Runpod

What we found

Runpod is a GPU cloud for AI developers. You can rent GPU machines by the second as Pods, choosing between Community Cloud, which runs on peer-to-peer providers with variable reliability, and Secure Cloud, in tier 3 and 4 data centres; run serverless endpoints whose workers scale to zero, with what Runpod says are sub-200-millisecond cold starts through its FlashBoot technology; or rent multi-GPU clusters of up to 64 GPUs, or reserved clusters on longer terms. Spot instances are cheaper but can be evicted when demand rises. It offers templates through its Hub, per-request public endpoints for popular models, network storage, a command-line tool, a Python SDK and an MCP server. Pods run in 31 regions, including Japan, India and Australia, but none in Southeast Asia.

Runpod, Inc., co-founded by chief executive Zhen Lu and Pardeep Singh, raised $20 million co-led by Intel Capital and Dell Technologies Capital in 2024 and a $100 million growth round led by Summit Partners in June 2026 at a $1 billion valuation. It says more than a million developers use it and that its serverless platform has handled more than 20 billion inference requests.

Prices changed on 27 September 2026. Community Cloud starts at $0.16 an hour for an RTX A5000, with an RTX 4090 at $0.34, an A100 SXM at $1.39 and an H100 SXM at $2.69. Secure Cloud costs more: $0.74 for an RTX 4090, $1.59 for an A100 and $3.49 for an H100 SXM, up to $7.89 for a B300, so the Community discount ranges from about 13% to more than half depending on the card. Serverless workers cost $0.58 to $9.98 an hour, and storage $0.05 to $0.14 per gigabyte a month. There is no free tier: accounts need a $10 minimum deposit, though referral bonuses and a startup programme, which prefers venture-backed companies, offer credits.

It suits developers and start-ups wanting cheap, flexible GPU access without a long-term cloud contract.

Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.