The two technologies take different approaches to protecting data while it is processed. SGX isolates selected code and data inside application-level enclaves, requiring developers to design or modify software to use them. TDX instead isolates entire virtual machines from the host, including the hypervisor and other VMs, so existing workloads can run inside confidential VMs without extensive changes. That makes TDX particularly attractive to cloud providers running large numbers of virtual machines, and Intel says it offers the scalability and ease of deployment needed for wider adoption of confidential computing.

Both technologies have faced security vulnerabilities over the years, including side-channel attacks and flaws in their attestation mechanisms, but Intel does not cite security concerns as a reason for dropping SGX. The company frames the move as reflecting the industry's preference for VM-level isolation over application-specific enclaves.

Intel also plans to expand TDX with new capabilities: TDX Connect for trusted I/O, which extends protection to data moving between a VM and external devices such as accelerators, plus post-quantum cryptography and enhanced attestation. No availability dates for these enhancements have been disclosed. Intel sees agentic AI as a key driver, arguing that autonomous agents can run in separate hardware-isolated VMs without software modifications, and it is working with Nvidia to coordinate confidential computing across CPUs and GPUs. TDX is already available through providers including Google Cloud, Microsoft Azure, Alibaba Cloud and IBM Cloud, with Red Hat, Canonical and VMware adding support for private clouds.