2026-09-28DataStorage
H100 rental prices are 15% above last year
The median on-demand H100 price was $3.42 per GPU-hour across 40 providers in the week of September 28. CME plans to launch GPU rental futures on October 5, pending regulatory review.
Live GPU prices on decentralized networks
Sources: public network APIs, with the method and fetch time shown for each number.
NVIDIA H100 (80GB)
Hourly rates across decentralized networks
Akash Network
Checked 2026-10-31 21:16 UTC
io.net
Not verified yet
Not verified yet
Aethir
Not verified yet
Not verified yet
Render Network
Not verified yet
Not verified yet
Live decentralized GPU price directory
Prices from public network APIs, updated every 6 hours.
Model class:
Showing 3 active decentralized GPU listingsView all 20+ Akash GPU models →
| Network | GPU model | VRAM | Hourly rate | Status | Source |
|---|---|---|---|---|---|
Akash NetworkChain: Cosmos | NVIDIA GeForce RTX 4090 (24GB) | 24 GB GDDR6X | $0.36/hr+5.9%, range $0.36–$0.36 Median of live Akash offers · How we calculate Based on 3 providers• Checked 2026-10-31 21:16 UTC | Active | Raw data ↗ |
Akash NetworkChain: Cosmos | NVIDIA A100 (80GB SXM4) | 80 GB HBM2e | $1.49/hr→0.0%, range $1.07–$1.91 Median of live Akash offers · How we calculate Based on 4 providers• Checked 2026-10-31 21:16 UTC | Active | Raw data ↗ |
Akash NetworkChain: Cosmos | NVIDIA H100 (80GB HBM3) | 80 GB HBM3 | $2.56/hr→0.0%, range $2.04–$3.16 Median of live Akash offers · How we calculate Based on 4 providers• Checked 2026-10-31 21:16 UTC | Active | Raw data ↗ |
Decentralized compute networks
Architectural overviews, consensus models, and technical profiles.
Chain: SolanaFounded 2017
Render Network
Render Network is a leading decentralized GPU network founded by OTOY, originally designed for distributed 3D rendering and now expanded into generative AI inference, NeRFs, and spatial computing. It connects node operators with spare consumer and workstation GPU capacity to digital creators and ML engineers requiring scalable graphics and compute power.
Target workloads:
3D Rendering & VFXAI Inference & Image GenNeRFs & Gaussian Splatting
Chain: SolanaFounded 2023
io.net
io.net is a state-of-the-art DePIN network that aggregates underutilized GPUs from independent data centers, crypto mining facilities, and consumer nodes into unified, low-latency clusters. Built on the Ray distributed computing framework, io.net allows AI engineers to deploy massive GPU clusters for model training, fine-tuning, and scalable batch inference without long-term contracts.
Target workloads:
AI Model Training & Fine-TuningBatch & Streaming InferenceDistributed Ray Clusters
Chain: CosmosFounded 2018
Akash Network
Akash Network is a sovereign Proof-of-Stake blockchain and permissionless compute marketplace built with the Cosmos SDK. It allows any cloud provider or data center to list idle CPU, GPU, memory, and storage capacity in an open reverse auction. Developers deploy containerized Docker applications using human-readable Stack Definition Language (SDL) files, with dynamic escrow billing.
Target workloads:
LLM Serving (vLLM, Ollama, TGI)Model Fine-Tuning & LoRAGeneral Containerized Web Services
Chain: ArbitrumFounded 2021
Aethir
Aethir is an enterprise-grade decentralized physical infrastructure network (DePIN) specialized in high-performance GPU compute. Rather than relying solely on consumer hardware, Aethir aggregates enterprise-grade NVIDIA H100, A100, and RTX cards distributed across global data center providers, delivering ultra-low latency compute for enterprise AI pipelines and interactive cloud gaming.
Target workloads:
Enterprise AI Model InferenceCloud Gaming & Real-Time RenderingInteractive Virtual Desktops
Best decentralized GPU by workload
Curated evaluations matching specific workloads with the optimal network and accelerator.
Workload guide
AI Inference
Low-latency, cost-effective serving for open-weight LLMs, vision, and speech models
Workload guide
3D Rendering & VFX
High-throughput parallel compute for OctaneRender, Blender, Cinema4D, and Unreal Engine
Workload guide
AI Model Fine-Tuning
Cost-effective training runs for LoRA, QLoRA, and domain adaptation with high-bandwidth accelerators
Recent DePIN compute developments
Sourced reporting on protocol releases, network growth and GPU pricing.
2026-08-24Aethir (via YesPress profile)
Aethir launches ACCELERATE to arrange AI infrastructure sites
Aethir says ACCELERATE gives access to 10 sites totaling up to 20 MW and projects up to $700 million in contracts by end of 2026. These are company projections, not completed contracts.
2026-08Salad Blog
Salad publishes architecture for putting its payment rails on-chain
Salad published the implementation design for milestone 3 of its Render Network integration, which routes every transaction through the burn-mint-equilibrium model. Customer RENDER payments would go to the customer's own on-chain settlement account.
Methodology & quality assurance
Read full methodology standards →No sponsored placements. Prices come from public network APIs and are shown with their method and fetch time. Anything we cannot source shows as 'Not verified yet'.
We do not currently use affiliate links. If that changes, links will be marked and listed on the affiliate disclosure page.