Hosts obsess over the hourly GPU price and set storage and bandwidth pricing once, on listing day, half at random. Renters see all three numbers in the quote. A machine with a sharp hourly rate and an aggressive storage price loses rentals silently: no error, no feedback, just renters clicking the listing below yours. Here is what the market actually charges and where to set your numbers.
The quote renters actually see
When a renter configures an instance on Vast.ai, the platform shows the full projected cost: the GPU hourly rate, plus storage at your per GB-month price scaled to the disk they request, plus bandwidth charges for data in and out. A training job that needs a 200 GB disk and pulls a 40 GB dataset is comparing totals, not hourly rates.
This is why storage and bandwidth are pricing decisions, not administrative details. They are part of your competitive price whether you think about them or not.
Storage: the market has already decided
As of mid-2026 market research, storage pricing on Vast.ai is effectively standard at about $0.20 per GB-month, and the number holds across GPU counts. Hosts with one card and hosts with eight charge the same, because renters treat it as a commodity with a known price.
That standardization cuts both ways:
- Price at the standard and storage disappears as a factor. Renters see the number they expected and move on to comparing what actually differs between machines.
- Price well above it and the quote looks wrong. A renter needing a 500 GB disk sees your storage line jump out, and because the cost shows up right in the rental quote, they do not need to do any math to notice. They just pick someone else. You never find out why.
- Price far below it and you gain little. Storage is rarely the deciding factor when it is at or under the expected number, so the discount buys almost no extra rentals.
High storage pricing is a silent killer
Bandwidth: more variance, more ways to lose
Bandwidth pricing varies far more between hosts than storage does, and that variance is where listings quietly die. The same mid-2026 research found extreme bandwidth prices are one of the most common hidden reasons an otherwise healthy machine does not rent. Data-heavy renters, the ones moving datasets and model checkpoints, filter hard on transfer costs, and a hostile bandwidth price reads as a trap.
A sane approach:
- Know your real cost. A flat-rate home or business connection means marginal transfer costs you almost nothing; a metered datacenter link means it costs something specific per TB. Price from that number, not from fear of abuse.
- Stay in the unremarkable range for your connection class. Your bandwidth price should never be the most interesting number in your listing.
- Do not use bandwidth pricing as a filter against heavy users. It filters out light users too, because they see the rate before they know their usage.
A worked quote comparison
Two machines, same GPU, and the renter needs 200 GB of disk for a month-long job with 100 GB of transfer. Prices as a renter compares them:
| Line item | Host A | Host B |
|---|---|---|
| GPU price, 720 hrs at $0.38 vs $0.40 | $273.60 | $288.00 |
| Storage, 200 GB-month at $0.20 vs $0.80 | $40.00 | $160.00 |
| Bandwidth, 100 GB | $2.00 | $25.00 |
| Renter's total | $315.60 | $473.00 |
Host B is only two cents more per hour in the search results and 50% more expensive in reality. Storage and bandwidth turned a competitive machine into an expensive one, and Host B's only symptom is a vague sense that the machine never rents. Numbers here are illustrative, drawn from mid-2026 market research; check current norms before you copy them forever.
Audit your own listing like a renter
The fastest way to catch a quote problem is to stop looking at your listing as a host. Open the search page as a renter would, find your machine, and configure a realistic instance against it: 100 GB of disk, then 500 GB, then a heavy-transfer scenario. Watch how the projected cost moves relative to the machines listed around yours. If your total drifts from competitive to expensive as the disk grows, your storage price is doing the damage. If light usage looks fine but the transfer line dominates a data-heavy configuration, bandwidth is the leak.
Ten minutes of this once in a while beats weeks of wondering why a well-priced machine underperforms. It is also the only view that shows you what the renter compares, which is the whole quote, in one column, next to everyone else's.
Set them once, sanity-check them yearly
The practical playbook: set storage at the standard rate, set bandwidth from your real connection cost plus a small margin, and then leave both alone. Recheck them when you change ISPs, when the platform changes its quote presentation, or once a year, whichever comes first. If you are still setting up your listing, our complete Vast.ai host walkthrough covers where these settings live. And to understand why the hourly number deserves the rest of your attention, read how Vast.ai pricing actually works.