Live · Decentralized GPU Infrastructure

Powering AI Through Decentralized GPU Infrastructure

Real AI workloads. Real GPU utilization. Transparent compute output powered by a globally distributed GPU network.

Network · Live
Updated every second
GPUs Online
4,128
Network Utilization
AI Inference Jobs Today
Tokens Processed This Week
Uptime 99.6%

Live Job Stream

Anonymized tasks executing right now
[L40-23]
GPU Usage
Flux Pro · Video Generation
95%
[A100-11]
Tokens / sec
Llama 3 8B · Inference
134
[RTX 4090-07]
Duration
LoRA Fine-tune
01:22:33
[H200-04]
Tokens / sec
Mixtral 8x22B · Inference
86
[RTX 5090-15]
GPU Usage
SDXL · Batch Image
90%
AI Models Powered by the Network

Who do we supply tokens to?

As a leading global token producer, our tokens are supplied to the world's mainstream AI model providers, covering key markets across North America, Europe, and the Middle East. The companies below are a sample of our strategic partners in the AI industry — a snapshot of our international footprint across the sector.

OpenAI logo

OpenAI

Origin: United States
Flagship models

GPT-5 · GPT-4.1 · o4-mini · gpt-image-2

Anthropic logo

Anthropic

Origin: United States
Flagship models

Claude Opus 4.6 · Claude Sonnet 4.5 · Claude Haiku

Google DeepMind logo

Google DeepMind

Origin: United States · UK
Flagship models

Gemini 2.5 Pro · Gemini Flash · Gemma

Meta AI logo

Meta AI

Origin: United States
Flagship models

Llama 3.1 · Llama 4 · Code Llama

xAI logo

xAI

Origin: United States
Flagship models

Grok 3 · Grok 2 · Grok Vision

Microsoft logo

Microsoft

Origin: United States
Flagship models

Phi-4 · Phi-3.5 · MAI-1

NVIDIA logo

NVIDIA

Origin: United States
Flagship models

Nemotron 70B · NVLM · Cosmos

AWS Bedrock logo

AWS Bedrock

Origin: United States
Flagship models

Nova Pro · Nova Lite · Titan

Mistral AI logo

Mistral AI

Origin: France
Flagship models

Mistral Large 2 · Codestral · Pixtral

Cohere logo

Cohere

Origin: Canada
Flagship models

Command R+ · Command R · Aya

Stability AI logo

Stability AI

Origin: United Kingdom
Flagship models

Stable Diffusion 3.5 · SDXL · SD3

Black Forest Labs logo

Black Forest Labs

Origin: Germany
Flagship models

FLUX.1 Pro · FLUX.1 Dev · FLUX Schnell

D

Databricks

Origin: United States
Flagship models

DBRX Instruct · DBRX Base

AI21 Labs logo

AI21 Labs

Origin: Israel
Flagship models

Jamba 1.5 Large · Jamba 1.5 Mini

TII Falcon logo

TII Falcon

Origin: United Arab Emirates
Flagship models

Falcon 180B · Falcon 2 · Falcon Mamba

Hugging Face logo

Hugging Face

Origin: United States · France
Flagship models

SmolLM · StarCoder · BigCode

What is AI Compute?

How the Qubrid AI Platform Works

From device deployment to token circulation — understand the Qubrid AI compute token model in five steps.

  1. 1

    Choose a Compute Device

    From RTX to B200, GPU configurations for inference, training, and HPC workloads. All devices are hosted in US Tier 3+ data centers.

  2. 2

    Deploy & Go Live

    Once provisioned, your device connects to the Qubrid AI compute network with enterprise networking and 24/7 NOC support — ready in minutes.

  3. 3

    Process AI Workloads

    Devices process AI training, inference, and HPC tasks from enterprise clients. The platform monitors device status, utilization, and task completion in real time.

  4. 4

    Record Compute Contributions

    Each completed compute task is recorded as a platform token — a traceable record of the device's compute output within the platform.

  5. 5

    Tokens Circulate on the Marketplace

    Tokens can be listed on the Token Hall, connecting with enterprise AI teams that need compute capacity. Supply and demand determine price, with transparent settlement and delivery.

Live Compute Network

A live view of the GPU floor

Status, workloads, regions, throughput — refreshed continuously, never simulated for show.

GPU Fleet · Right Now

A snapshot of the network. Hover a row for details on the full dashboard.

Updated every second
GPU ID Status Temp Power Util Workload Token Output

AI Workloads Running Now

The most common job types served on the network today.

Network Regions

Capacity is distributed across continents to minimise latency.

Device Availability

Participate in the network

Access shares of real GPU nodes. View real compute output. No fixed yield — full transparency.

RTX 4090 Node · Cluster #A12

RTX 4090 × 8
Online
Online
99.2%
Avg Utilization
74%
Workloads
Flux · SDXL · ComfyUI
Daily Token Output
1.2M
Estimated — varies with demand
Power Draw
421 W
Temperature
67°C
Available Shares 412 / 3,000
View Node

H100 Node · Cluster #B07

H100 SXM × 8
Online
Online
99.8%
Avg Utilization
88%
Workloads
Llama 3 · Mistral Large
Daily Token Output
4.8M
Estimated — varies with demand
Power Draw
488 W
Temperature
64°C
Available Shares 89 / 1,500
View Node

L40 Node · Cluster #C03

L40 × 8
Online
Online
98.7%
Avg Utilization
71%
Workloads
SDXL · Video Diffusion
Daily Token Output
0.9M
Estimated — varies with demand
Power Draw
312 W
Temperature
61°C
Available Shares 1,140 / 4,500
View Node
Developers

Deploy AI workloads in minutes

A production-grade API for inference, training, and rendering. No racks to manage, no contracts to negotiate.

REST API

OpenAI-compatible endpoints for chat, completions, and embeddings. Drop-in for existing apps.

Docker

Push your own container images and run them on idle GPUs across the network on demand.

vLLM

Tuned vLLM serving for Llama, Mistral, Mixtral, Falcon, and your private fine-tunes.

ComfyUI

Serverless ComfyUI workflows for SDXL, Flux, and custom image / video pipelines.

LoRA Training

Distributed LoRA fine-tuning with one-line job submission and automatic checkpoints.

Inference Endpoints

Dedicated endpoints with autoscaling, BYO-model, and per-second billing.

Live · ~/qubridus
# Chat completion · OpenAI-compatible
curl https://api.qubridus.com/v1/chat/completions \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "llama-3-70b",
"messages": [{ "role": "user", "content": "Hello." }]
}'
# → routed to the next available GPU · per-second billing
 
# Embeddings · drop-in for vector DBs
curl https://api.qubridus.com/v1/embeddings \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-d '{
"model": "text-embedding-3-large",
"input": "Qubrid AI compute network"
}'
# → 3072-dim vector returned in < 80ms p50
 
# Image generation · SDXL / FLUX
curl https://api.qubridus.com/v1/images/generations \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-d '{
"model": "flux-1-pro",
"prompt": "a midcentury data center, isometric",
"size": "1024x1024"
}'
# → signed URL, valid 24h · region-aware
 
# LoRA fine-tune · one-line submit
curl https://api.qubridus.com/v1/jobs \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-d '{
"task": "lora-finetune",
"base_model": "llama-3-8b",
"dataset": "s3://bucket/train.jsonl",
"epochs": 3
}'
# → auto checkpointing · resume across nodes
 
# Chat completion · OpenAI-compatible
curl https://api.qubridus.com/v1/chat/completions \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "llama-3-70b",
"messages": [{ "role": "user", "content": "Hello." }]
}'
# → routed to the next available GPU · per-second billing
 
# Embeddings · drop-in for vector DBs
curl https://api.qubridus.com/v1/embeddings \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-d '{
"model": "text-embedding-3-large",
"input": "Qubrid AI compute network"
}'
# → 3072-dim vector returned in < 80ms p50
 
# Image generation · SDXL / FLUX
curl https://api.qubridus.com/v1/images/generations \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-d '{
"model": "flux-1-pro",
"prompt": "a midcentury data center, isometric",
"size": "1024x1024"
}'
# → signed URL, valid 24h · region-aware
 
# LoRA fine-tune · one-line submit
curl https://api.qubridus.com/v1/jobs \
-H "Authorization: Bearer $QUBRIDUS_API_KEY" \
-d '{
"task": "lora-finetune",
"base_model": "llama-3-8b",
"dataset": "s3://bucket/train.jsonl",
"epochs": 3
}'
# → auto checkpointing · resume across nodes
 
$