Sovereign wealth funds want Perplexityās AI on their own hardware, with no data access for Perplexity
Perplexity is fielding requests for fully isolated deployments of its Computer agent platform as data sovereignty becomes a buying condition
Some of the world’s most cautious investors have a request for Perplexity: bring your AI to our servers, and do not look inside.
Sovereign wealth funds have asked Perplexity to run its product on hardware the funds own. That arrangement gives them full control and blocks Perplexity from seeing their data.
What the funds are asking for
The request centers on Perplexity Computer, the company’s agent platform. The funds want the whole system running on their own machines, powered by fine-tuned open-weight models.
Open-weight models are AI systems whose underlying parameters are published. Anyone with enough hardware can download and run them without calling home to the original developer.
Under this setup, Perplexity has no access to the data. It also gets no usage telemetry, which is the behind-the-scenes reporting that tells a software company how its product is being used.
CEO Aravind Srinivas described the requirements in an interview on October 12, 2026. He said sovereign wealth funds want completely self-contained systems. That includes the agent harness and the control loops, the scaffolding that tells an AI agent what to do and checks its work.
Updates, according to Srinivas, would arrive only as software. No remote peeking, no live connection required to keep things running.
The funds involved have not been publicly identified. No deal sizes or affected countries have been disclosed either.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
How Perplexity got here
On August 25, 2026, the company launched its Portable Computer. The system runs on Nvidia DGX Spark hardware, a compact machine designed to run AI workloads locally rather than in a distant data center.
Portable Computer keeps sensitive work on the device. Users can optionally hand off non-sensitive tasks to the cloud. Fully local workflows carry zero token costs, meaning users are not paying per unit of text processed the way they would with a cloud AI service.
In September 2026, Perplexity rolled out Hybrid Compute features for Apple silicon. The feature automatically detects sensitive information and keeps it processed locally.
Perplexity’s strategic partnership with Nvidia began around June 2025, aimed at government and enterprise customers in Europe and the Middle East. The company also launched US federal partnerships in November 2025.
What this means for AI buyers and builders
Sovereign wealth funds manage national wealth. Their research, positioning, and deliberations are among the most sensitive information a government holds.
That has trade-offs for the company. Telemetry is how AI firms learn where their products fail and how customers actually use them. Giving it up means flying partly blind on some of its most demanding accounts.
Nvidia sits on both sides of that equation. It sells chips into giant data centers, and through products like DGX Spark, it also sells hardware for running AI locally.
The appeal likely extends beyond sovereign funds. Finance, law, and health all handle information that cannot casually leave the building. Those sectors have been among the most hesitant to adopt cloud AI tools for exactly that reason.
The other question is execution. Running fine-tuned open-weight models on customer hardware, with updates shipped only as software and no remote visibility, is harder to support than a centralized service. Bugs are tougher to diagnose when the vendor cannot see the patient.