What is confidential computing? Protecting data while it runs

The three states of data

Data exists in three states, and each has traditionally needed its own protection. Data at rest — sitting in storage — is commonly protected by encryption, as is data in transit — moving across a network; our explainer on what encryption in the cloud is covers both in detail. The third state, data in use, is the data actively being processed in a system’s memory while a computation runs on it, and historically it has been the exposed one: to process data, a system has generally had to decrypt it first, which means it sits unprotected in memory for the duration of that processing, visible in principle to anything with sufficient privilege on that system.

Confidential computing closes that gap. It is specifically about protecting the in-use state — keeping data encrypted and isolated even while active computation is happening on it, rather than only before and after.

How it works, conceptually

The mechanism is a trusted execution environment, sometimes called a secure enclave — a hardware-isolated area of a processor where data is decrypted and processed in a way that is protected from everything outside that isolated area, including privileged software on the same machine, the host operating system, and the cloud provider’s own administrative access.

The practical effect is that a workload can run inside this enclave and produce results without the surrounding infrastructure — up to and including the provider running the hardware — being able to read the data it is working on. This is a hardware guarantee rather than a purely software or policy one, which is what distinguishes it from ordinary access controls and permissions.

Why it matters

Confidential computing lets organisations process highly sensitive data in the cloud with stronger guarantees than encryption at rest and in transit alone can provide — closing the one gap in the data lifecycle where information was previously exposed by necessity. That matters for data that is regulated, commercially sensitive, or simply too risky to expose even briefly during processing.

It also enables genuinely new scenarios: multiple parties can compute jointly on a shared dataset — for example combining data from separate organisations to run an analysis or a machine-learning process — without any party exposing its own raw inputs to the others or to the platform coordinating the computation. That kind of multi-party computation was previously difficult to trust without confidential computing’s hardware guarantees.

Why it is a growing 2026 trend

Three pressures are pushing confidential computing into mainstream use. Rising privacy regulation increases the cost of any data exposure, including the momentary exposure that ordinary processing has always involved. The growth of sensitive AI and machine-learning workloads, often trained or run on data that is commercially or personally sensitive, raises the stakes of processing that data anywhere outside the owner’s direct control. And multi-party use cases — organisations wanting to collaborate on data without fully trusting each other or a shared platform — are becoming more common as data collaboration itself becomes more valuable.

How it relates to sovereign cloud and zero trust

Confidential computing complements rather than replaces these adjacent ideas. Sovereign cloud (see our explainer on what sovereign cloud is) is about legal jurisdiction and control over where data lives and who operates the infrastructure; confidential computing is about protecting data during processing regardless of who operates the hardware, which makes the two genuinely complementary layers rather than overlapping ones — a sovereign deployment with confidential computing adds a technical guarantee on top of a legal one.

It also fits naturally into a zero trust posture (see our explainer on what zero trust security is), which starts from the assumption that no part of the surrounding infrastructure should be implicitly trusted. Confidential computing extends that assumption to the processing layer itself, rather than trusting the host system by default once other access controls are satisfied.

The honest note, and where it fits careers

Confidential computing is a powerful additional layer, not a silver bullet — it protects data during processing specifically, and it adds real complexity: workloads often need adapting to run inside an enclave, and the guarantees depend on trusting the specific hardware implementation. It solves the in-use exposure gap; it does not replace the need for sound access control, encryption at rest and in transit, or good identity management elsewhere in the system.

The major cloud providers offer confidential-computing options as part of their broader security and compute catalogues, generally as an alternative configuration for compute resources rather than a separate product line — the specifics are best checked against current provider documentation. For anyone building a security-focused cloud career, confidential computing sits alongside encryption, identity and zero trust as one more concept worth understanding conceptually, and it connects directly to the data-protection material covered in security-focused certification study.