Confidential Computing

Overview

Data has traditionally been protected in two states: at rest, when stored, and in transit, when moving across a network. The third state, in use, while actively being computed on, was the hardest to secure, because the processor needs the data in readable form to work on it. Confidential computing closes that gap, using hardware to keep data encrypted and isolated even during processing, shielded from the operating system, the hypervisor, and the infrastructure operator alike.

How it works

It relies on a trusted execution environment, a hardware-enforced secure enclave in which code runs and data is decrypted only inside protected memory the rest of the system cannot read. On the latest GPUs this protection extends to the accelerator, so that model weights and the data passing through them remain encrypted in GPU memory and verified against tampering, with cryptographic attestation proving the environment is genuine.

Why it matters

For the most sensitive workloads, sovereignty over where data lives can be reinforced by guarantees over who can see it while it is in use. Confidential computing means that even the party running the infrastructure cannot inspect a customer's data or models during processing. Combined with in-country residency, it offers regulated and security-critical organizations an unusually strong assurance over their most valuable assets.