Overview / Pricing

Pricing

Every configuration has a published price per GPU-hour. Reserve capacity for a month or a year and the discount is shown beside it — there is no quote-only tier for standard hardware.

Sample ratesPrices on this page are illustrative and will be replaced by the published rate card before launch. They are not offers.

Compute

Per-second billing, one-minute minimum. Stop any time.

Compute prices by configuration
AcceleratorPlanGPUsPer GPU-hourPer node-hourPer node-month
B300288 GB HBM3ecl.b300.x88$6.20$49.60$36,208
B200180 GB HBM3ecl.b200.x88$4.90$39.20$28,616
H200141 GB HBM3ecl.h200.x88$3.10$24.80$18,104
H10080 GB HBM3cl.h100.x88$2.40$19.20$14,016
MI300X192 GB HBM3cl.mi300x.x88$2.10$16.80$12,264
RTX PRO 600096 GB GDDR7cl.rtx6000.x88$1.45$11.60$8,468
cl.rtx6000.x22$1.55$3.10$2,263
L40S48 GB GDDR6cl.l40s.x44$0.95$3.80$2,774
cl.l40s.x11$1.05$1.05$767

Per node-month assumes 730 hours. Storage, network add-ons and taxes are extra.

Estimate a month

Choose a configuration and how you'll use it. The estimate updates as you go.

ESTIMATESample rates

Compute
Reservation discount
Confidential mode
Block storage
Per month

Request this capacity

Add-ons

Priced separately so the compute price stays comparable.

Confidential mode
+8% of compute
Measured boot, GPU confidential-computing mode and per-boot attestation reports. Supported GPUs only.
Block storage
$0.08 per GB-month
NVMe-backed volumes, encrypted with keys scoped to your project.
Object storage
$0.018 per GB-month
S3-compatible buckets in the same facility. No egress fee to your own nodes.
Private interconnect
$650 per port-month
Dedicated 10/100 Gb/s cross-connect to your own network or cage.
Dedicated IPv4
$3 per address-month
IPv6 /56 per project is included.

Support

Every account reaches an infrastructure engineer. The tiers change how fast, and how closely we work with you.

Support tiers
TierPriceFirst responseIncludes
StandardIncludedNext business day
  • Email and ticket support
  • Public status and incident reports
  • Hardware replacement
Priority8% of spend · $1,000 minimum1 hour for severity 1
  • Shared Slack or Teams channel
  • Named infrastructure engineer
  • Quarterly capacity planning
DedicatedCustom15 minutes for severity 1
  • On-call engineer for your enclosure
  • Change windows agreed with you
  • Joint incident review

Billing questions

How is on-demand usage billed?

Per second from the moment a node is handed to you until you stop it, with a one-minute minimum. Stopped nodes are not billed; volumes are, until you delete them.

What does a reservation guarantee?

That the configuration is held for you for the whole term, in the region you chose. You can run it or not — the capacity is yours either way, and it is billed monthly.

Can I change a reservation?

You can move a reservation to a newer accelerator generation at any time, crediting the remaining value. Shortening a term is not possible.

Is there a charge for data transfer?

Traffic between your own nodes, volumes and buckets in one facility is free. Internet egress is billed per GB above a monthly allowance that scales with your compute spend.

How do I pay?

By card for on-demand usage. Reservations and enclosures are invoiced, with net-30 terms after a credit check.