Skip to content

Modelyo

Israeli deep-tech startup building a 'confidential cloud' that lets regulated organizations run sensitive data and AI workloads on public cloud infrastructure while keeping everything encrypted even during use.

Visit Website ↗ + Add to Compare Claim This Company
50/100Incremental Innovator

Overview

Modelyo is an Israeli company, headquartered in Herzliya and founded in 2024 by CEO Benny Meir and VP R&D Michael Kiperberg, building what it calls a “confidential cloud.” The platform lets organizations run sensitive data, applications, and AI workloads — including inference, agents, and retrieval-augmented generation — on cloud infrastructure while keeping data cryptographically isolated at rest, in transit, and during active use, with customers retaining control of their own encryption keys and hardware security modules.

Modelyo’s platform works across Intel TDX, AMD SEV-SNP, Arm CCA, and NVIDIA confidential-computing GPU architectures, spanning hyperscalers, specialized “neoclouds,” sovereign clouds, and on-premise environments. In February 2026, the company joined the Confidential Computing Consortium, a Linux Foundation project, as a start-up member. It targets financial services, government and defense, and healthcare customers who need cloud elasticity without giving up control of sensitive data or model weights.

Modelyo’s differentiator is breadth of hardware support and its focus on “sovereign” deployment: rather than building for a single cloud or a single confidential-computing chip architecture, it aims to give regulated customers a consistent confidential-computing layer regardless of which infrastructure they ultimately choose.

Innovation Matrix Assessment

Innovation Velocity 5/10

Modelyo has shipped a multi-architecture confidential-computing platform and joined the Confidential Computing Consortium within about two years of founding, a reasonable but not exceptional pace for deep-tech infrastructure.

Operational Value 6/10

Letting regulated customers keep control of keys and model weights while still using public cloud elasticity solves a real compliance and data-sovereignty operational problem.

Market Momentum 4/10

Confidential Computing Consortium membership is independently verifiable, but seed-stage funding figures conflict across sources and no named paying customer was confirmed.

Category Disruption 5/10

Confidential/sovereign cloud computing is a genuine architecture shift, but Modelyo is one of several vendors, alongside OPAQUE, Evervault, enclaive, and Anjuna, pursuing the same general approach.

Real-World Efficacy 3/10

No independent benchmarks, audits, or named production deployments were found; the company's own anonymized customer testimonials are the only evidence of real-world use.

Enduring Relevance 7/10

As data-sovereignty and AI-governance regulation tightens, confidential cloud infrastructure that keeps data encrypted even during use will matter more over the next several years.

Why CISOs Should Care

Lets regulated organizations use public cloud and AI infrastructure for sensitive workloads without handing custody of encryption keys or model weights to the cloud provider.

What Makes It Different

Supports multiple confidential-computing hardware architectures (Intel TDX, AMD SEV-SNP, Arm CCA, NVIDIA GPUs) across hyperscale, neocloud, sovereign, and on-prem environments rather than locking into one stack.

The Matrix Verdict

50/100 — INCREMENTAL INNOVATOR

Modelyo is an Incremental Innovator at the early edge of a genuinely important category: its multi-architecture confidential-cloud approach is credible and its Linux Foundation consortium membership is real, but it is still seed-stage with no independently confirmed customer or funding detail to push it higher.

Editorial Note: Claims vs. Verified Findings

Confidential Computing Consortium membership and the company's own technical description of its hardware support are independently or directly verifiable; funding amount, founding year, and headcount are inconsistent across third-party aggregators and should be treated as approximate.

Sources