Why enterprises shift to open-source for AI sovereignty

This article summarizes an episode of 20VC with Harry Stebbings’s video series featuring Anastasios Angelopoulos, CEO of Arena.

Anastasios Angelopoulos, CEO of Arena / Photo credit: Anastasios Angelopoulos

The safest AI model for a business is not always the most powerful one. Anastasios Angelopoulos, co-founder and CEO of Arena, believes this reality will accelerate the adoption of open-source AI.

As open systems improve, buyers are prioritizing data control over raw performance or pricing. This demand for “AI sovereignty” creates an opportunity for American open-source leaders, even as Chinese developers release increasingly capable models.

Chinese open models changed the market

Choosing an AI model has become as much a question of trust as performance. Angelopoulos points to Kimi K3 as a turning point, arguing that Chinese open models are no longer simply catching up, they are beginning to outperform leading American models in some areas.

“Kimi actually beat all American models, including Fable, on some subset of tasks,” Angelopoulos says. “That doesn’t mean they are not distilling. They may still be using distillation as a substep, but distillation is only part of the story.”

If open models continue narrowing the performance gap while remaining cheaper and easier to customize, proprietary AI providers could face growing pricing pressure.

Businesses are already using stronger open models as leverage when negotiating with closed-model vendors.

Trust gaps create opportunities for American AI

This leverage comes with a catch. While companies want the high performance of foreign open models, they refuse to build their core infrastructure on technology they do not completely trust.

Angelopoulos views AI control as a critical business strategy. As software becomes commoditized, companies must protect themselves by owning their unique data and operational tools.

“Enterprises are going to want to own their own intelligence,” he says. “They’re going to want so-called AI sovereignty, which means owning your whole AI supply chain. You can take an open-source model, fine-tune it on your own company’s data, and own your stack end to end.”

This demand creates a massive opening for an American open-source leader. Because foreign models often fail enterprise security reviews, the winning provider will be the one that offers both top-tier capability and total system control.

Data becomes critical infrastructure

Owning the AI stack only matters if companies can train it with proprietary information. As foundation models improve, unique internal data becomes one of the few sustainable competitive advantages.

Angelopoulos argues that growing models will require increasingly specialized data, forcing both businesses and investors to rethink how they value data assets.

Testing controls how AI tools operate