English 
How safe is your cloud data once it actually has to be put to use? Files are secured against being viewed while they are parked in storage or moving across networks. But that data generally gets in your app’s hands, where it can be viewed in memory.
That creates a security blind spot that traditional cloud protection can’t fully solve. Confidential computing takes you to the last frontier, defending data while in use with hardware-based isolation and TEEs. It’s a game changer for financial, health and proprietary data that is too sensitive for public cloud.
The real challenge is how effectively an enterprise can implement it and yet retain performance and control.
What is Confidential Computing and How Does It Work?
Confidential Computing Confidential computing protects data in-use via a secure hardware environment, including the use of enclaves or TEEs. It protects the offloading of workloads between multiple workloads on the device and from the OS, hypervisor or cloud platform. Data remains in a secure environment or cloud where it can be acted upon without being seen.
The starting point is the encrypted data flow. Only after the workload has been authenticated, the key management service brings in the decryption key into the TEE. Within the chip, the TEE’s CPU decrypts the data before, during the processing, and after. This keeps it protected.
Security of legacy software relying on access control and systems rights Confidential computing leverages a hardware root of trust (AMD SEV, Intel SGX, ARM Trust Zone).
For example, AWS shows this with Nitro Enclaves. Nitro Enclaves generate a signed attestation document which includes the measurements of the PCRs of the enclave image, kernel, and application. The attestation document can be validated by an outside client, enabling an HSM to only provide secret key material if it successfully attests.
Why the Modern Enterprise Needs Confidential Computing?
Zero Trust can be summed up in a few words. Don’t automatically trust any user, device, or system that is within the corporate network. In public and hybrid cloud environments, confidential computing extends this trust model through the use of hardware-based protections. No privileged infrastructure component is ever allowed to directly access data stored inside a Trusted Execution Environment. This avoids data exposure by compromised hosts or malicious actors.
It’s the same protection that counts as organizations face regulations like GDPR, HIPAA and the Digital Personal Data Protection Act in India. But hardware-based isolation can prove data is protected when being processed, even against the host infrastructure, it can’t of itself prove the enterprise is compliant. They will have to show compliance requirements such as consent, access, retention and data governance.
The way we connect Industry 4.0 will impact how organizations work together. For example, to help banks stop money laundering, banks could identify patterns in their customers’ data and pass that info on without exposing what that data actually is. Microsoft describes how one bank, Scotiabank, used Azure Confidential Computing with Opaque to identify money laundering and maybe even human trafficking.
High-Impact Industry Use Cases
And the power of confidential computing is realized when organizations shift sensitive workloads to the cloud. ヘルスケア, financial services, and AI work requires access to significant amounts of data. But each industry has a different risk when that data becomes available for the taking.
Here’s a good one from the tech world. Healthcare and genomics is a great example. In today’s world, hospitals and research organizations are demanding more and more cloud infrastructure to support workloads analyzing patient records, diagnostic imaging, and genomics data. Confidential computing is evolving to support this workload in secure enclaves with fewer data access points, enabling data scientists to crunch private medical data and share insights across organizations.
Financial services are also working on this use case for cross-border payments. Banks want to track suspicious activity across payment rails to improve their anti-money laundering efforts, but releasing transactional records can reveal personally identifiable information or business insight that banks might prefer not to share. Confidential computing makes it feasible for banks to work with sensitive data sets.
Another source of risk is involved with the use of artificial intelligence and machine learning. Training data can include the customer data and models can be intellectual property. Processing both within secure environments helps protect these assets when enterprises use public cloud infrastructure.
Google Cloud’s general availability of the accelerator-optimized g4-standard-48 machine type reflects this growing demand for secure AI infrastructure. Its configuration combines a 5th Generation AMD EPYC Turin processor, AMD SEV, and one NVIDIA RTX PRO 6000 GPU. This gives enterprises a dedicated infrastructure option for confidential AI and ML workloads.
Challenges, Limitations, and Implementation
Confidential computing strengthens cloud security, but adopting it introduces practical challenges. Performance overhead remains one concern. Encryption, decryption, and isolation can add processing costs, potentially affecting latency in demanding workloads. However, the penalty depends on the hardware, workload, and implementation. In a controlled DeepSeek-R1 inference benchmark across eight NVIDIA B200 GPUs, confidential computing retained 96.1–98.2% of baseline output-token throughput, with mean per-token latency overhead of 1.2–4.3% across concurrency levels of 1–16. The test used Intel TDX and TensorRT-LLM.
Hardware dependency creates another challenge. Different platforms support different confidential computing technologies, which can complicate migration across cloud providers and increase the risk of vendor lock-in. Intel has disclosed that its future Confidential Computing investments will focus exclusively on TDX. Diamond Rapids will be its last next-generation Xeon platform to support SGX, although SGX support will continue well into the 2030s. Developers therefore have time to plan their transition.
The skill gap is as important. Legacy applications often need refactoring before running within the enclave. If engineering teams don’t understand workload isolation, attestation, key management and platform requirements, it can be costly and hard to sustain.
The Future Outlook
Confidential computing will probably be a major addition to enterprise cloud security as enterprises take their workloads to the cloud. The CCC facilitates this transition in open source to foster cooperation and prevent fragmentation. A more robust shared ecosystem will help make adoption less dependent on individual vendor implementations.
エッジ computing presents another opportunity. As more data is analyzed and acted on directly on IoT devices, industrial systems, and connected infrastructure, privacy and security will become more essential. Confidential computing can help enable secure, hardware-isolated environments on edge nodes and support the security of sensitive workloads that are not tied to a centrally hosted cloud. This will be especially valuable when IoT devices are widely distributed and process information that needs to be treated as sensitive and protected. The industry’s adoption of it will rely on making these safeguards available to a wider range of hardware platforms, cloud vendors, and edge deployments.
こちらもお読みください： QueryPie AI、AIを活用したペネトレーションテストサービスを開始した
結論
Securing data in the cloud has traditionally focused on data at rest and in motion. Confidential computing closes the gap of data in use and enables organizations to have more confidence in their most valuable workloads. But deploying it everywhere is neither feasible nor necessary.
IT teams should start by conducting an inventory of their cloud workloads and pinpointing workloads that pose the greatest privacy, regulatory and commercial risk. Financial, health and customer data, internal models, or any other type of sensitive information would be sensible candidates. Moving these workloads first will allow firms to ‘err on the side of caution’ in areas of greatest impact.
日本語