How Much Can You Earn Renting Your GPU to AI?
Updated May 2026 · Live marketplace data refreshed 1 minute ago · 95 GPUs tracked across Vast.ai, RunPod, io.net, Clore.ai
At current market rates, after the platform's fee. Power cost not yet subtracted — see the rig detail page for your personalized profit at your electricity rate.
The short answer
A modern datacenter GPU rents for $2–6/hour; a consumer RTX card for $0.20–0.80/hour. After the marketplace's cut (7% on RunPod, ~15% on Vast.ai, ~20% on io.net) and your electricity bill, a datacenter card netting $154.90/day at 100% uptime is normal right now. Consumer cards clear $5–25/day after costs. The big variable is uptime — rental marketplaces only pay while a renter is actively using your GPU, so real-world hosts average 60–85% utilization.
Top 8 earning GPUs on AI rental markets
Sorted by daily host profit at $0.10/kWh electricity. Click any row for live per-provider pricing, offer depth, and the profit calculator.
| GPU | Net profit / day |
|---|---|
| $152.02 | |
| $104.52 | |
| $78.45 | |
| $38.61 | |
| $43.40 | |
| $55.75 | |
| $19.56 | |
| $36.28 |
Where the host's money goes
The "host rate" you see on a marketplace is not what you earn. Four deductions between the renter's card and your bank:
RunPod takes ~7%, Vast.ai's internal surcharge is ~15%, io.net takes ~20%. This is deducted before the rate you see — we surface the host-net number directly so no manual math.
You're only paid while a renter is using the card. Hobby hosts average 60–75%; Enterprise targets 95%+. Our rig detail pages include an uptime-scenario slider so you can model this for your setup.
A full-tilt datacenter GPU (B200 at 1,000W) burns $2.40/day at $0.10/kWh, $4.80/day at $0.20/kWh. Residential electricity in most US states falls in that range; industrial rates are 2-4× cheaper.
Vast.ai pays out weekly in BTC/USD (min $20). RunPod uses Stripe Connect to USD (min $50). io.net pays in IO token. No fiat payout = you'll also pay an exchange spread when converting.
Which marketplace pays hosts the most?
Host-take-home percentages vary meaningfully. For the same raw rate, RunPod's Community Cloud hosts pocket the biggest slice. The three marketplaces below are the ones we scrape live host-net rates for; click through to start onboarding.
| Marketplace | Host keeps | Visit |
|---|---|---|
RunPod
|
93% | Sign up → |
Vast.ai
|
85% | Sign up → |
io.net
|
80% | Sign up → |
|
|
85% | Sign up → |
Not an endorsement — each marketplace has different renter demand, payout methods, and onboarding friction. We track 10+ P2P marketplaces total in the Rental Markets directory.
Real earnings examples
Three live snapshots from the top-earning GPUs we track. Numbers are host-net daily revenue at current marketplace rates, before electricity. Click any rig to see your personalized profit at your own electricity rate plus per-provider breakdown.
Setup checklist for first-time hosts
End-to-end, expect 1–3 hours from sign-up to your first paid rental session. The friction is mostly verification and bandwidth setup, not the GPU itself.
- Pick the marketplace. Datacenter card (A100/H100/B200) → Vast.ai or RunPod. Consumer GPU (RTX 4090, 4080, 3090) → Vast.ai's consumer tier or Salad. Crypto-payout believer → io.net.
- Create the host account. KYC ranges from email-only (Vast.ai) to Stripe Connect onboarding (RunPod, ~10 minutes with a passport).
- Install the host agent. One-line install on Linux (Ubuntu LTS recommended). Reboot once for Nvidia driver + nvidia-container-toolkit. Windows is supported on Vast.ai but Linux earns better.
- Open the right ports. 22 (SSH), 8080 (host agent), and a port range for renter SSH. CGNAT + dynamic IP works on Vast.ai via WireGuard tunnel; RunPod requires a static or port-forwardable address.
- Set your bid. Start at the marketplace median for your GPU and adjust weekly. Underprice by 5–10% on a brand-new listing to seed reviews; raise once you're rated.
- Watch utilization. First 2 weeks tell you whether your bid is viable. If utilization is <30%, drop the price. If it's >90% with a queue, raise it. Our rig detail pages show recent utilization across the marketplace so you have a benchmark.
When is AI rental worth it vs. mining?
For most modern GPUs, AI rental now pays more than mining does. Direct comparison: a Nvidia H100 SXM earns $40–60/day renting on AI markets; the best mining algorithm it can run (Autolykos2 for ERG, ETH-proxy chains) clears under $2/day. The inversion is recent — until 2023, mining was the dominant use case for consumer + datacenter cards. Post-Ethereum merge + AI training demand, rental took over. See GPU mining vs AI rental for a full breakdown.
When is AI rental worth it vs. mining?
For most modern GPUs, AI rental now pays more than mining does. Direct comparison: a Nvidia H100 SXM earns $40–60/day renting on AI markets; the best mining algorithm it can run (Autolykos2 for ERG, ETH-proxy chains) clears under $2/day. The inversion is recent — until 2023, mining was the dominant use case for consumer + datacenter cards. Post-Ethereum merge + AI training demand, rental took over. See GPU mining vs AI rental for a full breakdown.
FAQ
No. Vast.ai's consumer tier actively recruits RTX 3090, 4090, and even 3080 Ti hosts. Datacenter cards (A100, H100, B200) earn more absolute $/day but also cost 10-20× more upfront. The ROI-per-dollar-of-hardware is often better on consumer cards.
No — they're mutually exclusive use modes. When the marketplace client is running, it owns the GPU. Hosts typically pick one or dedicate separate cards. Our rig detail pages show "rental wins" or "mining wins" per GPU so you know which to pick.
Most marketplaces require symmetric 100+ Mbps with <30ms latency to major AWS/GCP regions. Data transfer in/out is capped per job but adds up — factor ~$5–20/month in bandwidth overage risk for high-usage hosts.
Generally no. Marketplaces require a desktop with a PCIe-slotted dGPU + dedicated cooling for sustained 100% TDP. Laptop GPUs throttle under sustained load and most renters refuse them. The exception is Salad Cloud — it accepts laptops because its workload is short-burst inference, not multi-hour training.
In 2026 the steady-state for a well-priced 4090 is 65–85% — meaning 15–35% of the time, the card is idle and earns nothing. Underpricing by 5–10% versus the marketplace median typically lifts utilization above 80% within a week. Live host-net rates per provider are tracked per GPU on the rig detail page.
Vast.ai pays out weekly in BTC or USD once you cross the $20 threshold. RunPod uses Stripe Connect with a 7-day rolling payout cycle (min $50). io.net pays in IO token per job — cleared on-chain within minutes. Onboarding KYC adds 1–2 days for fiat-paying providers.
Rental income is typically reported as self-employment or business income in most jurisdictions. USD payouts generate a 1099 in the US; crypto payouts (BTC via Vast.ai, IO token via io.net) have their own tax treatment. Consult a CPA — we don't give tax advice.