GPU Clouds — procurement report
ProductArena · rankings as of 2026-09-05 · evidence as of 2026-09-05 · 5 products · 54 judged requirements · 270 judged cells
Methodology: Every product is judged against a shared taxonomy of user stories using cited evidence — hands-on probes > repository code > independent community sources > vendor claims — never opinion. Full writeup: https://ultrametric.ai/productarena/methodology
Leaderboard
| # | Product | PA Score | Coverage score | Applicable cells | Confidence |
|---|---|---|---|---|---|
| 1 | Vast.ai | 40.1 | 39.6 | 49/54 | A |
| 2 | Runpod | 33.5 | 40.9 | 50/54 | A |
| 3 | Lambda | 24.1 | 34.2 | 46/54 | A |
| 4 | CoreWeave | 20.3 | 20.6 | 48/54 | C |
| 5 | Paperspace | 17.6 | 16.0 | 47/54 | B |
PA Score = agent-readiness blend (see methodology). Coverage score = weighted share of judged requirements met. Confidence = how much of the score rests on tested vs claimed evidence (A–D).
Uncertainty note
The current #1/#2 gap in this arena is not close enough to qualify for the multi-judge uncertainty pass (or the pass has not covered it yet) — no extra caveat applies beyond the per-product confidence grades above.
Buyer checklist (RFP)
The arena's 54 judged user stories as requirements, grouped by theme. Priorities mirror the story weights our scoring uses (3 = must-have, 2 = should-have, 1 = nice-to-have). Interactive version with per-requirement verdicts for the top products: /arena/gpu-clouds/checklist
Access connectivity — stories about access connectivity in this arenaAccess connectivity
Stories about access connectivity in this arena
- developerSSH into my GPU instance with my own keys and get root-level control of the environmentmust-have
- developerOpen Jupyter or connect my IDE (VS Code/Cursor) to the instance in one stepnice-to-have
- developerExpose ports to serve applications from my instance and connect instances over private networkingnice-to-have
Agenticness — how well agents can access and operate the productAgenticness
How well agents can access and operate the product
- ai-native userPlug MCP servers into this product so it can use their toolsmust-have
- ai-native userConnect an agent via an official MCP servermust-have
- ai-native userDrive the product through a documented public APImust-have
- ai-native userDelegate tasks to a built-in AI assistant inside the productmust-have
- ai-native userPoint an agent at llms.txt or agent-oriented docsshould-have
- ai-native userRun the product headlessly / in CI for automationshould-have
- ai-native userUse an official CLIshould-have
- ai-native userIssue scoped/least-privilege API credentials for an agentshould-have
- ai-native userBuild against official SDKsshould-have
- ai-native userSubscribe to events via webhooksshould-have
- ai-native userGet AI-generated insights and suggestions from my data inside the productshould-have
- ai-native userSet up automations that run autonomously in the backgroundshould-have
- ai-native userOperate the product with natural-language commandsshould-have
- ai-native userExplore an interactive API reference with runnable examplesshould-have
- ai-native userDownload a machine-readable API spec (OpenAPI or equivalent)should-have
- ai-native userRely on versioned APIs with a documented deprecation policyshould-have
- ai-native userTest against a sandbox environment without touching production datanice-to-have
Automation depth — how much of the product can run unattendedAutomation depth
How much of the product can run unattended
- ai-native userDefine rules that trigger actions automatically on eventsmust-have
- ai-native userPerform bulk operations across many items at onceshould-have
- ai-native userSchedule recurring jobs or workflowsshould-have
- ai-native userVersion, review, and roll back my automationsnice-to-have
Capacity availability — stories about capacity availability in this arenaCapacity availability
Stories about capacity availability in this arena
- ml engineerSee real-time GPU availability by type and region before I try to provision, instead of discovering stockouts by failureshould-have
- ml engineerChoose from current-generation datacenter GPUs (H100/H200/B200 class) as well as cheaper previous-generation optionsshould-have
- platform engineerSee documented quotas and instance limits and raise them through a defined processshould-have
Clusters scale — stories about clusters scale in this arenaClusters scale
Stories about clusters scale in this arena
- ml engineerProvision a multi-node GPU cluster with fast interconnect for distributed training without a sales cyclemust-have
- platform engineerSchedule jobs on managed Slurm or Kubernetes instead of building my own scheduler on raw nodesshould-have
Openness — open source, data portability, and self-hosting storiesOpenness
Open source, data portability, and self-hosting stories
- ai-native userExport all of my data in open formats and leavemust-have
- ai-native userSelf-host the core productmust-have
- ai-native userDo everything through the API that I can do in the UIshould-have
- ai-native userRead the product's source under an open licenseshould-have
Pricing billing — stories about pricing billing in this arenaPricing billing
Stories about pricing billing in this arena
- ml engineerI am billed at per-second or per-minute granularity and only while my instance is actually runningmust-have
- ml engineerSee the published per-GPU-hour price for every GPU type on a public pricing page without talking to salesmust-have
- ml engineerRent spot or interruptible GPU capacity at a deep discount with clearly documented preemption semanticsmust-have
- platform engineerPull usage and billing breakdowns programmatically to attribute GPU spend by team or workloadshould-have
- ai agentQuery the GPU catalog with live pricing and availability from a public or documented endpoint before committing any spendshould-have
- platform engineerLock in reserved or committed-use discounts for sustained GPU capacitynice-to-have
Privacy posture — data-handling and privacy storiesPrivacy posture
Data-handling and privacy stories
- ai-native userPrevent my data from being used to train AI modelsmust-have
- ai-native userChoose where my data is stored (region/residency)should-have
- ai-native userControl data retention and deletionshould-have
- ai-native userOpt out of telemetry and usage trackingshould-have
Provisioning lifecycle — creating, updating, and tearing down resources across their lifecycleProvisioning lifecycle
Creating, updating, and tearing down resources across their lifecycle
- ai agentProvision a GPU, monitor it, run a workload, and tear it down end to end through documented APIs, CLI, or MCP without a human in the consolemust-have
- developerProvision an on-demand GPU instance from the console or API and be running code on it within minutesmust-have
- ml engineerSet auto-shutdown timers or spend limits so a forgotten instance can't silently run up a huge billshould-have
- developerStart, stop, restart, and terminate instances programmatically and keep paying only for what is runningshould-have
Serverless endpoints — stories about serverless endpoints in this arenaServerless endpoints
Stories about serverless endpoints in this arena
- developerDeploy code to autoscaling serverless GPU workers that scale to zero, instead of managing always-on instancesshould-have
Storage data — storing and moving data — persistence, formats, durabilityStorage data
Storing and moving data — persistence, formats, durability
- developerAttach persistent network storage that survives instance teardown, so datasets and checkpoints outlive any single GPU rentalmust-have
- developerMove data in and out efficiently — S3-compatible endpoints, cloud-storage sync, or documented transfer toolingshould-have
Templates images — stories about templates images in this arenaTemplates images
Stories about templates images in this arena
- developerRun my own Docker image or custom machine template with my exact environmentshould-have
- developerLaunch from pre-built ML templates (PyTorch, CUDA, vLLM, ComfyUI) instead of assembling an environment from scratchshould-have
Trust governance — stories about trust governance in this arenaTrust governance
Stories about trust governance in this arena
- platform engineerVerify the provider's security and compliance posture (SOC 2, data handling, datacenter tiers) before putting proprietary models on itshould-have
- platform engineerManage team members with roles and scoped API keys so credentials and spend stay controllednice-to-have
Pricing signals
Extracted verbatim from each vendor's own pricing page — never converted, averaged, or derived. Products whose page prints no unit price are recorded as unclear, honestly.
| Product | Headline price | Unit | As of |
|---|---|---|---|
| Vast.ai | pricing unclear — The page describes pricing tiers and platform mechanics qualitatively but does not print any specific dollar figures or GPU-hour rates in the provided markdown. | 2026-09-07 | |
| Runpod | $0.27 | GPU-hour | 2026-09-07 |
| Lambda | $0.79 | GPU-hour | 2026-09-07 |
| CoreWeave | $2.7 | GPU-hour | 2026-09-07 |
| Paperspace | $0.76 | GPU-hour | 2026-09-07 |
Appendix: recorded probes
Hands-on probe recordings — transcripts/videos a human can replay, the strongest evidence tier. Watch them at https://ultrametric.ai/productarena/proofs
- CoreWeave
curl -si https://api.coreweave.com/v1beta1/cks/clustersterminal · recorded 2026-09-05 · exit 0 - CoreWeave
curl -s -X POST https://docs.coreweave.com/mcp -H 'Content-Type: application/json' -d '<jsonrpc initialize>'terminal · recorded 2026-09-05 · exit 0 - Lambda
curl -s https://cloud.lambda.ai/api/v1/openapi.json | head -c 400 && curl -si https://cloud.lambda.ai/api/v1/instancesterminal · recorded 2026-09-05 · exit 0 - Paperspace
HOME=$(mktemp -d) sh -c 'curl -fsSL https://paperspace.com/install.sh | sh && ~/.paperspace/bin/pspace version && ~/.paperspace/bin/pspace --help'terminal · recorded 2026-09-05 · exit 0 - Runpod
runpodctl --helpterminal · recorded 2026-09-05 · exit 0 - Runpod
runpodctl version # installed via `brew install runpod/runpodctl/runpodctl`terminal · recorded 2026-09-05 · exit 0 - Runpod
curl -s -X POST https://docs.runpod.io/mcp -H 'Content-Type: application/json' -d '<jsonrpc initialize>'terminal · recorded 2026-09-05 · exit 0 - Runpod
curl -si -X POST https://mcp.getrunpod.io/ -H 'Content-Type: application/json' -d '<jsonrpc initialize>'terminal · recorded 2026-09-05 · exit 0 - Runpod
curl -s https://rest.runpod.io/v1/openapi.json | head -c 300 && curl -si https://rest.runpod.io/v1/podsterminal · recorded 2026-09-05 · exit 0 - Vast.ai
curl -s 'https://console.vast.ai/api/v0/bundles/' | head -c 600terminal · recorded 2026-09-05 · exit 0 - Vast.ai
uvx --from vastai vastai --helpterminal · recorded 2026-09-05 · exit 0 - Vast.ai
uvx --from vastai vastai search offers 'gpu_name=RTX_4090 num_gpus=1' -o 'dph' | head -12terminal · recorded 2026-09-05 · exit 0
Cite as: ProductArena by Ultrametric Inc, GPU Clouds arena, rankings as of 2026-09-05 — https://ultrametric.ai/productarena/arena/gpu-clouds
License: © 2026 Ultrametric Inc. Brief quotation of individual verdicts, scores, or evidence excerpts is permitted with attribution to "ProductArena by Ultrametric Inc (ultrametric.ai/productarena)", as is use of the data to evaluate, contest, or contribute corrections. Bulk copying, redistribution, or use to build competing datasets requires prior written permission (see DATA-LICENSE in the repository).
No liability: rankings, verdicts, and scores are research outputs derived from the cited evidence at a point in time, provided "as is", without warranties. Ultrametric Inc accepts no responsibility for procurement, purchasing, or other decisions made in reliance on them — verify against the cited evidence before acting (https://ultrametric.ai/productarena/terms).