Cloud GPU Pricing 2026: Confidential VM prices for H100, H200 and RTX PRO 6000 | VoltageGPU

Live VoltageGPU pricing for Confidential VMs with an NVIDIA H100, H200 or RTX PRO 6000 Blackwell GPU, side-by-side with AWS, Google Cloud and Microsoft Azure. Every VM is an Intel TDX trust domain with NVIDIA GPU attestation you generate and verify yourself, per-second billing, no commitment. Your first top-up of $10 or more adds one hour on the RTX PRO 6000 Blackwell VM.

Cloud GPU Pricing Comparison 2026, VoltageGPU vs AWS vs GCP vs Azure

Comparison prices are public list prices from each provider’s pricing page (April and September 2026). VoltageGPU prices are read live from the provider inventory and update on this page.

Two ways to use VoltageGPU

• Confidential VM (Intel TDX, attestation generated by the tenant): RTX PRO 6000 Blackwell $3.80/hour, H100 $6.95/hour, single-GPU H200 $8.08/hour, each with the Intel TDX quote and the NVIDIA GPU attestation verified from inside the VM; 8x H100 node $55.62/hour with NVIDIA attestation of all eight GPUs (multi-GPU Protected PCIe mode, verified 10 September 2026). Self-service, boots in about two and a half minutes, one hour upfront then per second, unused time refunded on release.
• Confidential AI inference API, OpenAI-compatible, every model in a TEE, billed per token from the same balance.

Also available: Standard GPUs with no enclave, for data that is not sensitive, from a distributed provider network at /standard-gpus, typically from under $0.50 per GPU-hour on the smallest cards. Confidential AI agents for regulated teams have their own page at /agents/confidential. An AI coding agent can rent these machines itself, at these prices, with a limited key and a daily budget: see the agent quickstart.

Dedicated bare metal (no enclave, not confidential)

A whole 8x H100 SXM machine with 192 vCPU and a full NVLink mesh, root over SSH, no hypervisor, nobody else on it. Not confidential: no Intel TDX, no attestation; you check the hardware yourself (hwcheck script, systemd-detect-virt, nvidia-smi topo -m). One hour charged at start, then per second, with a mandatory automatic release (24 hours by default, 72 at most). Details at /dedicated.

• 8x H100 SXM (bare metal, 8x H100, 192 vCPU, no enclave, not confidential): $49.80 per hour for the whole machine, $6.23 per GPU-hour, billed per second with a mandatory automatic release of 72 hours at most (3 available now).

Sandboxes (no enclave, not confidential)

Small Ubuntu 24.04 or 26.04 boxes (1, 2 or 4 vCPU, 2 GB RAM, 10 GB disk, CPU only, SSH optional) created in seconds and billed per second. Idle, a sandbox pauses on its own after 300 seconds by default and bills its paused price until you resume it in seconds or delete it. Details at /sandboxes.

• Small sandbox (1 vCPU, 2 GB RAM, 10 GB disk, ubuntu-24.04 or ubuntu-26.04, no enclave, not confidential): $0.09 per hour running, $0.004 per hour paused, billed per second (2 available now).
• Medium sandbox (2 vCPU, 4 GB RAM, 20 GB disk, ubuntu-24.04 or ubuntu-26.04, no enclave, not confidential): $0.18 per hour running, $0.008 per hour paused, billed per second (2 available now).
• Large sandbox (4 vCPU, 8 GB RAM, 20 GB disk, ubuntu-24.04 or ubuntu-26.04, no enclave, not confidential): $0.35 per hour running, $0.008 per hour paused, billed per second (2 available now).

Why VoltageGPU

• Per-second billing, pay for the exact second the machine runs, not for whole hours
• No reserved-instance lock-in, no 1-year or 3-year commitments to unlock the listed price
• Your first top-up of $10 or more adds $3.80, one hour on the RTX PRO 6000 Blackwell Confidential VM
• Same Intel TDX confidential computing technology as Azure and Google Confidential VMs, with an open-source verifier and public dated evidence
• Bitcoin accepted alongside card payments

Cheapest Cloud GPU for LLM Inference 2026

• Open models running in TEEs (Qwen, DeepSeek, Kimi, GLM, Gemma, Mistral), priced per million input and output tokens: live prices on /models and from GET /v1/models
• OpenAI-compatible API at api.voltagegpu.com/v1, drop-in for OpenAI SDK, LangChain, LlamaIndex

Which Confidential VM to pick

• Development, attestation tests and 7B–70B inference: RTX PRO 6000 Blackwell, 96 GB, $3.80/hour
• LoRA / QLoRA fine-tuning and heavier inference: H100 80GB, $6.95/hour
• Models that need the memory: H200 141GB, $8.08/hour
• Multi-GPU training: the 8x H100 node, attested per GPU in Protected PCIe mode
• Training data and gradients stay encrypted in memory inside the Intel TDX trust domain

Frequently Asked Questions, Cloud GPU Pricing

What is the cheapest confidential GPU per hour in 2026?

On VoltageGPU, the RTX PRO 6000 Blackwell Confidential VM at $3.80/hour, and an H100 Confidential VM at $6.95/hour. AWS p5 ($4.30/hr) and Google a3 ($3.67/hr) are commodity instances without a trusted execution environment, so they are cheaper per hour but not comparable; Azure’s confidential NCC40ads H100 v5 lists at $8.90/hr (12 September 2026).

How do VoltageGPU prices compare to AWS, GCP, and Azure?

The only hyperscaler SKU with a GPU inside a trust boundary and a public price is Azure’s NCC H100 v5 (AMD SEV-SNP plus one H100 NVL, $8.90/hr list on 12 September 2026); Google Cloud sells confidential H100 on A3 High with Intel TDX at per-zone rates, and AWS has no confidential GPU. Dated, sourced comparison: four clouds compared.

Is there a free trial?

Signing up is free. Using a machine needs a prepaid balance, and your first top-up of $10 or more adds $3.80 automatically, one hour on the RTX PRO 6000 Blackwell Confidential VM, once per account.

Is there a minimum commitment or reserved-instance discount?

No. The prices on this page are the price you pay, with no contracts and no reserved instances. Per-second billing means a five-minute experiment costs five minutes. Minimum top-up is $5.

How does VoltageGPU billing work?

The first hour is charged when the VM starts, then billing is per second; the unused part is refunded when you release the machine.