Perplexity bets the most important AI computer is the one already on your desk

Perplexity AI has unveiled a hybrid inference system that automatically decides, mid-task, whether to process data on a user’s local device or route it to cloud-based AI models. Michael Nuñez reports for VentureBeat that CEO Aravind Srinivas demonstrated the technology at Computex 2026 during Intel’s keynote, using confidential deal documents to show the system in action.

The core idea is straightforward: sensitive data such as financial records or health information stays on the device, while tasks requiring more complex reasoning are sent to powerful cloud models. The user does not need to configure anything. The system makes those decisions on its own.

Why this matters for businesses

For regulated industries like finance, healthcare, and law, keeping sensitive data off third-party servers is often a legal or contractual requirement. Perplexity’s system offers a middle path: local processing for confidential content, cloud processing for everything else. The company says it will ask users for permission before sending any sensitive task to the cloud.

The announcement fits into a broader enterprise push. Earlier this year, Perplexity introduced Computer for Enterprise, with integrations for tools like Salesforce, Snowflake, and SharePoint, as well as SOC 2 Type II certification.

A crowded but still open field

Apple, Google, and Microsoft all offer some form of local AI processing. What Perplexity claims sets its system apart is dynamic, task-level routing decided automatically during execution rather than set in advance.

The product is not yet available. Perplexity says the hybrid inference feature will launch in the coming weeks. Whether the technology performs as demonstrated outside a controlled stage environment remains to be seen.