Vast.ai is the cheapest way to rent a GPU — a 4090 from ~$0.14/hr, an H100 from ~$1.79/hr (24 Sep 2026). But reliability depends on the host, not the platform. Here's what to know before you rent.
This post contains affiliate links. If you buy through them, Aliteq may earn a commission — at no extra cost to you. Prices verified at publish time.
Share
Vast.ai is the cheapest way to rent a GPU, and it's not close. It's a marketplace — independent hosts list their cards and compete on price — so as of 24 Sep 2026 an RTX 4090 goes for about $0.14/hour and an H100 from roughly $1.79/hour on the spot market (our own tracker). But 'cheapest' hides a catch most reviews skate past: on Vast, reliability depends on the host, not the platform. Here's what to actually know before you rent — how the pricing works, how the reliability really works, and who should skip it. This is a documented-fact review, part of our cloud-GPU pricing pillar.
What Vast.ai costs right now
Prices are set by the marketplace, so they move, but the shape is stable. Hosts list at a price and you pay per second while your container runs; spot (interruptible) listings are the cheapest — typically well below the on-demand rate — while on-demand costs more and won't be reclaimed. Storage and egress are billed on top. The figures below are the cheapest live offers from our tracker, captured 24 Sep 2026 (the median runs higher, and the RunPod column is on-demand for context).
Vast is a network of independent providers — its host documentation describes datacenter-verified hosts (which submit documentation and hold to uptime baselines) alongside unverified ones. You don't just pick a GPU; you pick a host, using the signals Vast shows for every listing: a reliability score (a 0–1 rating from the host's uptime and interruption history), a DLPerf score (measured deep-learning throughput — not every H100 performs the same, depending on cooling and power), tested network speed, and host ratings and reviews. Filtering by these is the whole skill of using Vast well.
The reliability question (the real catch)
Here's the part that decides whether Vast is cheap or expensive for you: reliability is host-specific, not platform-wide. A verified data-centre host with redundant power and a long uptime history is genuinely reliable. A solo operator running a gaming rig is not. Public reviews note that on the cheapest unverified hosts, restart overhead and lost training progress can push the real cost well above the sticker price — so the headline rate isn't always the real one. Spot instances can be reclaimed with only minutes' notice, so checkpoint anything you can't lose, and don't put sensitive or regulated data on a host you can't vet.
Referral link
Rent the cheapest capable GPU on Vast.ai
The marketplace almost always has the lowest spot price — a 4090 from about $0.14/hr, an H100 from ~$1.79/hr (24 Sep 2026). Filter to a verified host and checkpoint your work.
Referral link — we may earn a commission at no cost to you. Prices on our compare page are the provider's live figures, cheapest first; this never changes the ranking.
Frequently asked
Is Vast.ai safe and legit?
Yes — it's an established marketplace (running since 2018) used widely for AI, with per-second billing and containerised instances. The nuance is reliability, not legitimacy: you're renting from independent hosts of varying quality, so vet the host's reliability score, verification status and reviews. For non-sensitive, interruptible work it's a fine trade for the low price.
Is Vast.ai reliable?
It depends entirely on the host you pick, not on Vast as a platform. A verified data-centre host with strong uptime is dependable; a cheap unverified host is not. Use the reliability score and verification filter, choose on-demand or a high-reliability host for long jobs, and checkpoint on spot instances so a reclaim doesn't cost you a run.
What's the difference between spot and on-demand on Vast?
Spot (interruptible) listings are the cheapest — typically well below the on-demand price — but the host can reclaim the machine with minutes' notice. On-demand costs more and stays yours until you stop it. Use spot for interruptible inference and experiments you can checkpoint, and on-demand for anything you can't afford to lose.
Vast.ai or Runpod — which should I use?
Vast.ai for the lowest possible rate if you'll vet hosts and checkpoint; Runpod for a managed, polished experience at a slightly higher price with less to think about. We compare them directly in our Runpod vs Vast vs Lambda piece.
Referral link
Managed pods + serverless
Want managed reliability and ready-made templates instead of vetting hosts yourself? Runpod's on-demand H100 is about $1.99/hr (24 Sep 2026) — a small premium for a lot less to worry about.
Vast.ai earns its reputation: it's the cheapest rate you'll find, if you rent smart — verified hosts, checkpointing, spot for throwaway work. Compare live prices on our cloud-GPU compare page, read the pricing pillar for the full market, and see the Runpod review for the managed alternative.
