GPU Mesh explained: how the wholesale GPU capacity network works
What is GPU Mesh and how does wholesale GPU capacity sharing work?
GPU Mesh is hosted·ai's wholesale GPU capacity sharing network for service providers. It connects two types of participants: GPU infrastructure owners who want to monetise idle or underutilised capacity, and neoclouds or operators who need GPU compute without the CAPEX of owning hardware. Suppliers list their available compute on GPU Mesh with a region, GPU type, availability window, and wholesale price. Buyers subscribe to those pools at wholesale rates and sell access to their end customers at retail pricing through their own branded portal, powered by hosted·ai's GPUaaS software. GPU Mesh handles the matchmaking, supply verification, and the wholesale commercial relationship. Critically: publishing pools to GPU Mesh does not make them exclusively available through the Mesh. Suppliers can sell those same pools to their own direct customers at the same time, running both channels simultaneously through the hosted·ai admin panel.
The two sides of GPU Mesh
GPU Mesh is a two-sided market. Understanding both sides is necessary to understand how the network creates value.
Suppliers: GPU infrastructure owners. Suppliers are companies with GPU infrastructure. Using hosted·ai, they create virtual GPU pools with optimised utilisation and overcommit, then publish those pools to the Mesh. Other service providers subscribe to those pools and re-sell them to customers. Suppliers are paid for GPU resources that are consumed. They do not need to build a retail GPU cloud product or acquire end customers directly. GPU Mesh provides the demand channel. Publishing a pool to the Mesh does not remove it from direct sales — suppliers run both simultaneously.
Buyers: operators who need supply. Buyers are service providers that need GPU infrastructure. Using hosted·ai, they subscribe to GPU pools on GPU Mesh and re-sell that infrastructure as GPUaaS to their end customers. Buyers are billed for GPU resources their customers consume — not for subscribing to a pool. They can sell GPU Mesh resources alongside any on-premises GPU infrastructure they already have. The source of GPU Mesh infrastructure is invisible to the buyer's end customer.
GPU Mesh infrastructure tiers: Premium, Standard, and Agile
GPU Mesh infrastructure is provided in three tiers. Each tier has a different sharing (overcommit) ratio and workload scheduling configuration, suited to different buyer use cases and supplier monetisation targets.
Premium tier: 1x sharing. No GPU sharing. Highest performance. Best for workloads that need minimum latency and highest security. Uses temporal scheduling — workloads have full isolated access to GPU resources while they have priority.
Standard tier: 2x sharing. High performance with a balance of security and cost efficiency. Also uses temporal scheduling, which offers the highest security for shared environments — workloads are swapped in and out of GPUs in the pool and have full access while they have priority.
Agile tier: 4x sharing. Highest efficiency. Best for workloads that need optimal pricing. Uses spatial scheduling, which allows workloads to co-exist on GPUs simultaneously. This maximises utilisation efficiency at the cost of lower isolation.
The sharing ratio controls how many virtual GPUs are available per physical GPU. With 8 GPUs in a pool on the Agile tier (4x sharing), 32 virtual GPUs are available for customers to subscribe to simultaneously. For suppliers, higher sharing ratios mean more revenue per physical GPU. For buyers, higher sharing ratios mean lower wholesale costs and more competitive retail pricing.
GPU Mesh pricing: how wholesale rates work
Each tier has a wholesale price per hour, per GPU type, set by hosted·ai to reflect typical retail GPUaaS market rates. Wholesale Mesh prices are typically between 50% and 80% of retail.
Buyers set their own retail price for their GPUaaS offering and pay the wholesale price only when their customers consume from the GPU pool. Subscribing to a pool location is free. Buyers must have credit in their GPU Mesh wallet to use Mesh resources; auto top-up options are available.
Suppliers are paid at the wholesale price when their GPU resources are consumed. Payment is credited to the supplier's GPU Mesh wallet, with weekly payment as the most frequent option and multiple options for transferring credits to a destination of choice.
How capacity is matched and verified
GPU Mesh is not a self-serve marketplace where any listed capacity is immediately purchasable. Infrastructure owners go through a verification process before their capacity is listed. This covers hardware specifications (GPU type, VRAM, node count), connectivity (InfiniBand, network uplink speed and redundancy), and operational baseline (uptime history, support model).
Once verified, capacity is listed with the specifics that buyers need to make a sourcing decision: exact GPU type, VRAM per GPU, node count available, InfiniBand yes or no, internet uplink, location (region, not necessarily exact data centre), pricing model (reserved or on-demand), minimum term, and go-live timeline.
What buyers can do with sourced capacity
Capacity sourced through GPU Mesh is managed through the hosted·ai platform in the same way as owned hardware. Buyers configure GPU pools, set overcommit ratios, define products, set pricing, and open capacity to customers through their portal. From the customer's perspective, the experience is identical whether the underlying capacity is owned hardware or GPU Mesh supply.
GPU overcommit applies to GPU Mesh capacity the same way it applies to owned hardware. A buyer who sources 10 H100 nodes through GPU Mesh and applies 5x overcommit is selling the effective capacity of 50 H100 access slots to their customers. This is why overcommit is critical to the economics: at a 1:1 ratio, the spread between wholesale cost and retail price is tight. With 3x to 5x overcommit, the margin is substantial.
GPU Mesh and packet.ai
GPU Mesh capacity feeds into packet.ai, hosted·ai's elastic neocloud. packet.ai is the retail distribution layer for the GPU Mesh network: it sources capacity from GPU Mesh participants and makes it available to developers and enterprise AI teams at competitive per-minute pricing. B200 capacity is live on packet.ai at $3.75/hr, sourced through the GPU Mesh network from hardware partners operating Blackwell infrastructure.
For GPU Mesh capacity owners, this means their hardware can be utilised by packet.ai demand even when they do not have direct retail customers of their own. For packet.ai customers, it means access to a diversified supply base across multiple infrastructure providers and regions.
Regions covered by GPU Mesh
GPU Mesh currently covers capacity in North America, Europe, the Middle East, Africa, and Asia-Pacific. H100 80GB is the most widely available GPU type. B200 and H200 capacity is available in North America and select EU regions. A100 and older generation supply is available across most regions. Buyers with regional requirements, for example an EU CSP who needs capacity within EU legal jurisdiction, can specify regional constraints and GPU Mesh will match them only with verified capacity within that geography.
Getting access to GPU Mesh
GPU Mesh is available to hosted·ai platform customers. Operators using the platform can access GPU Mesh sourcing as part of their subscription. Capacity owners who want to list supply on GPU Mesh go through the verification process independently of the platform subscription.
For a detailed walkthrough of how to become a supplier or buyer on GPU Mesh, download the hosted·ai GPU Mesh Guide.