GPU Cost Calculator: Rent vs Buy
Compare costs for buying and renting GPU hardware, using live rental pricing. For more considerations and context, see our Rent vs Buy companion article.
Buy
Indicative price. Edit to match your build.
Specs / performance
32GB VRAM @ 1,792 GB/s; 575W TDP
Model sizePrefillDecode
- $7,000
- upfront
- $21–23
- monthly ongoing
Costs over 3 years$7,805
- Hardware
- $7,000
- Electricity (@ 107 kWh / month)
- $805
Assets after 3 years$3,584
- Hardware
- $3,584
Net Costs$4,221
Average monthly net cost over 3 years:$117
Rent
/ hrThe lowest available on-demand rate across 3 providers.
Specs / performance
32GB VRAM @ 1,792 GB/s; 575W TDP
Model sizePrefillDecode
- $0
- upfront
- $81–89
- monthly ongoing
Costs over 3 years$3,063
- On-demand rental (@ 173 hrs / month)
- $3,063
Assets after 3 years$0
Net Costs$3,063
Average monthly net cost over 3 years:$85
HardwareElectricityBuy (net)Rent
Show these numbers as a table
| Elapsed | Buy (cash) | Buy (net) | Rent |
|---|---|---|---|
| At purchase | $7,000 | $0 | $0 |
| 1y | $7,260 | $1,660 | $991 |
| 2y | $7,529 | $3,049 | $2,012 |
| 3y | $7,805 | $4,221 | $3,063 |
Assumptions behind these numbers
- Usage: 8h/day, 5 days/week, or a total of 173 hours/month.
- On-demand instances are only enabled and billed while in use.
- Electricity at $0.20/kWh, rising 3%/year. Power draw is 100W plus 90% of the GPU's TDP — 618W average while in use.
- Hardware value changes -20%/year, so $7,000 of kit is worth $3,584 after 3 years. Net Costs are the cash spent minus that end value; the dashed Buy (net) line plots that net cost of ownership to date.
- Purchase prices are approximate. Edit the buy price to match your specific build.
- Rental rates are the lowest on-demand price across all the cloud providers we track.
- Rental prices rise 3%/year from today's rate.
- Supported model size and prefill/decode speed estimates are indicative. See our companion article for speed test results on a range of hardware. Always test hardware before purchasing.
- Model size classes are based on those discussed in our companion article, in line with Artificial Analysis' open model size classes. In some cases, 4-bit quantisation is required.
- This calculator focuses on single-GPU setups aimed at single-user inference, with a total purchase cost below $20k. For high-demand multi-user inference, consider larger multi-GPU pods.