CERTAIN CONTENT THAT APPEARS ON THIS SITE COMES FROM AMAZON. THIS CONTENT IS PROVIDED 'AS IS' AND IS SUBJECT TO CHANGE OR REMOVAL AT ANY TIME.
This hub tracks GPU clouds and VPS/CPU providers in one place. Prices are list rates with a verification state; rows marked Measured carry Argusic's own operational data (spin-up time, real billed cost, failure rate) from running our benchmark fleet, each number linked to its run evidence. We sell no hosting and belong to no provider.
How we measure and rank
Default order is price, ascending, with verified prices first. Argusic-measured columns come from our production workloads, not synthetic tests: every pod we rent logs its spin-up seconds, preemptions, and the rate it was billed at. Community signals never affect order, and neither do affiliate commissions. Reviews are intentionally absent: this page publishes what we measured and verified, not opinions.
GPU Cloud vs. CPU Cloud: Which One Do You Need?
The two tabs above answer two different questions, and picking the wrong one is the easiest way to overpay for compute. A GPU cloud rents access to a graphics card, an H100, an A100, an RTX 4090, billed by the hour or the second. A CPU cloud, usually sold as a VPS, rents a slice of an ordinary server billed by the month. Neither is better than the other; each is built for a different kind of job.
Reach for a GPU cloud when the workload itself needs a graphics processor: training or fine-tuning a model, running local inference on an open-weight LLM, generating images or video in bulk, or anything that calls CUDA or a tensor core directly. A CPU-only VPS cannot speed that work up; it either runs on a GPU or it is not worth paying to run at scale. Argusic's own list of tested self-hosted AI apps is a quick way to see which open-source tools actually expect a GPU underneath before you rent one.
Reach for a CPU cloud when the workload is bound by network, memory, or plain logic instead: a website, a database, a CI runner, most Docker deployments, and most AI agents, which spend their time calling a model's API rather than running the weights themselves. A VPS with enough RAM for the process and a fast enough disk for the database usually costs a fraction of even the cheapest GPU-hour; browse every subject Argusic tests and most of it runs this way.
As a rule of thumb: if the job is run this model, rent a GPU by the hour only while it runs. If the job is keep this service up, rent a VPS by the month. Argusic tests both sides of that choice the same way it tests everything on the home page: install it, run it, and record what actually happened.
Frequently asked questions
Which provider has the cheapest verified H100 right now?
The summary line at the top of the GPU tab computes this live from the catalog: the lowest verified on-demand H100 rate among tracked providers, with its verification date.
What does the Measured badge mean?
Argusic runs its own benchmark fleet on some of these providers. Measured rows carry operational data from that usage, each number linked to run logs. Where a metric is not recorded yet, the row says so instead of showing a figure.
Do affiliate links change the ranking?
No. Order derives only from prices, stated criteria, or measurements. Providers without any affiliate program list on equal footing, and the disclosure sits at the top of the page.
Why is a row marked not yet verified?
It is listed for coverage: the provider exists and the plan is real, but Argusic has not yet confirmed the price against the provider's own pricing page or API. Such rows show no figure until they are verified.
What is the difference between API-verified and Owner-verified?
API-verified rows are refreshed daily straight from the provider's own pricing API, and the response is stored as the source. Owner-verified rows were checked by hand against the provider's own console or billing on the date shown. Both carry a verification date; neither is a guess.
What is the Featured partners tab?
It shows only the providers Argusic has an affiliate partnership with: their Visit button carries a tracked link, and the partnership is the only reason they are grouped there. Prices, verification states, and the default sort are identical to the other two tabs; being a partner changes no number and no rank anywhere else on the page.
How do I narrow the list to only the plans I care about?
Use the filters on the left: datacenter location, price range, GPU model or RAM and vCPU minimums, billing interval, provider type, and stock. Every option shows how many offers still match, options with zero matches grey out, and active filters render as removable chips above the list with a Clear all button.
Does the Argusic measured facet change the default sort order?
No. Sorting stays price, ascending, unless a measured sort is picked from the dropdown. The Measured facet and badge exist to separate what Argusic itself recorded from what a provider states on its own pricing page, so the two are never confused.
Does a low price mean the plan is available right now?
Only if its stock state says so. Each offer carries an in-stock, out-of-stock, or unknown state from the same source as its price, and the Availability filter can hide anything not confirmed in stock.
How often are the prices on this page refreshed?
Providers with a public pricing API are re-fetched daily, and every attempt, successful or not, is recorded in an operations log. Providers without an API are checked by hand, and the row carries the date of that check instead.
Signed-in GitHub users can discuss providers and report price changes. Comments live on GitHub Discussions, moderated there, and never affect rankings.
Comments never affect rankings or any number on this page.