Making Confidential AI effortless with Intel TDX and iExec

No need to repeat how AI is transforming industries. Let’s talk about what really mattered for developers: how painful it used to be to build actual Confidential AI applications.

For years, securing AI workloads meant wrestling with Intel SGX, which sounded great in theory but came with a catch: Compatibility struggles with SGX.

For developers wanting confidentiality, they had to modify their entire application and pray that the missing syscalls wouldn’t break everything. Half the time, they were debugging compatibility issues instead of actually building AI solutions. It was a time sink, a headache, and a roadblock to real-world adoption.

Then came Intel TDX and suddenly, building confidential dApps didn’t have to be this hard.

With TDX, builders run AI workloads in secure VMs without rewriting a single line of code. No more battling syscall limitations, no more forced compatibility fixes, just a seamless, high-performance confidential environment that works out of the box.

This shift is already making waves, and iExec is at the forefront of it. As a Gold Member of the Intel Partner Alliance, iExec has been leading the way in Confidential Computing. Its AI model monetization solution was featured in Intel’s AI catalog, reinforcing its role in shaping the future of privacy-first AI.

And it doesn’t stop there. Through NVIDIA’s Inception Program, iExec gained early access to advanced GPU technology, unlocking even greater scalability and security for AI workloads. 

Meaning iExec and Intel® TDX make Confidential AI effortless.

What Is Intel TDX and Why Does It Matter for Confidential AI?

Intel® Trust Domain Extensions (Intel® TDX) is a hardware-isolated trusted execution environment (TEE) designed to enhance data confidentiality and integrity in virtualized environments.

Built into Intel’s 4th Generation Xeon® Scalable processors (codenamed Sapphire Rapids), Intel TDX introduces Trust Domains (TDs) to isolate virtual machines (VMs) from the hypervisor, BIOS, System Management Mode (SMM), and even cloud service providers (CSPs). This isolation ensures that AI workloads run in a secure execution environment, free from unauthorized access, even by cloud providers.

For AI applications, security and privacy are crucial, particularly when handling sensitive datasets, proprietary models, or confidential computations. Intel TDX provides hardware-based protections that significantly reduce the attack surface while maintaining high performance.

Key Hardware Capabilities for AI

• CPU Acceleration for AI: Intel TDX is optimized for AI workloads running on Intel Xeon Scalable Processors, leveraging AVX-512 and other CPU optimizations to accelerate deep learning and machine learning models.

• Hardware-Assisted Confidentiality & Integrity: Intel TDX leverages Multi-Key Total Memory Encryption (MKTME) and Physical Address Metadata Table (PAMT) to protect AI data and models from unauthorized access by encrypting memory using multiple encryption keys.
• Ultra Path Interconnect (UPI) Encryption: Ensures that data transmitted between processors remains encrypted, securing AI models in distributed computing environments.
• Compatibility with Leading AI Frameworks: Intel TDX seamlessly integrates with popular AI frameworks, including TensorFlow, PyTorch, and BERT, ensuring developers can deploy AI models without major modifications.
• Reduced Noisy Neighbor Effect: AI workloads often require predictable performance. Intel TDX mitigates performance interference by isolating workloads at the hardware level, reducing contention for CPU and memory resources.
• Transparent Huge Pages (THP) Support: This feature optimizes memory management, ensuring efficient AI execution in cloud environments.

• Compatibility with NVIDIA Confidential Computing (CC) GPUs: Intel TDX is compatible with NVIDIA H100 GPUs, allowing AI workloads to leverage powerful hardware accelerators while maintaining data confidentiality.

Enhanced Security Features

Intel TDX employs several mechanisms to safeguard AI workloads:

• Secure Arbitration Mode (SEAM): A new CPU mode that supports Trust Domains (TDs) and ensures a clear separation between the hypervisor and the guest environment.
• TDX Module: A software module that runs in SEAM mode to manage the TDX architecture, ensuring that data inside a TD remains confidential.
• Remote Attestation: Verifies the integrity of the execution environment, allowing AI developers to confirm that their workloads are running on a trusted platform.
• Reduced Attack Surface: By isolating workloads from the host environment, Intel TDX minimizes the risk of unauthorized access and data breaches.

Intel TDX combines these features to create a trusted execution environment (TEE). This helps organizations deploy AI models safely in the cloud. It ensures data confidentiality, integrity, and compliance.

Intel TDX vs. Intel SGX: What’s the Difference?

Both Intel TDX and Intel Software Guard Extensions (SGX) are designed for Confidential Computing technology, but they serve different use cases. iExec supports both technologies, giving developers the flexibility to choose the best fit for their AI applications.

Secure Enclaves for Trusted Execution with Intel SGX

Intel SGX enables the creation of secure enclaves, isolated memory regions that protect sensitive computations from external access. This is ideal for:

• Privacy-preserving AI
• Secure key management
• Cryptographic operations
• Confidential data processing at an application level

A More Scalable Approach for Confidential AI with Intel TDX

Unlike SGX, which protects applications at a granular level, Intel TDX isolates entire virtual machines. This makes it a better choice for large-scale AI workloads, where bigger memory and multi-key encryption are required. 

TDX’s Secure Arbitration Mode (SEAM) further strengthens workload isolation, making it ideal for privacy-focused AI applications that need to process large datasets securely.

Why Intel TDX + iExec is the Best Choice for Confidential AI

The combination of Intel TDX and iExec creates new possibilities for Confidential AI, enabling secure, decentralized, and scalable execution of AI workloads.