#1
Voltage Park
US neocloud selling on-demand and reserved bare-metal NVIDIA HGX H100 clusters from $1.99/hr; acquired TensorDock (2025) and merged with Lightning AI (Jan 2026).
from $1.99 per GPU/hr
LLM inference
Rent training-class GPUs by the hour for LLM inference and fine-tuning, pay with crypto.
Short answer: Voltage Park is our top-rated inference provider as of 2026-09-24 (7.8/10), from $1.99 per GPU/hr.
#2
Lambda (Lambda Labs)
Established US GPU cloud: on-demand H100/B200 instances, 1-Click Clusters and Superclusters, billed per minute with no ingress/egress charges.
from $0.690 per GPU/hr
#3
Akash Network
Open permissionless marketplace for containerised GPU/CPU compute; providers bid, tenants pay per block.
from $0.090 per GPU/hr min $0.500
#4
Together AI
Centralized open-model inference and GPU-cluster provider with published per-token and per-GPU rates; fiat-only and not decentralized.
from $1.99 per GPU/hr min $5
#5
Runpod
Developer GPU cloud with published per-second rates across Community and Secure Cloud, plus serverless endpoints and clusters.
from $0.160 per GPU/hr min $10
#6
Prime Intellect
US (Delaware) marketplace for aggregated on-demand GPU instances and token-priced hosted RL training/inference, with an open-source AI research profile.
from $0.470 per GPU/hr
#7
Vast.ai
Host-set GPU marketplace with per-second billing; on-demand, interruptible and reserved tiers, plus serverless and clusters.
no KYC
from $0.088 per GPU/hr
#8
Thunder Compute
Centralized GPU cloud built on the company's own GPU-virtualisation layer: cheap per-minute A100/L40/H100/A6000 instances; no crypto, no token.
from $0.350 per GPU/hr
#9
Hyperbolic
US curated GPU cloud and OpenAI-compatible inference; H100 $3.19/hr, H200 $3.99/hr on-demand with hourly billing and $5 never-expiring credits.
from $3.19 per GPU/hr min $5
#10
Aleph Cloud (Aleph.im)
Decentralized cloud (VPS, GPU instances, storage, confidential VMs) on a node network; paid via ALEPH/USDC/card credits.
no KYC
from $0.055 per GPU/hr
#11
Spheron Network
Singapore-run GPU aggregator reselling capacity from 5+ neoclouds; H100 from $2.64/hr on-demand, per-minute billing.
from $0.480 per GPU/hr
#12
Theta EdgeCloud
Hybrid cloud+edge GPU marketplace with cheap consumer-GPU rates (RTX 3090 $0.14/hr, A100 $1.99/hr) and TFUEL-linked node rewards; cloud-launched in 2024. Facing an LA lawsuit alleging fraud/token manipulation.
from $0.050 per GPU/hr
#13
Nosana
Solana-based self-serve GPU marketplace for AI inference, CI and agent workloads; live price table, consumer-GPU supply.
from $0.048 per GPU/hr
#14
Gonka (Gonka.ai)
Cosmos-SDK L1 for decentralized open-weight LLM inference; hosts earn GNK, developers use OpenAI-compatible brokers.
price on request
#15
Fluence
Swiss-DAO multi-provider neocloud; H200 from $2.80/hr with zero egress, but Console access is still gated to an Alpha VM testing program and top-ups run only in USDC via CopperX.
from $2.80 per GPU/hr min $10
#16
io.net
Solana-based aggregated GPU/CPU network: clusters, VMs, containers, inference, and Intel TDX confidential compute.
price on request
#17
Aethir
Distributed enterprise/gaming GPU cloud (Aethir Earth for AI, Aethir Atmosphere for cloud gaming) with staking and slashing.
price on request
#18
Clore.ai
Permissionless peer-to-peer GPU marketplace paying in crypto, with on-demand and biddable spot tiers and no disclosed legal entity.
watch watch
from $0.027 per GPU/hr
#19
Exabits
Tokenised enterprise-GPU compute project (GB200/H100/H200/RTX 5090 claims) with a JS-only site, no public pricing and no disclosed legal entity.
unverified verify
price on request
#20
Flux
DePIN cloud with a FluxEdge GPU marketplace; its UK operator entered administration on 4 September 2026.
watch watch
price on request
// FAQ
How much VRAM does my model need?
A rough rule is 2 bytes per parameter at FP16 (so a 7B model needs ~14 GB plus KV-cache). The <a href="/tools/gpu-workload/">GPU workload calculator</a> estimates it and picks a fitting GPU.