◦ 01 · Distributed GPU marketplace

Rent compute.
By the second.
From anyone’s GPU.

HAON PowerHub is a direct line between workers who need GPU time and miners who lend out their idle rigs. Prepaid credit, per-tick capture every 10 seconds, end-to-end encrypted tunnels. You pay only for the compute you actually consumed.

Start renting→Lend your GPU↓

◉ Marketplace · Live

EU-WEST · ALPHA

Miners online

0rigs

Sessions live

0running

Avg rate

$0.00/ hour

Margin

5.1×× electricity

HAON net

+$1.90per day (net)

Kaspa mining

+$0.18per day (net)

Ravencoin mining

+$0.05per day (net)

Bitcoin mining

-$1.40per day (loss)

Numbers shown are indicative marketplace state. Prices and availability update per session.

LIVE ENERGY FEED · EU RESIDENTIAL

DE €0.33•ES €0.14•PT €0.15•UK €0.29•FR €0.20•avg €0.22•

◦ 02 · Why we built this

The GPU shortage
isn’t a 2021 story.
It’s structural. It’s today.

Graphics cards power modern AI — chat, images, science. But the foundries that build them can’t keep up. The queue is years long, and the hyperscalers take the whole shelf. Here’s why, and what a marketplace can actually fix.

◦ Where it breaks

Five steps.
Three of them jammed.

Design

Nvidia / AMD

Two firms design almost every high-end chip on the market.

Bottleneck

Foundry

TSMC · Taiwan

A single foundry produces the leading-edge nodes. Any disruption ripples globally.

Bottleneck

Packaging

Advanced 2.5D / 3D assembly (CoWoS)

The step that fuses logic die and HBM. Capacity is structurally undersized.

Bottleneck

Memory

HBM · high-bandwidth memory

Without HBM3/3E the card can’t ship. The longest waiting line in the industry.

Final GPU

The card you buy

Every step above has to clear before one finished unit reaches a shelf.

◦ 01

Factories are concentrated.

Most advanced chips come from a single island (Taiwan). Any disruption — political, weather, logistics — ripples through the whole planet.

◦ 02

Memory is the choke point.

Even when the chip is ready, without HBM it can’t ship. Today, HBM is the longest waiting line in the industry.

◦ 03

Hyperscalers pre-book the supply.

Google, OpenAI, Meta, Amazon pre-buy entire production runs. The smaller player never sees the inventory — it’s sold before it’s made.

◦ What it costs you today

Three ways to get GPU time.
One of them adds up.

WorkloadRent a GPU serverPay per API callHAON · idle GPU

Run a local LLM

Llama 3 · 8B · one million tokens

~$60 / day

A10 server, idle or not

~$10

GPT-4 output tokens

~$0.15

billed by the second

Generate 100 images

ComfyUI · Stable Diffusion XL

~$5

2 h of A10 compute

$40 – $80

DALL·E / Runway per image

~$0.40 – $0.80

actual minutes only

Two hours to experiment

Fine-tune or tinker, short session

~$20 / day

monthly rental, minimum

not viable

fine-tuning is 10–100× cost

~$0.80

just the 2 hours

* Prices indicative; the marketplace is live-priced and varies by GPU model and region. Billing by the second with a 15-minute minimum.

Run a local LLM

Llama 3 · 8B · one million tokens

Rent a GPU server
A10 server, idle or not
~$60 / day
Pay per API call
GPT-4 output tokens
~$10
HAON · idle GPU
billed by the second
~$0.15

Generate 100 images

ComfyUI · Stable Diffusion XL

Rent a GPU server
2 h of A10 compute
~$5
Pay per API call
DALL·E / Runway per image
$40 – $80
HAON · idle GPU
actual minutes only
~$0.40 – $0.80

Two hours to experiment

Fine-tune or tinker, short session

Rent a GPU server
monthly rental, minimum
~$20 / day
Pay per API call
fine-tuning is 10–100× cost
not viable
HAON · idle GPU
just the 2 hours
~$0.80

* Prices indicative; the marketplace is live-priced and varies by GPU model and region. Billing by the second with a 15-minute minimum.

◦ The answer

There are millions of idle GPUs.
They just aren’t connected.

◦ The idle side

Your GPU sleeps most of the day.

Every gaming rig, creative workstation, and small studio has a GPU that sits at low utilisation for hours on end. HAON turns that idle time into income — you set the rate, you set the hours, you stop whenever you want.

◦ The demand side

People who need compute, can’t get it.

Researchers, artists, small businesses, students — none of them have a pre-booked slot in a data centre. HAON opens that supply to them, billed by the second, no contract, no minimum month.

Lend your GPU→Start renting→

◦ 02 · Protocol

Three moves.
Nothing else.

◦ 01

Miner rents out a GPU

Install the agent, set a price, run it. No open ports — the agent tunnels through a broker, NAT-friendly. Works on Windows, macOS, and Linux.

◦ 02

Worker opens a session

Pick a machine by GPU, runtime, and region. Stream Ollama, ComfyUI, or custom HTTP workloads through an end-to-end encrypted tunnel.

◦ 03

Ticks bill every 10s

Locked balance captures per 10-second tick. You pay only for consumed compute; the unused remainder refunds to your wallet when the session closes.

◦ 03 · Pricing intuition

Roughly
a third
of hyperscaler.

Miners set their own hourly rate. The marketplace clears where demand meets supply, and the platform takes 5% on consumed compute — nothing on the top-up.

GPURuntimeHAON · hrHyperscaler · hr

RTX 4090 · 24GOllama 70B$0.80~$2.40

RTX 3090 · 24GComfyUI SDXL$0.45~$1.60

RTX 4080 · 16GOllama 13B$0.55~$1.90

RTX 3080 · 10GCustom HTTP$0.30~$1.20

* Indicative. Marketplace is live-priced; figures shown are typical alpha clears observed in EU-WEST. Hyperscaler column is on-demand list (AWS / GCP / Lambda, averaged).

◦ 04 · Two sides

Pick yours.

◦ For miners

Idle GPU?
Cash it out.

Install the agent, set your hourly rate, walk away. One command, no port-forwarding, no static IPs.

◦ Windows — one-click installDownload install-miner.cmddouble-click the file · no terminal · no admin · auto-starts at login

+ other platforms / terminal install

macOS · Linux

$ curl -sSL https://haon.run/install-miner.sh | bash

Windows · PowerShell

$ iwr -useb https://haon.run/install-miner.ps1 | iex

◦ For workers

Need a GPU?
Open a session.

Top up prepaid credit with Stripe, browse the marketplace, click open. Your workload is tunnelled to the chosen machine over an encrypted channel; billing fires every 10 seconds.

Windows - one-click installDownload HAON-Setup-Windows.exeWindows 10 / 11 (x64) · double-click

macOSHAON-Setup-Mac.dmg LinuxHAON-Setup-Linux.AppImage

Create an account→

+ power-user · pip install

any platform · Python 3.12+

$ pip install haon-worker && haon-worker login

◦ 05 · Alpha posture

Small, honest,
still a test.

Payments
Stripe test mode only. Use test cards (4242 4242 4242 4242). No real charges will clear.
Data
Tunnels are end-to-end encrypted. HAON does not proxy, store, or inspect your workload payloads — only session metadata for billing.
Scope
Single region (EU-WEST / Hetzner staging), ~5 runtimes, a small miner pool. Expect rough edges. Tell us about them.

alpha@haon.run→Response within one UTC business day.

Rent compute.By the second.From anyone’s GPU.

The GPU shortageisn’t a 2021 story.It’s structural. It’s today.

Five steps.Three of them jammed.

Three ways to get GPU time.One of them adds up.

There are millions of idle GPUs.They just aren’t connected.

Your GPU sleeps most of the day.

People who need compute, can’t get it.

Three moves.Nothing else.

Miner rents out a GPU

Worker opens a session

Ticks bill every 10s

Roughlya thirdof hyperscaler.

Pick yours.

Idle GPU?Cash it out.

Need a GPU?Open a session.

Small, honest,still a test.

Rent compute.
By the second.
From anyone’s GPU.

The GPU shortage
isn’t a 2021 story.
It’s structural. It’s today.

Five steps.
Three of them jammed.

Three ways to get GPU time.
One of them adds up.

There are millions of idle GPUs.
They just aren’t connected.

Three moves.
Nothing else.

Roughly
a third
of hyperscaler.

Idle GPU?
Cash it out.

Need a GPU?
Open a session.

Small, honest,
still a test.