Confidential Cloud Is Now in Public Beta
Published
Today we're launching Confidential Cloud in public Beta.
You know the feeling. You press send on a prompt, and you wonder where it went. And who can read it. Everything you typed about yourself, your customers, and your company now sits on someone else's server. Should you have sent it?
For companies, the stakes are higher. Their data is their value: source code, customer records, research, model weights. Companies can't afford to hand that over. So the most sensitive work either misses out on AI, or proceeds without the protection it deserves.
Confidential Cloud solves this.
It's the cloud for AI where your data stays protected from end to end: encrypted at rest, in transit, in CPU memory, and protected in GPU memory. Protected by the hardware itself, not by a promise.
The infrastructure provider can't read your data. The data center can't read it. We can't read it.
And don't take our word for it. Every Confidential Cloud instance, every service returns a verifiable attestation report that covers its hardware and software, signed in hardware by Intel, AMD, or NVIDIA. Verify the report before you submit any data.
How it's different
Every cloud says it won't look at your data. But that's only a contract. A promise.
Confidential Cloud is different. It runs entirely on Trusted Execution Environments (TEEs) and GPUs in confidential computing mode: hardware that protects a machine's memory from the host and reports exactly what software runs inside. Zero data retention (ZDR) and zero operator access (ZOA) become provable, not promised.
Until now, private AI meant your own hardware, on premises. No longer. We built the stack that makes the cloud private, too: from firmware and measured images to confidential GPUs and verifiable attestation. And we made it work at scale. Confidential Cloud is the fastest, simplest way to use AI privately and securely.
What's in the Beta
- Confidential GPUs. Confidential NVIDIA B300 GPUs, up to eight per Confidential VM.
- Confidential VMs. CPU VMs with Intel TDX and AMD SEV-SNP. On demand, billed by the hour.
- Fully attested. Verify the hardware and software before you use it.
- Built for agents. An MCP server exposes the whole cloud as tools. Your agents can launch and verify their own machines.
All available through one console and one API.
How it works
1. sign up
2. add a card
3. launch a confidential VM
4. verify the attestation report
5. build
What you can build
- Private inference. Prompts and completions that no one else can read.
- Private agents. Agents hold credentials, memory, and customer data. Give them a cloud that keeps data sealed.
- Private model weights. Serve a model without ever revealing the weights.
- Private training and fine-tuning. Use the data you couldn't move to a cloud before.
What's next
- Confidential inference. Shared capacity and dedicated.
- Persistent block storage. Restart a VM and keep your data.
- More GPUs. New B300 capacity.
- Managed Kubernetes. Confidential clusters without the setup.
- Version 1.0. Coming soon.
AI should be private by default. Introducing Confidential Cloud.
Create an account and launch your first confidential VM today.