Start Building →
For GPU providers

Turn idle GPUs into recurring revenue.

Add your spare GPU capacity to the packet.ai marketplace. We bring the customers, billing, support, and provisioning agent. You bring the hardware — and keep it under your control at all times.

Apply to become a provider →
Sample provider · last 30 days
$12,450
+18% MoM
Active GPUs24
Average utilization87%
Next payout · Feb 28$6,225.00
87%
Average fleet utilization across packet.ai providers
$2.84
Average $/GPU/hr paid to providers, blended across SKUs
14 days
Payout cycle — bi-weekly, direct to your bank account
99.5%
Uptime SLA from providers; we screen for stability
How it works

We handle the hard parts.

Focus on keeping your hardware running. We bring AI developers, run the billing engine, answer 24/7 support tickets, and take care of provisioning. You operate the metal.

Zero sales overhead
No sales team needed. Your GPUs are instantly discoverable by thousands of AI developers worldwide on packet.ai's marketplace.
Bi-weekly payouts
Predictable income every two weeks. We handle billing, invoicing, payment collection, and currency conversion. Direct deposit to your bank account.
Real-time transparency
Full visibility into occupancy, revenue per hour, GPU-by-GPU earnings, and historical trends — straight from the provider dashboard.
Scale flexibly
Add 1 to thousands of GPUs. Scale up or down based on your capacity. No minimums, no long-term commitments, no penalties for withdrawing.
Premium customers
Verified businesses running production AI workloads. No crypto miners, no abuse, no nuisance traffic — protected by AUP enforcement and KYC.
Simple integration
A lightweight provisioning agent handles VM lifecycle and monitoring on your hosts. Minimal overhead. You keep root and full control of the metal.
Apply to become a provider →
The process

Three steps. About a week.

From application to your first dollar of marketplace revenue, typically in under seven days.

Apply

Submit your infrastructure details. We verify GPU model and BIOS revision, network quality (latency, jitter, peering), power redundancy, and hosting environment.

Connect

Install the lightweight hosted.ai provisioning agent. Hosts stay under your control — we orchestrate VMs through the agent; you keep root and physical access.

Earn

Your GPUs go live on the marketplace. Track occupancy, $/hr, and earnings in real time through the dashboard. First payout lands on the next bi-weekly cycle.

Apply to become a provider →
Eligible hardware

We accept high-performance NVIDIA GPUs.

Datacenter-grade silicon for AI/ML workloads. Tier determines the rate band. Blackwell gets premium pricing; Ampere still earns solid returns.

Blackwellpremium
NVIDIA B200
180 GB HBM3e
Highest tier pricing
Hopperpremium
NVIDIA H200
141 GB HBM3
Top tier pricing
Hopperhigh
NVIDIA H100
80 GB HBM3
Competitive pricing
Amperestandard
A100 80G
80 GB HBM2e
Solid returns pricing
Blackwellstandard
RTX 6000 Pro
96 GB GDDR7
Cost-effective pricing
Ada Lovelacestandard
L40S
48 GB GDDR6
Solid returns pricing
Blackwellstandard
RTX 5090
32 GB GDDR7
Cost-effective pricing
Ada Lovelacestandard
RTX 4090
24 GB GDDR6X
Entry tier pricing
Stable network connectivity · 1 Gbps+
Enterprise-grade hosting / datacenter
Reliable power & cooling, N+1 ideal
99%+ uptime capability
Apply to become a provider →
Provider dashboard

Real-time visibility, down to the GPU.

Live utilization, $/hr, occupancy state, and earnings for every node. Export CSVs for accounting. API access for custom monitoring.

provider.packet.ai · fleet overviewLIVE
Monthly revenue
$12,450
▲ +18%
Avg utilization
87%
▲ +4%
Active GPUs
24
— 0
Uptime
99.97%
▲ +0.02%
GPUUtilizationRevenueStatus
H100-01
94%
$3.20/hr
occupied
H100-02
88%
$3.20/hr
occupied
A100-01
76%
$2.10/hr
occupied
A100-02
0%
$0.00/hr
available
B200-01
92%
$5.40/hr
occupied
Next payout · February 28$6,225.00
Apply to become a provider →
Your GPUs, your terms

Complete flexibility. No lock-in.

Withdraw anytime

Need your GPUs back? Withdraw capacity at any time. If currently occupied, we require 14 days of notice to migrate the customer workload smoothly. No financial penalty.

No contracts

No long-term contracts or minimum commitments. Add capacity when you have it, withdraw when you need it. Month-to-month, plain language, no enterprise legal department required.

Apply to become a provider →
Provider FAQ

Questions before applying.

If you don't see your question, email providers@packet.ai. We typically reply within one business day.

What's the revenue split?
Provider rate cards are tier-based by GPU SKU. As a general rule, providers earn the majority of customer-billed $/hr after platform fees. Detailed rate cards are shared during onboarding based on your hardware and region.
What hardware do I need to qualify?
Datacenter-grade NVIDIA GPUs (B200, H200, H100, A100 80G, RTX 6000 Pro, RTX 5090, RTX 4090, L40S). Hosts must run in a real datacenter or comparable hosting environment with 1 Gbps+ networking, redundant power, and stable cooling. We don't accept home-lab or consumer-grade setups.
How does the provisioning agent work?
The hosted.ai agent runs on your hosts and exposes a minimal, audited API for VM lifecycle management. It does not transmit any customer workload data back to us — workloads run isolated on your hardware. Source-level documentation is provided to vetted providers.
Do you take physical or network control of my hardware?
No. You keep full root, BMC, and physical access to your systems. The agent orchestrates VMs over a defined API. You can pause it, audit it, or pull the plug at any time.
Who handles customer support?
We do. packet.ai's 24/7 support team handles all tier-1 and tier-2 customer issues. We escalate to providers only for hardware-level incidents (e.g. node down, fan failure) via your on-call channel of choice.
How are payouts processed?
Bi-weekly, in USD, via direct deposit (ACH or SWIFT). Payouts include a CSV breakdown of revenue per node per hour. We support providers in the US, EU, UK, and most major OECD countries.
What regions are you most interested in?
Active demand across US (east, central, west), EU (Frankfurt, Amsterdam, Dublin, Paris), and UK (London). We also onboard providers in Asia-Pacific and Latin America selectively based on customer demand and latency profile.
Are workloads isolated between customers?
Yes. Each customer workload runs in an isolated VM (KVM) or container with dedicated GPU passthrough. No multi-tenant sharing of a single GPU between customers unless they explicitly opt into the Dynamic tier (and even then, isolation is preserved at the scheduler level).
Apply to become a provider →

Ready to monetize your GPU capacity?

Join our network of verified providers and start earning from idle GPUs. Applications reviewed within 2 business days.

Apply to become a provider →Contact us