per cardMemory isolated per tenantContinuously, by demandYes, with a guaranteed floor
Monetize your AI Grid

Add the business layer. Products, demand, getting paid.

hosted·ai turns your NVIDIA sites into GPU-as-a-Service products that sell, bill and pay out, under your brand. It also connects your AI Grid to a global demand network, bringing you buyers for tokens and rentals from day one. NVIDIA’s reference design doesn’t change.

Products

Sell tokens, virtual GPUs, on-demand, short-term and long-term capacity from the same fleet, dedicated or shared. Change the mix as demand moves, site by site.

Demand

The token exchange and hosted·ai marketplaces bring buyers at launch, served on your capacity under your brand. Your own customers always come first.

Getting paid

Every token and GPU hour is metered per tenant, site and GPU. hosted·ai collects from exchange buyers and pays you out automatically.

Scaling AI Grid inference revenue

The token exchange: inference throughput and scale

Enterprise demand for edge inference is the growth vector for AI Grid. Inference platforms need capacity now, at peaks and in regions where they have none. hosted·ai enables the token exchange, an inference marketplace between operators, so you serve your clients, fill idle capacity, and have access to more when your sites are full.

Supply

Your gridYour own demand runs first. Idle GPUs earn from everyone else.
Partner gridsWhen you’re full, your customer is served on a partner. You keep the customer.

The exchange

hosted·ai token exchange
  1. Apply your policy
  2. Place on capacity that fits
  3. Meter every token
  4. Pay out whoever served it

Demand

Your customersOn your endpoint, under your brand. Served first.
Inference platformsPeaks, new regions, low latency. Pay per token.
Committed platformsToken offtake, committed up front.

Operators invented roaming. The home network keeps the customer; the serving network gets paid. The token exchange does the same for inference.

Sovereign AI, by design

You keep the customer, the brand and the control

hosted·ai is fully self-hosted. The controller and every GPU node run in your environment, and you set the policy enforced on every request.

Residency by countryLatency targets per modelPricing per GPU typePartners you buy from and sell toOverflow: when and where
Isn’t AI Grid just MEC all over again?

Earlier multi-access edge computing (MEC) programs put compute in the network with no products, no buyers and no billing. hosted·ai adds all three: many tenants per GPU, multiple ways to sell each card, demand from the token exchange, and metering and payouts from day one.

Can it meet the highest sovereignty tier?

It’s designed for it. Residency policy decides which data, models and logs may leave which country, and tenant memory is partitioned and zeroed between tenants in Security mode. Formal certification is assessed per deployment, with your auditors.

How do we finance the build?

Pre-sell part of each site on long-term contracts or token offtake, then sell the rest by the hour and by the token. The contracted part supports financing; the rest lifts revenue per GPU.

Could we build this from open source?

Parts of it. Projects like HAMi can share a GPU between containers. They don’t give you one supported product across every site: VMs and containers, pricing, metering, payouts, the exchange and buyers.

Who controls the grid?

You do. You set products, prices, placement, partners and residency. hosted·ai supplies the platform and the demand, not a competing cloud.

Make your AI Grid earn.

Book your AI Grid review: a 30-minute call on your sites, your workloads and the products your grid could sell, covering commercials, technology or both.

Thanks. We have your request and will be in touch to arrange your AI Grid review.
Oops! Something went wrong while submitting the form.