Prem launches Enclave for secure sovereign AI clusters

Prem AI has launched Enclave, a product designed to protect distributed GPU clusters used to run AI models on sensitive data.

The launch centres on a system that applies confidential computing and post-quantum encryption across an entire AI cluster rather than a single machine. Prem said this lets organisations run large open-weight AI models while keeping models and data encrypted throughout processing.

The approach addresses a problem facing companies and public bodies that want to use AI on regulated or confidential information without sending that data to external providers. Enclave is intended to let customers deploy AI on existing infrastructure, whether on premises or in the cloud, without adding new hardware.

Based in Switzerland, Prem operates in sectors including legal, healthcare, finance and government. It said the product is already being used in production by customers and design partners in government, defence, legal and healthcare.

Prem said Enclave includes hardware-verified attestation, does not store data and does not allow it to leave the enclave during processing. It also provides cryptographic proof that information remains inaccessible, including to Prem itself.

Market shift

The launch comes as organisations face tighter security rules and growing scrutiny over data sovereignty. Concerns about where data is processed, who can access it and how it is protected have grown as businesses test large AI models on internal records, legal documents, health information and other sensitive material.

At the same time, governments and large institutions are paying more attention to post-quantum cryptography as a longer-term security requirement. Prem is positioning Enclave around those concerns, arguing that the cost and technical complexity of sovereign AI deployments have until now limited broader adoption.

The system is built on Intel TDX, AMD SEV-SNP and NVIDIA Confidential Computing, according to Prem. It supports multi-GPU attestation, GPU-level encryption, post-quantum encryption and zero data retention.

Prem said the product can be deployed on standard GPU hardware in two to four weeks. It is being offered through customer-managed infrastructure, private cloud environments and an application programming interface.

The launch expands Prem's wider push into what it calls sovereign AI, or AI systems run with tighter control over data, infrastructure and access. Earlier this year, the company introduced Fluso, a private workspace for agentic AI, and now says Enclave completes the underlying infrastructure layer of that platform.

Prem also disclosed operational figures intended to show existing use of its software. It said it has processed more than 100 million clinical health reports, delivers more than 3,000 hours of workflow automation each quarter and supports more than 50 custom AI pipelines in production.

Simone Giacomelli set out the company's case for the product in a statement issued alongside the launch.

"Sovereignty in AI means owning the intelligence that compounds value for your organisation. Until now, putting AI to work on your most sensitive data meant handing that data to someone else and trusting them with it. Enclave ends that trade-off by enabling organisations to run frontier AI on their own data, with mathematical proof to verify that it was never seen," said Simone Giacomelli, Founder and Chief Executive Officer, Prem AI.

The company also cited support from one of its investors.

"The global shift toward sovereign and verifiable AI is one of the most important infrastructure transitions of this decade. I've backed Simone and the team since day one because they're building the architecture institutions need to adopt AI with confidence," said Jim Breyer, Founder, Breyer Capital.

For buyers, the commercial question will be whether broad cluster-level protection can be introduced without slowing deployment or adding major operational complexity. Prem's pitch rests on the claim that organisations can keep control of sensitive workloads while still using large open-weight models on standard GPU environments.

That message is likely to resonate most in sectors where internal data cannot easily be shared with outside platforms and where auditability matters as much as model performance. Prem said Enclave is available now through customer-managed infrastructure, private cloud environments and the Enclave API.