Start Building →
Live demand · refreshed continuously

GPU infrastructure demand board.

Active infrastructure requests from packet.ai customers. If you can supply the GPU capacity we're looking for, express interest below and our team will be in touch within one business day.

Get notified of new demandApply as provider
Open requests14
Total GPUs sought5,472+
SKUsB300 · B200 · H200 · H100
GPU
Region
Sort
Showing 14 of 14

128 H200 nodes

Worldwide
Medium urgency

For a Middle East partner — looking for 1,000+ H200 GPUs. Short-term requirement, flexible on minor spec deltas.

Hardware
1024+ H200
Region
Worldwide
Minimum
Min: 12 months
GPU type
H200
GPUs/node
8
Nodes
128
Total GPUs
1024+
Win price
$1.80 / hr / GPU
12 months preferred
Express interest

128 B200 Nodes

Europe
Medium urgency

128 B200 nodes, 1-year commit, Europe. Frankfurt / Amsterdam / Paris all acceptable.

Hardware
1024+ B200
Region
Europe
Minimum
Min: 12 months
GPU type
B200
GPUs/node
8
Nodes
128
Total GPUs
1024+
Win price
$2.30 / hr / GPU
12 months preferred
Express interest

128 H100 nodes

Worldwide
Medium urgency

128 nodes H100, 12 months, worldwide. Standard HGX networking and storage acceptable.

Hardware
1024+ H100
Region
Worldwide
Minimum
Min: 12 months
GPU type
H100
GPUs/node
8
Nodes
128
Total GPUs
1024+
Win price
$1.60 / hr / GPU
12 months preferred
Express interest

100 B200 / H200 nodes

Worldwide
Medium urgency

100 nodes B200 or H200, 12 months, worldwide. Dedicated InfiniBand mandatory for multi-node training synchronization. Pricing target: B200 $3.00/GPU/hr, H200 $1.70/GPU/hr.

Hardware
800+ B200
Region
Worldwide
Minimum
Min: 12 months
GPU type
B200
GPUs/node
8
Nodes
100
Total GPUs
800+
Win price
$3.00 / hr / GPU
12 months preferred
Express interest

64 B200 nodes

California
Medium urgency

64 nodes B200, California preferred, 12 months. Must include high-speed interconnect, dedicated bandwidth, and 24/7 technical support.

Hardware
512+ B200
Region
California
Minimum
Min: 12 months
GPU type
B200
GPUs/node
8
Nodes
64
Total GPUs
512+
Win price
$2.80 / hr / GPU
12 months preferred
Express interest

64–128 B200 nodes

US · UK
Medium urgency

64–128 B200 nodes, US Tier 3 DC or UK, 12 or 24 months. Workload: LLM training & fine-tuning. NVIDIA Quantum-2 InfiniBand 400 Gb/s or RoCEv2 mandatory. 5th-gen NVLink/NVSwitch (1.8 TB/s). 1.5 TB HBM3e aggregate per node. High-performance local NVMe.

Hardware
512+ B200
Region
US · UK
Minimum
Min: 12 months
GPU type
B200
GPUs/node
8
Nodes
64–128
Total GPUs
512+
Win price
$2.80 / hr / GPU
24 months preferred
Express interest

20 H200 nodes

Flexible
Medium urgency

20 H200 nodes, 1-year commit. Flexible on region and exact spec.

Hardware
160+ H200
Region
Flexible
Minimum
Min: 12 months
GPU type
H200
GPUs/node
8
Nodes
20
Total GPUs
160+
Win price
$1.40 / hr / GPU
12 months preferred
Express interest

16 B300 nodes

United States
Medium urgency

16 B300 nodes, 12 months, Texas/USA (Tier 3 or equivalent). NVIDIA InfiniBand (6.4 Tbps) and NVIDIA Spectrum (2×400Gbps north/south). 100 Gbps dedicated internet per node. 2 PB hot storage included.

Hardware
128+ B300
Region
United States
Minimum
Min: 12 months
GPU type
B300
GPUs/node
8
Nodes
16
Total GPUs
128+
Win price
$4.40 / hr / GPU
12 months preferred
Express interest

16 B200 nodes

United States
Medium urgency

16 nodes B200, US, 6–12 months, US Tier 3 DC. Standard interconnect and storage.

Hardware
128+ B200
Region
United States
Minimum
Min: 6 months
GPU type
B200
GPUs/node
8
Nodes
16
Total GPUs
128+
Win price
$3.90 / hr / GPU
12 months preferred
Express interest

10 H100 nodes

Flexible
Medium urgency

10 H100 nodes, 1-year commit. Open on region — competitive H100 rates wanted.

Hardware
80+ H100
Region
Flexible
Minimum
Min: 12 months
GPU type
H100
GPUs/node
8
Nodes
10
Total GPUs
80+
Win price
$1.25 / hr / GPU
12 months preferred
Express interest

4 B200 Nodes

Flexible
Medium urgency

4 B200 nodes, 1-year commit. Production inference and fine-tuning workload.

Hardware
32+ B200
Region
Flexible
Minimum
Min: 12 months
GPU type
B200
GPUs/node
8
Nodes
4
Total GPUs
32+
Win price
$2.30 / hr / GPU
12 months preferred
Express interest

4 H200 nodes

European Union
Medium urgency

4 nodes H200, 6 months, EU. Standard HGX network configuration acceptable.

Hardware
32+ H200
Region
European Union
Minimum
Min: 6 months
GPU type
H200
GPUs/node
8
Nodes
4
Total GPUs
32+
Win price
$1.85 / hr / GPU
6 months preferred
Express interest

1 H200 node

United States
Medium urgency

1 node H200, US, 12 months. Must support Intel TDX (BIOS enabled) and HGX Hopper Firmware 1.7.0+. SEV/SNP-compatible. Internal NVLink/NVSwitch GPU-to-GPU; RoCEv2 acceptable. Dedicated public IPv4 (no CGNAT), full OS-level firewall control. Ubuntu 25.10.

Hardware
8+ H200
Region
United States
Minimum
Min: 12 months
GPU type
H200
GPUs/node
8
Nodes
1
Total GPUs
8+
Win price
$2.00 / hr / GPU
12 months preferred
Express interest

1–2 B200 nodes

United States
Medium urgency

1–2 nodes B200, US preferred, 6 months (open to longer). High-speed NVMe local disks required for minimized model startup and checkpoint loading. Workload: optimized inference. Standard HGX networking is sufficient. 2 TB+ system RAM.

Hardware
8+ B200
Region
United States
Minimum
Min: 6 months
GPU type
B200
GPUs/node
8
Nodes
1–2
Total GPUs
8+
Win price
$3.00 / hr / GPU
6 months preferred
Express interest

Become a GPU provider

Have GPU infrastructure you want to monetize? Join our provider network and get access to enterprise customers with verified KYC and predictable payouts.

Apply as provider →
Provider FAQ

Supplying capacity.

Everything providers want to know before expressing interest on a demand entry.

How is the 'win price' set?
It's the rate the buyer has indicated they're ready to transact at — informed by their target, current market rates, and the term they're committing to. Suppliers can come in below the win price for higher selection probability, or at the win price with stronger terms (region, networking, SLA).
What does 'Express Interest' actually do?
Submits your capacity proposal to our sourcing team. We verify your hardware, region, and timing match the request, then route qualified offers to the buyer within 24–48 hours. Buyers shortlist 1–3 providers per deal. No fee to express interest; commissions only apply on closed deals.
How do I get notified of new demand?
Hit “Get notified of new demand” above and tell us which SKUs and regions you supply. We email a daily digest of matching requests, or a real-time webhook for high-volume suppliers. Filter by GPU type, region, total volume, and commitment length.
What information do buyers see about my offer?
Your offer’s GPU SKU, region/datacenter tier, networking spec, total available capacity, $/hr rate, commitment terms, and provider rating. Your identity stays anonymous until the buyer selects you for negotiation, so you can offer aggressive pricing without affecting your public rate card.
How fast do deals typically close?
Small fleets (under 16 nodes) close in 3–10 days from express-interest to PO signed. Multi-node InfiniBand cluster deals (64+ nodes) typically close in 2–6 weeks, with technical validation calls in the middle. The urgency tag on each card reflects buyer-stated timeline.
What if I can fulfil only part of the request?
Partial fulfilment is welcome and common, especially on large multi-region requests. Buyers regularly accept 2–3 suppliers per deal to hit their total capacity target. Submit your portion and we'll match it against complementary offers.
Do I need to be an existing packet.ai provider to express interest?
No — anyone with eligible datacenter-grade NVIDIA GPUs can submit. New suppliers go through our standard provider verification (hardware specs, network quality, hosting environment) before contract — typically under one week.
Are these the only open requests?
No. The board shows public-facing demand we're comfortable sharing openly. We also have NDA-bound enterprise requests that don't appear here — if you're a vetted provider, we route those privately based on your supply profile. Email providers@packet.ai to opt in.