+ Add Provider
DohoHub / News / GPU hosting for AI, by price: from $21 to 2x RTX 5090
Pricing Sep 19, 2026 · 16:26 UTC

GPU hosting for AI, by price: from $21 to 2x RTX 5090

GPU hosting for AI, by price: from $21 to 2x RTX 5090

The full GPU ladder from our database: $21 VPS to dual RTX 5090 at $604.45/mo. K80 24 GB vRAM at $64.45 is the best vRAM-per-dollar we track — $2.69 per GB.

**What does it actually cost to rent a GPU server for AI work in September 2026? We pulled every active GPU plan from our database — from a $21 VPS to a dual-RTX 5090 box — so the price ladder is visible end to end.

The metric that matters for AI is vRAM, not cores: your model has to fit in memory first. Prices below are per month, as captured from provider sites on September 19, 2026.

Entry level: small models, inference, dev

  • GPU Mart — Express GPU VPS, 2 GB vRAM: $21/mo
  • GPU Mart — Dedicated P1000, 4 GB vRAM: $37/mo — dropped 42% last week, we caught it in our price history
  • Vultr — A16, 2 GB vRAM (fractional): $43/mo
  • GPU Mart — Dedicated P600, 2 GB vRAM: $49/mo
  • Vultr — A40, 2 GB vRAM (fractional): $55/mo

The sweet spot: 24 GB vRAM under $80

  • GPU Mart — Dedicated K80, 24 GB vRAM: $64.45/mo — $2.69 per GB, the best vRAM-per-dollar in our database (dropped 41%, logged here)
  • Servarica — Bee GPU VPS, 24 GB vRAM: $79/mo — $3.29 per GB

A K80 is old silicon — no tensor cores, slower than anything modern. But for batch jobs that run overnight, 24 GB of vRAM at $64 is a tool few people price-check. Two providers moved this tier the same day in September.

Modern cards: RTX 40/50 series

  • GPU Mart — RTX 5060 VPS, 8 GB vRAM: $85/mo
  • GPU Mart — RTX A4000 VPS, 16 GB vRAM: $119/mo — $7.44 per GB
  • GPU Mart — RTX 4060 dedicated, 8 GB: $149/mo
  • Hetzner — GEX44, 20 GB vRAM (L40S-class): €232.30/mo
  • GPU Mart — RTX A5000 dedicated, 24 GB: $269/mo
  • GPU Mart — RTX 5090 VPS, 32 GB vRAM: $291.85/mo — the cheapest 32 GB card we track
  • OVHcloud — RISE-GPU-1, 12 GB: C$207.99/mo

The big box

  • Database Mart / GPU Mart — Enterprise multi-GPU, 2x RTX 5090, 64 GB vRAM total: $604.45/mo — dropped 30% on September 11, full story here. At −$254.55/mo versus the old price, that is over $3,000 a year saved on one node.

Why we are running this one

GPU pricing is volatile — the same plan moved 30–42% in the last two weeks alone. Every number above comes from our price monitor, not from a provider's marketing page. Live list: dohohub.com/deals

#gpu #ai #vram #pricing #inference
← All news