packet.ai

Mixed fleetGPU specialist

7 GPU models, a mix of consumer cards and datacenter accelerators. Sold on-demand. Single cards only. 4 of 7 SKUs are quote-only — the price is not published at all.

single GPUquoted in USD4 quote-onlyregistered in United States

Read the review Source page ↗

Cheapest offer

$0.39

Offers tracked

GPU models

Egress

not tracked

Storage

not tracked

Last change

7h ago

Catalog

GPUCardsBillingCost /hr

NVIDIA RTX 4090 24GBRTX 4090 24GB× 1on demand$0.39 Deploy

NVIDIA L40S 48GBL40S 48GB× 1on demand$0.92 Deploy

NVIDIA A100 80GBA100 80GB× 1on demand$1.43 Deploy

NVIDIA H100 SXM5 80GBH100 SXM 80GB× 1not publishednot published Deploy

NVIDIA H200 SXM5 141GBH200 141GB× 1not publishednot published Deploy

NVIDIA RTX 5090 32GBRTX 5090 32GB× 1not publishednot published Deploy

NVIDIA RTX PRO 6000 Blackwell 96GBRTX 6000 Pro 96GB× 1not publishednot published Deploy

3 of 7 offers carry a published price. The rest are “contact us” SKUs, shown rather than hidden.

How this provider sells

Hardware owner

The provider, in owned or leased capacity built for GPU rental.

Price set by

The provider, one price per SKU.

You are buying from

An operator whose whole business is renting these cards.

Every card here has a single price once region and terms are fixed.

Registered company

You contract with

Hosted AI Inc

United States · Delaware

San Jose, California (headquarters)

2 sources

“Packet.ai is a product of Hosted AI Inc, a Delaware corporation”
packet.ai/terms0f253c16947e

“a Delaware corporation headquartered in San Jose, California.”
packet.ai/contact5c26a4b62250

Quoted from packet.ai’s own pages, read 2026-09-12. A missing line means the site does not say, not that the answer is no.

Cheapest rate over time

Not enough history yet. Price history comes from closing an offer row when its price changes, and the collector does not do that yet — so there is one observation per offer and nothing to chart. This fills in once that lands.