Vultr GPU pricing changes and market comparisons

Vultr GPU pricing rates

Vultr sells GH200.

The rate is $1.99.

This price remains the lowest in the current market comparison. DeployBase tracks the 8x HGX H100 80GB bare metal server at $23.92 per hour. This total equals $2.99 per GPU-hour. ComputePrices lists a different H100 SXM rate at $2.30 per GPU-hour. Which figure represents the true cost for an H100? Vultr also sells A100 80GB PCIe instances at $2.397 per hour. This configuration includes 24 vCPUs and 80GB of system RAM. A 1-year prepaid term for an A100 PCIe instance at $2.397 per hour drops the cost to $1.80 per hour, which creates an estimated $5,230 in annual savings at full-time utilization. The A100 SXM rate is $2.60 per hour. This premium covers NVLink interconnect and higher inter-GPU bandwidth. The lineup also includes A40 at $1.86 per hour and A16 at $0.51 per hour. Vultr provides several A16 configurations, including 4-GPU setups at $0.512 per hour and 16-GPU setups at $0.574 per hour. As of December 2025, preemptible AMD MI325X and MI355X plans exist. Users can also secure up to 35% savings on dedicated access with 3-year GPU committed-use contracts.

Market comparisons

Spheron undercuts Vultr on A100 prices.

The Spheron on-demand PCIe rate is $1.82 per hour.

Vultr’s A100 PCIe rate is $2.397 per hour. Vultr wins on GH200. The Vultr GH200 rate is $1.99 per hour while Spheron lists it at $3.02 per hour. Spheron provides H100 spot instances at $2.21 per hour for PCIe. This price beats Vultr’s bare-metal per-GPU rate. Vultr lacks a published spot tier for H100. Vultr also lacks a PCIe H100 SKU. The Vultr H100 bare metal node edges out Spheron by 12% on on-demand rates. Spheron’s H100 SXM spot rate is $2.91 per hour. Spheron’s H100 on-demand node rate is $3.38 per hour, while the Vultr node is $2.99 per hour. You can check live rates.

Service details

Vultr bundles compute, storage, and interconnect.

This bundling applies to the A100, H100, and GH200 tiers. Vultr provides the Talon service for fractional A100 access. This service starts at $90 per month. It delivers 1/20th of the compute power of a full A100. It provides 4 GB of GPU memory. The Talon service uses Nvidia MIG technology for instances with 10 GB of memory or more. Nvidia vGPU software uses time slicing to share compute resources between virtual machines. Vultr operates in 32 global data center locations. These locations span the US, Latin America, EMEA, and APAC. Users pay $0.01 per GB for bandwidth overages after they use their 2 TB/month free egress allowance. Vultr uses a 730-hour monthly billing model for most products. The Talon service remains in beta and stays limited to the New Jersey location. This service was developed with Nvidia participation to help with machine learning inference.