Apple has started using new M5 chips inside its Private Cloud Compute servers, the infrastructure that handles cloud based requests for Apple Intelligence. The change shows the company is preparing for heavier AI workloads and more advanced assistant features that rely on secure remote processing instead of running fully on device.
The software also references a new component called Private Cloud Compute Agent Worker, which runs a special version of iOS built around an agent style architecture designed to process AI tasks. iOS 26.4 already includes code that connects to this system, which signals that upcoming features are tied closely to this backend change.
According to Apple’s latest Private Cloud Compute software release, the new architecture operates on hardware identified as model J226C powered by the M5 chip, confirming a major shift in server silicon strategy.
New server architecture and AI direction
For years, the company used M2 Ultra chips inside Private Cloud Compute servers even after newer generations arrived, and reports once suggested a move to M4 hardware, but that transition never widely happened. Now the jump directly to M5 suggests Apple waited for a larger performance and efficiency improvement before rebuilding its AI infrastructure.
The company also recently partnered with Google to use Gemini models for certain assistant capabilities, so stronger server hardware becomes necessary to handle hybrid processing between Apple’s systems and external models.
Apple is also designing dedicated AI server processors. Analyst Ming-Chi Kuo says mass production begins in the second half of 2026 with deployment planned for 2027. This indicates the current M5 rollout is an intermediate step before fully custom data center silicon arrives.
Research environment and security testing
Apple also published details about a Virtual Research Environment that lets security researchers boot a simulated Private Cloud Compute node on Apple silicon Macs. The environment can run inference requests, generate secure enclave attestations, and allow controlled testing of the system’s privacy protections.
Researchers can load custom code, adjust restricted execution modes, and inspect how requests process data in order to detect possible vulnerabilities. Apple specifically encourages investigation into “ways to execute unattested code,” “exploitable flaws in request processing,” and “any potential violations of our privacy protections.”
The company already builds Private Cloud Compute servers in Houston as part of its domestic infrastructure investment, and the move to M5 hardware shows the backend is evolving alongside upcoming AI features rather than reacting after launch.
