Spheron
6.1/10Spheron is a GPU cloud marketplace that gives AI teams on-demand, spot and reserved NVIDIA and AMD GPUs from several partner data centers through one account.
Data last updated November 2026 · Our methodology
Overview
Spheron pools GPU capacity from several partner data centers and sells it through one dashboard, one REST API and one bill. Its providers include Verda, Massed Compute and several Spheron-branded pools. You can rent VMs or bare metal nodes with root SSH access and a dedicated IP. Hardware ranges from an RTX 4090 up to H100, H200, B200, B300 and AMD MI300X, and multi-node clusters with InfiniBand are available. It also sells CPU nodes, containers, persistent volumes, and managed inference endpoints for open models. An Agent GPU Sandbox lets MCP clients such as Claude Code or Cursor launch GPU VMs within a spend cap you set. Pricing is per GPU-hour with no contract. On-demand runs are billed per minute after a 20-minute minimum. Spot capacity is roughly 26 to 61 percent cheaper but can be reclaimed without notice. Reserved commitments and custom clusters of 8 to 512+ GPUs are quoted by sales. Bandwidth and NVMe storage are included in the hourly rate, and payment can be made by card, bank transfer or the USDT and USDC stablecoins. The service suits startups, AI labs and ML engineers who want cheaper H100-class hardware than the hyperscalers without signing up with each provider separately. It is not a general web host: there is no managed WordPress, managed database or CDN. The main limitation is that the experience depends on the partner. Regions, compliance certifications such as ISO 27001, SOC 2 or HIPAA, and storage behaviour all differ by provider. On top of that, independent user reviews are almost nonexistent.
Our Rating
AI Software6.3/10
Hosting & Infrastructure5.9/10
Who Spheron is best for
Best software for DevelopersGood fit
ML engineers get root SSH, Docker and framework templates, a REST API and MCP-based agent provisioning.
There are no contracts, billing is per minute and spot prices are low, which suits AI startups watching GPU spend.
Less suited
Stanford, UC Berkeley and the University of Washington are listed as customers for research compute.
AI SaaS products can use the managed inference endpoints or reserved clusters, but it offers no general app-hosting stack.
Reserved clusters up to 512+ GPUs and partner-level SOC 2 and HIPAA exist, but compliance varies by provider and there is no published SLA for GPU rentals.
Pros & Cons
Pros
- One account, one API and one bill cover GPUs from several partner data centers, so you avoid lock-in to a single provider.
- Published H100 on-demand pricing ($3.96/hr) is well below the AWS and Azure list rates Spheron cites, and spot is cheaper still at $2.21/hr.
- The hourly rate includes NVMe storage and bandwidth, and the docs state there are no ingress or egress charges.
- Choice of VM or bare metal with root SSH, multi-GPU nodes up to 8x, and InfiniBand clusters for distributed training.
- The REST API covers deployments, volumes, SSH keys and teams, and an MCP-based Agent GPU Sandbox lets AI coding agents launch GPUs under a spend cap.
- Accepts cards, bank transfer and the USDT and USDC stablecoins.
Cons
- Regions, compliance certifications and volume features all depend on the partner provider, so they are not uniform.
- There is almost no independent user feedback: Trustpilot shows zero reviews and no G2 or Reddit discussion turned up.
- GPU rentals carry no published uptime SLA. The docs mention only 99% uptime, without SLA terms or credits.
- No Kubernetes, Terraform provider or CLI is documented. Automation relies on the REST API and MCP.
- Spot instances can be reclaimed without notice, and on-demand runs have a 20-minute minimum.
- Live marketplace prices change, and some GPUs show only stale 'last known' rates.
Key Features
Spheron Pricing
Managed inference
- Quote on request
- Zero data retention
- Guaranteed concurrency and an uptime SLA
- OpenAI-compatible API and dedicated endpoints for open models
On-demand
- Not reclaimable
- NVMe storage and bandwidth included
- Priced per GPU-hour and billed per minute
- VM or bare metal with root SSH and a dedicated IP
- 20-minute minimum runtime (60 minutes for 2x MI300X)
- Examples: H100 $3.96/hr, A100 $1.62/hr, RTX 4090 $0.59/hr, B200 $10.17/hr
Reserved / Custom clusters
- Contact sales
- Guaranteed capacity at volume rates
- Sourcing typically takes 24-48 hours
- Clusters of 8 to 512+ GPUs with InfiniBand
- GB200, GB300 and Vera Rubin are reservation-only
Spot
- About 26-61% below on-demand
- Separate pool that can be reclaimed without notice
- Examples: H100 $2.21/hr, B200 $5.25/hr, L40S $1.24/hr
Frequently Asked Questions
On-demand GPUs are priced per GPU-hour and billed per minute with no contract, after a 20-minute minimum runtime (60 minutes for 2x MI300X). Stopped instances are billed at the disk rate. Reserved capacity and custom clusters are quoted by sales.
On-demand instances are dedicated and cannot be reclaimed. Spot instances come from a separate pool at roughly 26 to 61 percent off but can be reclaimed without notice.
Spheron says the hourly rate includes NVMe storage, bandwidth and a dedicated IP, and the docs state there are no ingress or egress fees. Persistent volumes that exist independently of instances can also be created through the API.
It depends on the provider. Spheron says HIPAA, ISO 27001 and SOC 2 Type I/II are available, and the docs attribute ISO 27001 and GDPR to Verda and HIPAA and SOC 2 Type II to Massed Compute. A GDPR data processing agreement is published.
Spheron began in 2020 as a web3 infrastructure project. Its current docs state it is not a blockchain network but an aggregated GPU cloud that pools capacity from partner data centers.
Yes. A REST API with bearer-token authentication covers deployments, pause and resume, volumes, SSH keys, teams and balance. An MCP endpoint also lets agents such as Claude Code provision GPU VMs within a per-hour spend cap.
Payment is by credit card, bank transfer, or the USDT and USDC stablecoins.
Spheron lists the US, Canada and Europe, and its docs also mention Asia Pacific, including India (Noida and Chennai). You choose the region at deploy time based on which providers have capacity.
Similar Tools
See all alternatives
Bare metal cloud provider that offers dedicated servers, GPU servers and virtual machines on demand, billed hourly or monthly, for developers and infrastructure-heavy businesses.
French cloud provider offering shared web hosting, VPS, dedicated servers and an OpenStack-based public cloud for businesses and developers.
Vultr is an unmanaged cloud infrastructure provider that offers virtual machines, bare metal, GPUs, Kubernetes, storage and managed databases, aimed at developers and businesses.
Ranked near Spheron in AI Software
See Spheron in the listSpheron is #231 in AI Software by our score.
