In this conversation with Yang Li, co-founder and COO of Cosine, we discuss what sovereign AI actually means in practice. While the term has become increasingly common, Li argues that it comes down to three things: control, certainty, and resilience. Rather than relying on cloud-based AI services, Cosine builds foundation models that can run entirely inside a customer’s own infrastructure, including fully air-gapped environments where systems have no connection to the public internet.
“Cosine is a model lab. We pre-train and post-train our own foundational models, and we produce the product layer as well, similar to Claude Code and OpenAI Codex. But fundamentally, we can deploy it in a fully air-gapped environment,” said Li.
This ability has made Cosine particularly attractive to defence organisations and other highly regulated industries where sending sensitive information to a cloud provider is not an option. Customers include companies working with classified information as well as financial institutions that want greater control over how their data is handled.
One obvious question is how an air-gapped AI system stays current. Li said Cosine doesn’t push continuous updates in the way cloud providers do. Instead, every customer agrees on an update schedule that matches the rest of its software infrastructure. Some organisations prefer weekly updates, while others only update periodically. Cosine also provides customers with the same tools its own engineers use, allowing them to fine tune and post-train models using their own internal data without exposing it outside their organisation.
“We definitely don’t update as frequently as someone who’s part of a cloud solution,” Li said. “People always feel like if you’re using ChatGPT or Claude Code, the model feels very different from week to week, but that’s because they’re constantly updating it behind the scenes. We serve our customers through a licence agreement, so we agree ahead of time how often they want updates.”
Although Cosine began by building AI coding tools, Li said the same reasoning and problem-solving abilities that make a model good at programming also translate into research, document generation, customer service workflows, and other enterprise tasks. The company remains focused on software development, especially legacy programming languages that have become increasingly difficult to support as experienced developers retire.
“We started off as focused primarily on coding,” he said. “In certain legacy coding languages, we’re twice as good as the frontier models, including COBOL, Fortran, Ada and, to a certain extent, even .NET.”
That expertise is already solving practical problems. Li described speaking with a defence contractor that had aircraft sitting idle because only a handful of engineers still understood the Fortran code running parts of the avionics systems. AI models trained specifically for those languages can help maintain and modernise software that remains critical to military and government operations.
The company itself has grown alongside the rapid rise of generative AI. Although Cosine was founded in 2022, Li noted that widespread adoption of AI coding assistants has only taken off during the past year. Today the company is backed by the UK Government’s Sovereign AI Industry Partnership and works with organisations including BAE Systems, Babcock, Leonardo, PwC, Thales and HSBC. It is also training a new 1.35 trillion parameter mixture-of-experts model designed to deliver frontier-level performance while remaining deployable in fully sovereign, air-gapped environments.
