Run the model where the data lives.
GPU capacity on quote, a private inference endpoint inside your private cloud, and Sealed for the data no one may read.
Quoted per GPU-hourOperator named in the quote
On your own devices or ours, inside your private cloud's own network. Billed by the hour, no minimum term.
What you get
Four things to know.
- Private model hosting
- Inference on local data
- Training runs on quote
- Government and health data
GPU capacity, quoted in writing
The quote names the operator, the facility, the card class, the count, the window and the price per GPU-hour.
A private inference endpoint
Your models answer inside your private cloud. Prompts and data do not leave your network.
Sealed confidential inference
Data stays unreadable while in use, including to the cloud operator, and a receipt proves it.
CPU inference on the largest machines
The 32 and 64 vCPU sizes run smaller models without a GPU.
Price
What a GPU quote states
| Line | Stated |
|---|---|
| Operating entity and facility | Named before signature |
| Card class and count | Named |
| Availability window | Dated |
| Price | Per GPU-hour |
| Minimum term | Stated |
| Provisioning and hardware downtime | Not billed, written into the quote |
Data, model and answers stay on your devices and inside your private cloud.[1]
Not yet
- No self-serve GPU size in the console yet; capacity is quoted.
Where it runs
On our nodes or your own devices, inside your private cloud's own network. If a node fails, work moves across the network under the policy you set.
Get started
Start on the free tier today.
Or talk to Triple X about running it on your own devices.
- 01
Join the waitlist.
Tell us what you want to run and where. No credit card.
TakesTwo minutes
- 02
Call the endpoint.
Private inference from the console or the API. GPUs on quote.
TakesMinutes
- 03
Bring your own servers.
Triple X sets up your private cloud on them.
WhoTriple X
We never ask for your root credentials, API keys or bank details by email.
All productsAt a glance
The numbers a buyer asks for.
What a GPU quote states, what CPU inference costs, and where the data stays, each with where to check.
| What | Triple X Cloud | Where to verify |
|---|---|---|
| GPU price | Per GPU-hour, on quote | Your written quote |
| Operator and facility | Named before signature | Your written quote |
| Minimum term | Stated in the quote | Your written quote |
| Downtime | Not billed; written into the quote | Your written quote |
| CPU inference | 32 vCPU $0.384; 64 vCPU $0.768 an hour | Rate card, August 1, 2026 |
| Private endpoint | Inside your private cloud | Docs and the console |
| Sealed | Unreadable in use; a receipt per run | Sealed page and a sample receipt |
| Self-serve GPU | Not yet; capacity is quoted | This page, October 2026 |
| Exit cost | Transfer from $0.01 per GB | Pricing page and your contract |
Questions
What buyers ask before they sign.
In the order they come up. Every answer ends with where to check it.
10 answers
All questionsCan I run my own model privately?
Yes. A private inference endpoint answers inside your private cloud, so prompts and data do not leave your network. Smaller models run on CPU on the 32 and 64 vCPU sizes without a GPU. Larger models need quoted GPU capacity.
Proofthe inference endpoint description on this page and the virtual machine size list.[1]
What does a GPU cost?
On quote. No public GPU rate is published. The quote names the operator, the facility, the card class, the count, the availability window, the price per GPU-hour and the minimum term. Provisioning time and hardware downtime are not billed.
Proofa written GPU quote from Triple X and the GPU note on the Pricing page.
What does CPU inference cost?
The published rate card: a 32 vCPU machine is $0.384 an hour and a 64 vCPU machine is $0.768. Hourly, no minimum term. Stop the machine and the meter stops. Outbound transfer is from $0.01 per GB.
Proofthe published virtual machine rate card, August 1, 2026.[2]
What does it cost to leave?
Outbound transfer from $0.01 per GB for your model weights and data, and nothing else, unless a GPU quote carries a minimum term, which the quote states. Models and data are standard files you can copy out.
Proofthe transfer bands in the published pricing and the minimum term line in your GPU quote.[2]
Who can see my prompts, data and weights?
Inside your private cloud, you. The platform underneath sees that a workload exists and its size, not its contents. For data no one may read, Sealed keeps it unreadable while in use, including to the operator, and a receipt proves the run.
Proofthe access table on the Security page and the Sealed page.
Can I run inference on GPUs I already own?
Yes. Your GPU servers join your private cloud, and the platform schedules, networks and bills the work on them. Triple X quotes the platform charge in writing. Hosted GPU capacity comes from a named operator and facility.
Proofthe Sovereign page and a written quote from Triple X naming the operator and facility.
Where does a hosted GPU job run?
In the facility named in the quote, in the region you choose, inside your private cloud. Data, model and answers stay on your devices and inside your private cloud. Nothing is placed before the quote is signed.
Proofthe operating entity and facility line in your GPU quote.
What happens when a GPU node fails?
Hardware downtime is not billed, and that is written into the quote. Work restarts under the policy you set, inside your private cloud first. Checkpoint training runs to object storage so a restart loses minutes, not days.
Proofthe downtime line in your GPU quote and the failover rules on the Reliability page.
What proof exists for confidential inference?
The platform holds a SOC 2 Type II attestation with an unqualified opinion, available under NDA. A Sealed run produces a receipt you can check. Ask Triple X for a sample receipt and the attestation report before you commit regulated data.
Proofthe SOC 2 Type II report under NDA and a sample Sealed receipt from Triple X.[3]
How do I start, and what is not included?
Start CPU inference on the free tier today, then ask Triple X for a GPU quote in writing. There is no self-serve GPU size in the console yet. Bring your model files and a container image that serves them.
Proofthe Not yet list on this page, checked October 2026.
No questions under this term yet. Ask Triple X.