Perplexity launches 'Hybrid Compute' to split AI tasks between local and cloud processing
The AMW Read
Extends Perplexity's known local-compute push (Nvidia-backed Portable Computer) into a software privacy/cost split, but remains a product feature update rather than a structural shift.
Perplexity launches 'Hybrid Compute' to split AI tasks between local and cloud processing
Perplexity has rolled out "Hybrid Compute" for Perplexity Computer on Mac, a feature that routes AI workloads between an on-device model and cloud-based models. A "privacy gate" determines which data leaves the device: cloud models handle reasoning, web search, and planning, while the on-device model processes tasks involving sensitive information and private files.
The move follows Perplexity's August launch, with Nvidia, of Portable Computer, a locally run AI agent device built to eliminate per-token inference costs, and comes amid reported Nvidia investment talks that could value Perplexity above $30 billion. Hybrid Compute applies the same local-processing logic in software form, addressing two pressures at once: enterprise and consumer demand for data locality, and the cost structure of routing every query through cloud inference. Per the AI Market Watch index, Perplexity has raised $1.72 billion to date (the index tracks roughly 5,000 companies; this is coverage, not a census).
For builders, the privacy-gate pattern β deciding locally what's sensitive enough to keep on-device before any cloud call β is a template other agent products handling private files or credentials may need to adopt as enterprise buyers push back on blanket cloud processing. For investors, the more relevant signal is cost: if on-device inference offsets even a fraction of cloud token spend at Perplexity's user scale, hybrid local/cloud architecture becomes a margin lever that agent companies without a hardware partnership like Perplexity's Nvidia tie-up will struggle to replicate quickly.
