AIToolVerdict
中文

Hourly cloud compute and per-GPU hourly GPU rental — from $2.50/month for 1 vCPU and 0.5 GB of RAM, with the first tier that has usable memory at $6, up to 192 vCPU and an NVIDIA A16 at $0.471 per GPU per hour.

Rating
4.2 4.2 out of 5
Starting price $2.50 per month; each plan is listed with both an hourly and a monthly price
Free trial No
View deal · Vultr Read review

Last updated:

Pros

  • Cloud GPU is billed per GPU per hour, so a few hours of experimentation has a small and knowable price — no card to buy, power or cool
  • Every plan lists both an hourly and a monthly price ($0.004/hour on the $2.50 tier, $0.009 on the $6 tier, $0.045 on the $30 tier, $0.112 on the $75 tier), so you can validate hourly before committing to a month
  • Five compute lines on one account, from $2.50/month to 192 vCPU and 1536 GB of RAM
  • The $6 High Performance tier is the cheapest plan with usable RAM (1 GB, 25 GB NVMe, 2 TB of bandwidth) and the right application server to put beside a rented GPU
  • Dedicated vCPU starts at the $30 General Purpose tier (4 GB RAM, 30 GB NVMe, 4 TB of bandwidth), so the shared-CPU lines are a choice rather than the only option
  • The $75 Storage Optimized tier trades CPU for disk: 150 GB of NVMe at the same 4 TB of bandwidth, five times the General Purpose tier
  • Kubernetes Engine has a free control plane, so you pay only for worker nodes, and managed MySQL and PostgreSQL, load balancers, Block Storage and S3-compatible Object Storage sit in the same account
  • Large public API and documentation surface, which is what makes the platform automatable

Cons

  • Automatic backups are not included and add 20% to the base fee, which turns the $2.50 instance into $3.00/month
  • Snapshot storage is billed separately at $0.05/GB/month: 100 GB of snapshots costs $5.00/month, twice the price of the instance being snapshotted
  • DDoS protection is a $10/month add-on for 10 Gbps of mitigation, and a Windows licence is not included in the base price
  • The cheapest GPU line, the A16 at $0.471 per GPU per hour, is positioned for virtual desktops and light graphics rather than model inference, which sits on the GH200 at $1.990 and the PCIe A100 at $2.397
  • The $2.50 tier's 0.5 GB of RAM is below what an application server needs, pricing varies by data centre location, and patching, firewalling and upgrades are yours

Score breakdown

Features 5.0
Ease of use 4.0
Value for money 4.0
Support 4.0

Pricing

Cloud Compute — Regular Performance

$2.50 per month · $0.004/hour

  • 1 vCPU, 0.5 GB RAM, 10 GB storage and 0.5 TB of bandwidth
  • Previous-generation Intel with shared vCPU — the cheapest way onto the platform
  • 0.5 GB of RAM is below what an application server needs, so it fits a DNS resolver, a tunnel endpoint or an SSH jump box and little else

Cloud Compute — High Performance

$6.00 per month · $0.009/hour

  • 1 vCPU, 1 GB RAM, 25 GB NVMe storage and 2 TB of bandwidth
  • AMD EPYC or Intel Xeon with shared vCPU and NVMe storage
  • The cheapest tier with enough RAM to sit in front of a rented GPU as the application server

Optimized Cloud Compute — General Purpose

$30.00 per month · $0.045/hour

  • 1 vCPU, 4 GB RAM, 30 GB NVMe storage and 4 TB of bandwidth
  • Dedicated vCPU — the processor is not shared with other customers
  • The first tier that holds an application, a small database and a worker at the same time

Optimized Cloud Compute — Storage Optimized

$75.00 per month · $0.112/hour

  • 1 vCPU, 8 GB RAM, 150 GB NVMe storage and 4 TB of bandwidth
  • Five times the storage of the General Purpose tier at the same 4 TB of bandwidth
  • For datasets and busy databases where disk capacity, not CPU, sets the ceiling

Who it is for

  • Work that needs GPU compute for a few hours rather than a permanent GPU server — per-GPU hourly billing keeps an experiment at a few dollars
  • Anyone running model inference: Vultr positions the GH200 at $1.990 and the PCIe A100 at $2.397 for it, not the cheapest A16
  • Applications that need a real application server — the $6 tier is the cheapest plan with usable RAM and the machine to pair with a rented GPU
  • Workloads that need dedicated vCPU, which starts at the $30 General Purpose tier with 4 GB of RAM, 30 GB of NVMe and 4 TB of bandwidth
  • Storage-heavy services — the $75 Storage Optimized tier pairs 150 GB of NVMe with 8 GB of RAM
  • Teams wiring cloud resources into an existing pipeline: large public API and docs, plus a Kubernetes Engine control plane that costs nothing
  • Work that needs a large machine briefly: VX1 reaches 192 vCPU, 768 GB and 10 TB at $0.060 to $5.760 per hour

Alternatives

InterServer

Top pick

The cheapest sustainable renewal rate in shared hosting — $7/month after a $2.50 first month, billed month to month with no multi-year prepay, and unlimited storage, transfer, email and FTP included.

4.2 4.2 out of 5
Starting price: $2.50 first month, then $7/month

Best for: Anyone who has been burned by a renewal invoice and now compares the post-discount rate first: $7/month against SiteGround's $17.99, about $220 saved over three years

Read review View deal · InterServer

VPS.org

Root-access cloud VPS from $2.50/month, billed hourly with a monthly cap — 11 plans from 512 MB to 96 GB, and upgrades keep your data and IP.

3.8 3.8 out of 5
Starting price: $2.50 per month, billed hourly and capped at 672 hours

Best for: Developers who want root access and are willing to run their own server

Read review View deal · VPS.org

BandwagonHost

KiwiVM KVM VPS from $49.99/year, and the CN2 GIA/CTGNet China-optimised line from $49.99/month — with free datacenter migration without data loss and a claimable SLA.

4.3 4.3 out of 5
Starting price: $49.99 per year (20G KVM PROMO VPS); the China-optimised Osaka CN2 GIA line starts at $49.99 per month

Best for: Buyers whose audience or team is in mainland China and who need a stable return route rather than the cheapest possible box — this is the host the CN2 GIA premium exists for

Read review View deal · BandwagonHost