Cheapest Monthly GPU Rental in 2026: A100 & H100 Price Comparison
Written by the ApexVPS team • Last updated: September 2026 • 9 min read
What does a month of GPU rental actually cost in September 2026? For an A100 80GB running all 720 hours, the range runs from about $317 on Vast.ai’s marketplace (from $0.44/hr, if that rate held all month) to about $2,225 of hourly billing at Paperspace — with ApexVPS’s dedicated flat monthly rental at $1,290. For an H100 80GB, the span is roughly $1,433 (RunPod Community Cloud) to $2,369 (Lambda on-demand), with ApexVPS flat at $1,990. Every figure below is a public list price with a linked source, and every monthly equivalent is a simple 720-hour multiplication.
Two things matter beyond the raw number: whether the price is fixed (a flat monthly invoice) or variable (marketplace hours that reprice constantly), and whether the machine is dedicated and guaranteed or interruptible. We flag both on every row.
A100 80GB: one month, every major option
| Provider | Rate (A100 80GB) | Month of 720 h | Model |
|---|---|---|---|
| Vast.ai | from $0.44/hr (PCIe) | from ≈ $317 | Marketplace hours; prices update hourly; interruptible hosts |
| RunPod (Secure Cloud) | $1.39/hr | ≈ $1,001 | Per-second billing from a prepaid balance |
| TensorDock | from $1.42/hr | from ≈ $1,022 | Marketplace/on-demand hours |
| ApexVPS | $1,290/month flat | $1,290 — fixed | Dedicated server; unlimited usage; no meter; crypto, no KYC |
| Vultr | ~$2.40/hr | ≈ $1,728 | On-demand hourly VM |
| Lambda | $2.79/hr | ≈ $2,009 | On-demand hourly |
| Paperspace | ~$3.09/hr | ≈ $2,225 | On-demand hourly |
Public rates checked September 2026 (sources below). “From” marketplace rates assume the cheapest listed host stays available and unchanged for 720 hours — a best case, not a quote.
H100 80GB: one month, every major option
| Provider | Rate (H100 80GB) | Month of 720 h | Model |
|---|---|---|---|
| RunPod (Community Cloud) | from $1.99/hr | from ≈ $1,433 | Community hosts; per-second billing |
| Vast.ai | ~$2.00/hr typical | ≈ $1,440 | Marketplace hours; $0.90–$2.27/hr by host tier |
| TensorDock | $2.25/hr (SXM) | ≈ $1,620 | On-demand hours |
| ApexVPS | $1,990/month flat | $1,990 — fixed | Dedicated server; unlimited usage; no meter; crypto, no KYC |
| RunPod (Secure Cloud PCIe) | $2.89/hr | ≈ $2,081 | Per-second billing from a prepaid balance |
| Lambda | $3.29/hr | ≈ $2,369 | On-demand hourly |
Public rates checked September 2026 (sources below). Marketplace “typical” rates vary by host and move hourly.
How to read these tables honestly
The cheapest sticker prices all come from hourly marketplaces, and they are real — for as long as the host stays listed, at that price, with the machine actually free. A 720-hour month at a marketplace rate is a theoretical floor: hosts get occupied, prices reprice upward, and interruptible instances can stop mid-run. The flat monthly rentals are the opposite trade: you pay more than the theoretical floor and in exchange the price, the machine and the terms do not move for the whole term.
Note what a flat rental includes that an hourly GPU rate does not: at ApexVPS the monthly price covers the whole dedicated server — vCPU, RAM, NVMe storage and a bandwidth allowance (4–12 TB depending on the plan) — while hourly GPU listings frequently meter storage and bandwidth separately.
When hourly wins, when flat monthly wins
- Hourly wins for short jobs, experiments, weekend fine-tunes and anything interruptible: pay $3 for two hours of A100 and stop.
- Flat monthly wins when a GPU must run all month and the invoice must be known in advance: production inference, month-long training, research clusters on a grant budget.
- The break-even rule: divide the flat monthly price by the hourly rate. ApexVPS’s $1,290 A100 breaks even with a $2.79/hr Lambda rate at about 462 hours (64% of the month); the $1,990 H100 breaks even with a $3.29/hr rate at about 605 hours (84%).
- Tier matters: on H100, ApexVPS’s flat $1,990 undercuts even some hourly secure-cloud rates (≈$2,081 for RunPod PCIe over a full month).
So what is the cheapest monthly GPU rental in 2026?
Cheapest sticker price: a marketplace A100 at Vast.ai’s “from $0.44/hr” — if you accept host variance and interruptions. Cheapest fixed, dedicated, guaranteed price: ApexVPS’s A100 at $1,290/month and H100 at $1,990/month, unlimited usage, no hourly meter, payable in crypto with no KYC and a 30-day money-back guarantee. If that profile fits, the ApexVPS monthly GPU plans from $249 (T4, A100, H100) are the direct next step.
Sources (all checked September 2026)
- Vast.ai — A100 PCIe live pricing (from $0.44/hr)
- RunPod — cloud GPU pricing (A100 PCIe $1.39/hr, H100 PCIe $2.89/hr)
- RunPod — H100 PCIe Community Cloud (from $1.99/hr)
- TensorDock — A100 80GB from $1.42/hr
- TensorDock — homepage: H100 SXM on-demand at $2.25/hr
- Thunder Compute — provider table: Lambda A100 $2.79/hr, H100 $3.29/hr; Vast.ai H100 ~$2.00/hr (Aug 2026)
- Spheron — Vultr cloud GPU rate card: A100 80GB ~$2.40/hr (Aug 2026)
- Dataoorts — cheapest cloud GPU providers 2026: A100 ~$3.09/hr (Paperspace-class on-demand)
- ApexVPS — GPU plans and specifications
Frequently Asked Questions
What is the cheapest monthly GPU rental in 2026?
By sticker price, marketplace hourly rates win: Vast.ai lists A100 80GB from $0.44/hr (≈$317 per 720-hour month) and H100 around $2.00/hr (≈$1,440). Among fixed dedicated monthly plans, ApexVPS is the cheapest in this comparison: A100 80GB at $1,290/month and H100 80GB at $1,990/month, both with unlimited usage and no hourly meter.
Is hourly or monthly GPU rental cheaper?
Hourly is cheaper below the break-even point; flat monthly above it. Divide the monthly price by the hourly rate: ApexVPS’s $1,290 A100 breaks even with a $2.79/hr on-demand rate at ~462 hours (64% of a month). Run less than that and hourly wins; run 24/7 and the flat plan is both cheaper than most on-demand rates and entirely predictable.
Why are marketplace GPU prices so much lower?
Marketplaces aggregate spare capacity from independent hosts, so prices are set host-by-host, reprice hourly, and the cheapest tiers are often unverified or interruptible. You trade predictability and guarantees for price. A flat monthly rental prices the removal of that variance.
Does ApexVPS bill GPU servers by the hour?
No. ApexVPS rents dedicated GPU servers (T4, A100, H100) at flat monthly prices — $249, $1,290 and $1,990 respectively — with unlimited usage, no hourly metering, no prepaid balance, crypto-only payment and a 30-day money-back guarantee.