- Lugano-based Prem AI is seeking $100M at a $500M+ valuation to let hedge funds and law firms run AI on their own servers
- It has launched Fluso, an encrypted AI workspace that runs agents inside a customer’s own infrastructure
- Known backers include Fan Zhang of Sequoia Capital China and David Maisel, founding chairman of Marvel Studios
Prem AI, a Lugano-based startup that lets hedge funds and law firms run artificial intelligence on their own servers instead of sending data to OpenAI or Anthropic, is seeking to raise $100 million in a Series A at a valuation of at least $500 million, Bloomberg reports.
The company expects to close the round in the third quarter of 2026.
The fundraiser comes alongside the launch of Fluso, an encrypted AI workspace that runs agents and automates tasks entirely inside a customer’s own infrastructure, whether that is a private cloud, a virtual private cloud, or a fully air-gapped on-premise environment. The product is built for organisations where data leaving the building is simply not an option.
The problem with renting AI
Every API call to OpenAI or Anthropic is also a data transfer. Trading strategies, legal documents, and client records are stored on third-party servers governed by someone else’s terms of service, under a US legal regime that can compel disclosure.
For a hedge fund or a law firm, that is not an abstract concern. It is a competitive and regulatory exposure that compounds over time. The sovereign AI question, once mostly a debate about countries and cloud dependency, now applies to any firm whose edge lives in its data.
Simone Giacomelli co-founded SingularityNET, the decentralised AI network, before starting Prem in 2023. The company is built around a single architectural bet: that firms handling the most sensitive data will eventually refuse to rent intelligence from a third party, and will demand infrastructure they can inspect, own, and control.
How Prem works
The startup combines self-hosted model deployment with what it calls a ready-to-use application layer, giving customers a full AI stack, from inference to agent automation, without requiring them to assemble it from parts.
Fluso, the new workspace product, runs on open-weight models, meaning every parameter can be audited. Data never enters training pipelines. A free tier also positions it as a consumer-facing entry point into Prem’s broader infrastructure play.
The self-hosted category is not empty. Databricks offers on-premise AI, and open-source models from Meta and Mistral give enterprises a starting point. But Prem packages the full stack as a finished product rather than a toolkit.
The difference, in Giacomelli’s framing, is the application layer: security without usability is infrastructure that never gets used.
Backers and the money trail
Known investors from previous rounds include Fan Zhang, co-founder of Sequoia Capital China, and David Maisel, the founding chairman of Marvel Studios, who engineered its $4 billion sale to Disney in 2009. Prem has not confirmed which investors are participating in the current round.
Prem raised a $14 million seed round in April 2024, followed by a $6.1 million bridge at a $200 million valuation. If the Series A closes above $500 million, it would represent a 2.5x valuation step-up in roughly two years. The company intends to use the funds to grow its team and accelerate product development.
