Skip to content
Web Development13 min read

Confidential Computing for Business Applications: Protecting Data While It Is Being Processed

What confidential computing is, how trusted execution environments and attestation protect data in use, real business use cases, limits, and how to adopt it.

01

The short answer

Confidential computing protects data while it is being processed. Workloads run inside a hardware-isolated trusted execution environment (TEE) with encrypted memory, so the host operating system, hypervisor and cloud operator can't read the data. Attestation lets you verify exactly what code and configuration are running before you release keys or data to it. It complements encryption at rest and in transit, and it is worth considering when sensitive data must be processed on infrastructure you don't fully control, or shared between parties who don't fully trust each other.

This guide explains how it works, where it helps a business, its limits, and how to adopt it without rebuilding everything.

02

Three states of data

StateTypical protectionWhat it guards against
At rest (stored on disk or in a database)Disk, database and backup encryption with managed keysStolen disks, leaked backups, unauthorised storage access
In transit (moving across networks)TLS and network encryptionInterception between services and users
In use (in memory while processed)Confidential computing: hardware-isolated, memory-encrypted TEEs with attestationA compromised or curious host OS, hypervisor, cloud administrator or co-located workload
03

How it works

  • Trusted execution environment. The processor isolates a workload, either a whole virtual machine (such as AMD SEV-SNP, Intel TDX or Arm CCA realms) or a smaller enclave within an application (such as Intel SGX), and encrypts its memory with keys held in the hardware.
  • Attestation. The hardware produces signed evidence of what is running: the platform, firmware and a measurement of the workload. A verifier checks it against expected values. The IETF's RATS architecture (RFC 9334) describes the roles involved: the attester, the verifier and the relying party.
  • Key release. A key management service releases decryption keys only to a workload whose attestation matches policy. This is what ties the protection together: data stays encrypted until it reaches verified code.
  • Confidential GPUs. Recent data-centre GPUs, starting with NVIDIA's H100, can run in a confidential mode alongside a CPU TEE, which extends protection to AI training and inference.
04

Options on the main clouds

Checked in November 2026. Availability varies by region and machine type, so confirm the current status before designing around a specific option.

ProviderMain options
AWSNitro Enclaves: isolated environments with no persistent storage, interactive access or external networking, whose attestation documents can be checked by AWS KMS before keys are released
Microsoft AzureConfidential VMs on AMD SEV-SNP and Intel TDX, confidential GPU VMs with H100, confidential containers on Azure Container Instances, confidential AKS nodes, Azure Attestation and Managed HSM; some container and clean-room features are in preview
Google CloudConfidential VMs on AMD SEV, SEV-SNP and Intel TDX, with confidential GPU options, and Confidential Space for collaboration between parties who don't trust each other
05

Business use cases

Use caseWhy confidential computing helps
Processing highly sensitive data in the cloudFinancial, health or identity data can be processed with the cloud operator technically excluded, which can help with risk assessments and customer due diligence
Multi-party data collaborationSeveral organisations (for example banks checking for fraud, or a retailer and a brand measuring campaign results) combine data for an agreed analysis without any party, or the operator, seeing the others' raw data
Key and secret protectionSigning, decryption and tokenisation run inside attested environments so keys are never exposed to the host; see secrets management
Confidential AI inferencePrompts and data sent to a model are processed in a protected environment, reducing exposure to the model host. Apple's Private Cloud Compute, announced in 2024, is a prominent example of this design.
Protecting model weights and IPProprietary models or algorithms run on customer or partner infrastructure without exposing the code or weights
06

Limits and risks

Confidential computing is a strong control for a specific threat: someone with access to the infrastructure reading data in memory. It is not a general security fix.

  • Application vulnerabilities still matter. A bug or weak access control inside the protected workload can still leak data.
  • Side channels. Researchers have published attacks against several TEE technologies over the years, including Foreshadow against Intel SGX and CacheWarp against AMD SEV; vendors generally respond with microcode and firmware updates, so patching and attestation of firmware versions matter.
  • Attestation must actually be checked. Without a verifier and key-release policy, a TEE gives much less assurance.
  • Operational complexity. Debugging, monitoring and logging are harder when operators deliberately can't see inside the workload.
  • Performance and cost. Expect some overhead and a narrower choice of machine types and regions.
  • Trust moves, it doesn't disappear. You still trust the hardware vendor and its attestation infrastructure.
07

How to adopt it

StepWhat to do
1. Define the threatWho must not see the data in use: the cloud operator, your own administrators, a partner? If the answer is nobody beyond your normal controls, you may not need it
2. Choose the isolation levelConfidential VMs or containers need few or no code changes; enclaves give a smaller trusted boundary but need more application work
3. Design attestation and key releaseDefine the expected measurements, the verifier and the policy under which keys are released
4. Separate sensitive processingMove only the sensitive part (for example matching, scoring or decryption) into the protected environment, behind a clear API
5. Plan operationsLogging that doesn't leak sensitive data, monitoring, patching of firmware and images, and incident response
6. MeasurePerformance and cost against a normal deployment, using your own workload

Handling highly sensitive data?

ZSpace Labs designs secure application architectures for regulated data, including confidential computing where it is the right control, attestation-based key release and secure multi-party data workflows.

Start a Project
08

When you probably don't need it

For many business applications, strong access control, encryption at rest and in transit, good key management, audit logging and a trusted cloud provider already address the realistic risks. Confidential computing earns its cost when the operator itself is part of the threat model, when contracts or regulators expect technical exclusion of the operator, or when several parties must compute on combined data that none of them may see.

09

Conclusion

Confidential computing closes the last gap in data protection, data in use, with hardware isolation and attestation. It is mature enough for production on the major clouds, but it is a targeted control, not a cure-all. Start with a clear threat model, protect only the sensitive part of the workload, and make attestation and key release the centre of the design. For where your data is stored and how portable it is, see cloud vendor lock-in and digital sovereignty.

For preparing cryptography for future quantum threats, see post-quantum cryptography readiness.

Exploring confidential computing?

Tell us what data you process, who must not see it and where it runs. ZSpace Labs will help you decide whether confidential computing fits and how to adopt it. Discuss your security architecture.

Start a Project
FAQ

Common questions.

Protecting data while it is being processed, by running computation inside a hardware-based trusted execution environment (TEE) whose identity and configuration can be verified through attestation. The Confidential Computing Consortium, a Linux Foundation project, defines it as the protection of data in use by performing computation in a hardware-based, attested trusted execution environment.

Encryption at rest protects stored data, and encryption in transit protects data moving across networks. Both leave data decrypted in memory while it is being processed, where the operating system, hypervisor or cloud operator could in principle access it. Confidential computing protects that third state, data in use, with hardware isolation and memory encryption.

Yes. AWS offers Nitro Enclaves, Microsoft Azure offers confidential virtual machines on AMD SEV-SNP and Intel TDX, confidential GPU VMs and confidential containers, and Google Cloud offers Confidential VMs and Confidential Space. Availability varies by region and machine type, and some features are still in preview.

No. It reduces who can see data in use, including the cloud operator and administrators, but it doesn't fix vulnerabilities in the application running inside the protected environment, weak access control or poor key management. Researchers have also published side-channel attacks against several TEE technologies over the years, generally followed by vendor fixes.

There is usually some overhead from memory encryption and isolation, and the amount depends on the technology and workload. Measure it with your own workload before committing, particularly for I/O-heavy or GPU workloads.

Get in touch

Have a project in mind?

Whether you're building a new digital product, improving an existing website, or looking to automate part of your business — let's talk.