Domyn CEO argues enterprises must own AI models

In an interview with The Information, Uljan Sharka warns that relying on major cloud APIs compromises critical data sovereignty.

Companies in heavily regulated industries need to own AI models and proprietary data rather than routing critical workloads through application programming interfaces (APIs) provided by major cloud vendors. That was the central argument made by Uljan Sharka, founder and chief executive of European startup Domyn, during an appearance on The Information’s daily show TITV on October 9, 2026.

Speaking during a segment on corporate AI ownership, Sharka warned that the standard software-as-a-service setup creates legal and strategic vulnerabilities. When regulated enterprises send confidential records through third-party APIs, they risk exposing internal workflows and forfeiting actual ownership of the core technology running their business.

Why regulated firms choose to own AI models

Sharka pitched his case directly to organizations in banking, insurance, pharmaceuticals, and defense. In these industries, strict data-protection laws and trade-secret mandates prevent internal records from touching commercial models managed by outside providers.

According to Sharka, conventional Model-as-a-Service offerings do not meet those compliance standards. Domyn proposes an alternative architecture: providing customers with underlying source code and running continuous pre-training directly on private corporate records. Model weights and parameters stay locked within the client’s secure perimeter, eliminating the need to rent compute time or query closed proprietary algorithms.

A $1 billion debt package and proprietary compute

Domyn started in Milan in 2016 under the name iGenius. In June 2025, the company rebranded as Domyn to focus entirely on digital sovereignty for corporate clients. Sharka’s work on sovereign European systems earned him a place on the TIME100 AI list in 2026, curated by TIME Magazine.

To build the private computing infrastructure backing this model, the startup secured a financing package exceeding $1 billion. The core of that effort is a dedicated supercomputer facility in Italy, developed in partnership with chipmaker Nvidia and tech group G42. Notably, around 90% of that total capital consists of structured debt rather than equity funding.

Operational tradeoffs and hardware depreciation

That heavy reliance on debt introduces significant operational risk. Artificial intelligence hardware loses economic value quickly. New generations of graphics processing units (GPUs) hit the market every eighteen to twenty-four months, making older server clusters far more expensive to run per unit of compute. Paying off substantial debt obligations while hardware depreciates requires consistent, high-margin revenue.

So far, Domyn has not released audited financial statements proving the long-term viability of this balance sheet. The company has also not published independent benchmarks comparing its proprietary models against industry standards from OpenAI, Anthropic, or Google. For now, claims regarding the system’s technical edge and commercial sustainability rest on the firm’s own statements.

Digital sovereignty versus operational convenience

Sharka has an obvious commercial interest in pushing this thesis, given that Domyn’s core product is dedicated on-premises infrastructure. Even so, his pitch reflects growing anxiety among European chief information officers and compliance teams.

With regulatory mandates such as the European Union’s AI Act taking effect, corporate liability for privacy failures and data leakage is increasing sharply. For executives reporting to regulatory boards, the initial speed of hooking up a commercial API is increasingly measured against the long-term threat of vendor lock-in and lost data sovereignty.

Sources

• The Information: TITV episode interview, October 9, 2026
• TIME Magazine: TIME100 AI 2026 profile
• The Next Web: $1 billion debt financing and Italian supercomputer
• Domyn: official rebranding announcement